Maintained by Difan Deng and Marius Lindauer.
The following list considers papers related to neural architecture search. It is by no means complete. If you miss a paper on the list, please let us know.
Please note that although NAS methods steadily improve, the quality of empirical evaluations in this field are still lagging behind compared to other areas in machine learning, AI and optimization. We would therefore like to share some best practices for empirical evaluations of NAS methods, which we believe will facilitate sustained and measurable progress in the field. If you are interested in a teaser, please read our blog post or directly jump to our checklist.
Transformers have gained increasing popularity in different domains. For a comprehensive list of papers focusing on Neural Architecture Search for Transformer-Based spaces, the awesome-transformer-search repo is all you need.
1.
Sun, Xiaoxue; Wang, Hongpeng; Song, Pei-Cheng
Compact Training-Free NAS with Alternating Evolution Game for Medical Image Segmentation Proceedings Article
In: Gee, James C.; Alexander, Daniel C.; Hong, Jaesung; Iglesias, Juan Eugenio; Sudre, Carole H.; Venkataraman, Archana; Golland, Polina; Kim, Jong Hyo; Park, Jinah (Ed.): Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, pp. 108–118, Springer Nature Switzerland, Cham, 2026, ISBN: 978-3-032-05325-1.
@inproceedings{10.1007/978-3-032-05325-1_11,
title = {Compact Training-Free NAS with Alternating Evolution Game for Medical Image Segmentation},
author = {Xiaoxue Sun and Hongpeng Wang and Pei-Cheng Song},
editor = {James C. Gee and Daniel C. Alexander and Jaesung Hong and Juan Eugenio Iglesias and Carole H. Sudre and Archana Venkataraman and Polina Golland and Jong Hyo Kim and Jinah Park},
url = {https://link.springer.com/chapter/10.1007/978-3-032-05325-1_11},
isbn = {978-3-032-05325-1},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
booktitle = {Medical Image Computing and Computer Assisted Intervention – MICCAI 2025},
pages = {108–118},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) has shown significant potential in designing deep neural networks for medical image segmentation. However, even emerging training-free NAS frameworks often incur substantial computational costs and lengthy search times. To address the critical challenges of computational efficiency and architecture interpretability, the paper proposes a compact training-free NAS framework based on an Alternating Evolution Game (AEG-cTFNAS). The proposed method alternates the search and contribution evaluation of the encoder and decoder within the UNet architecture via alternating games. It employs a truncated normal distribution for compact encoding, sampling, and updating to minimize computational overhead, while Bayesian inference is utilized to estimate the contribution of each block, adaptively adjusting the search strategy and facilitating process visualization. Experimental results on two benchmark datasets reveal that AEG-cTFNAS outperforms both manually designed architectures and NAS-based algorithms, underscoring its efficacy and potential on medical image segmentation. Code is available at https://github.com/spcity/AEG-cTFNAS.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) has shown significant potential in designing deep neural networks for medical image segmentation. However, even emerging training-free NAS frameworks often incur substantial computational costs and lengthy search times. To address the critical challenges of computational efficiency and architecture interpretability, the paper proposes a compact training-free NAS framework based on an Alternating Evolution Game (AEG-cTFNAS). The proposed method alternates the search and contribution evaluation of the encoder and decoder within the UNet architecture via alternating games. It employs a truncated normal distribution for compact encoding, sampling, and updating to minimize computational overhead, while Bayesian inference is utilized to estimate the contribution of each block, adaptively adjusting the search strategy and facilitating process visualization. Experimental results on two benchmark datasets reveal that AEG-cTFNAS outperforms both manually designed architectures and NAS-based algorithms, underscoring its efficacy and potential on medical image segmentation. Code is available at https://github.com/spcity/AEG-cTFNAS.
2.
Pandala, Madhavi Latha; Periyanayagi, S.
Optimal explainable vision transformer framework for skin cancer diagnosis with neural architecture search feature learning Journal Article
In: Biomedical Signal Processing and Control, vol. 112, pp. 108723, 2026, ISSN: 1746-8094.
@article{PANDALA2026108723,
title = {Optimal explainable vision transformer framework for skin cancer diagnosis with neural architecture search feature learning},
author = {Madhavi Latha Pandala and S. Periyanayagi},
url = {https://www.sciencedirect.com/science/article/pii/S1746809425012340},
doi = {https://doi.org/10.1016/j.bspc.2025.108723},
issn = {1746-8094},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
journal = {Biomedical Signal Processing and Control},
volume = {112},
pages = {108723},
abstract = {Skin cancer is one of the most prevalent and life-threatening diseases worldwide, making early and accurate detection crucial for improving patient survival rates. Traditional diagnostic methods rely on manual examination by dermatologists, which is subjective and time-consuming. To address these challenges, this research presents an advanced Optimal XAI based Skin Cancer Classification Network (OXAI-SCC-Net) framework for automated skin cancer detection and classification. The proposed methodology integrates multiple novel techniques to enhance accuracy and robustness. Initially, a Neural Architecture Search-Large Network (NASL-Net) is employed for feature extraction, leveraging automated deep-learning architecture search to optimize feature learning. To tackle class imbalance, a Support Vector Machine-Adopted Synthetic Oversampling Technique (SSOT) is utilized, which improves upon Synthetic Minority Oversampling Technic (SMOTE) by generating synthetic samples based on Support Vector Machine (SVM) decision boundaries, ensuring a balanced dataset. Further, Hippopotamus Optimization Algorithm with Explainable Artificial Intelligence (HOA-XAI) is applied for feature selection, reducing computational complexity by selecting the most informative features while minimizing redundant ones. Finally, a Vison Transformer Convolutional Neural Network (VT-CNN) classifier is trained on the optimized dataset to classify skin lesions into different categories. The proposed OXAI-SCC-Net method achieved an accuracy of 99.12%, with precision, recall, and F1-score each at 99.13% on ISIC-2019 dataset. This indicates highly consistent and reliable performance across all evaluation metrics compared to state-of-art approaches.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Skin cancer is one of the most prevalent and life-threatening diseases worldwide, making early and accurate detection crucial for improving patient survival rates. Traditional diagnostic methods rely on manual examination by dermatologists, which is subjective and time-consuming. To address these challenges, this research presents an advanced Optimal XAI based Skin Cancer Classification Network (OXAI-SCC-Net) framework for automated skin cancer detection and classification. The proposed methodology integrates multiple novel techniques to enhance accuracy and robustness. Initially, a Neural Architecture Search-Large Network (NASL-Net) is employed for feature extraction, leveraging automated deep-learning architecture search to optimize feature learning. To tackle class imbalance, a Support Vector Machine-Adopted Synthetic Oversampling Technique (SSOT) is utilized, which improves upon Synthetic Minority Oversampling Technic (SMOTE) by generating synthetic samples based on Support Vector Machine (SVM) decision boundaries, ensuring a balanced dataset. Further, Hippopotamus Optimization Algorithm with Explainable Artificial Intelligence (HOA-XAI) is applied for feature selection, reducing computational complexity by selecting the most informative features while minimizing redundant ones. Finally, a Vison Transformer Convolutional Neural Network (VT-CNN) classifier is trained on the optimized dataset to classify skin lesions into different categories. The proposed OXAI-SCC-Net method achieved an accuracy of 99.12%, with precision, recall, and F1-score each at 99.13% on ISIC-2019 dataset. This indicates highly consistent and reliable performance across all evaluation metrics compared to state-of-art approaches.
3.
Wang, Xingyu; Ji, Junzhong; Liu, Gan; Xiao, Yadong
PE-RBNAS: A robust neural architecture search with progressive-enhanced strategies for brain network classification Journal Article
In: Medical Image Analysis, vol. 107, pp. 103813, 2026, ISSN: 1361-8415.
@article{WANG2026103813,
title = {PE-RBNAS: A robust neural architecture search with progressive-enhanced strategies for brain network classification},
author = {Xingyu Wang and Junzhong Ji and Gan Liu and Yadong Xiao},
url = {https://www.sciencedirect.com/science/article/pii/S1361841525003597},
doi = {https://doi.org/10.1016/j.media.2025.103813},
issn = {1361-8415},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
journal = {Medical Image Analysis},
volume = {107},
pages = {103813},
abstract = {Functional Brain Network (FBN) classification methods based on Neural Architecture Search (NAS) have been increasingly emerging, with their core advantage being the ability to automatically construct high-quality network architectures. However, existing methods exhibit poor robustness when dealing with FBNs that have inherent high-noise characteristics. To address these issues, we propose a robust NAS with progressive-enhanced strategies for FBN classification. Specifically, this method adopts Particle Swarm Optimization as the search method, while treating candidate architectures as individuals, and proposes two progressive-enhanced (PE) strategies to optimize the critical stages of population sampling and fitness evaluation. In the population sampling stage, we first utilize Latin Hypercube Sampling to initialize a small-scale population, ensuring a broad search range. Subsequently, to reduce random fluctuations in searches, we propose a PE supplementary sampling strategy that identifies advantageous regions of the solution space, and performs precise supplementary sampling of the population. In the fitness evaluation stage, to enhance the noise resistance of the searched architectures, we propose a PE fitness evaluation strategy. This strategy first evaluates individual fitness separately using both original data and artificially constructed noise-augmented data, then combines the two fitness scores through a novel progressive formula to determine the final individual fitness. Experiments were conducted on two public datasets: the ABIDE I dataset (1,112 subjects, 17 sites), and ADHD-200 (776 subjects, 8 sites), using AAL/CC200 atlases. Results demonstrate that PE-RBNAS achieves state-of-the-art performance, with 72.61% accuracy on clean ABIDE I data (vs. 71.05% for MC-APSONAS) and 71.82% accuracy under 0.2 noise (vs. 68.15% for PSO-BNAS). The results indicate that, compared to other methods, the proposed method demonstrates better model performance and superior noise resistance.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Functional Brain Network (FBN) classification methods based on Neural Architecture Search (NAS) have been increasingly emerging, with their core advantage being the ability to automatically construct high-quality network architectures. However, existing methods exhibit poor robustness when dealing with FBNs that have inherent high-noise characteristics. To address these issues, we propose a robust NAS with progressive-enhanced strategies for FBN classification. Specifically, this method adopts Particle Swarm Optimization as the search method, while treating candidate architectures as individuals, and proposes two progressive-enhanced (PE) strategies to optimize the critical stages of population sampling and fitness evaluation. In the population sampling stage, we first utilize Latin Hypercube Sampling to initialize a small-scale population, ensuring a broad search range. Subsequently, to reduce random fluctuations in searches, we propose a PE supplementary sampling strategy that identifies advantageous regions of the solution space, and performs precise supplementary sampling of the population. In the fitness evaluation stage, to enhance the noise resistance of the searched architectures, we propose a PE fitness evaluation strategy. This strategy first evaluates individual fitness separately using both original data and artificially constructed noise-augmented data, then combines the two fitness scores through a novel progressive formula to determine the final individual fitness. Experiments were conducted on two public datasets: the ABIDE I dataset (1,112 subjects, 17 sites), and ADHD-200 (776 subjects, 8 sites), using AAL/CC200 atlases. Results demonstrate that PE-RBNAS achieves state-of-the-art performance, with 72.61% accuracy on clean ABIDE I data (vs. 71.05% for MC-APSONAS) and 71.82% accuracy under 0.2 noise (vs. 68.15% for PSO-BNAS). The results indicate that, compared to other methods, the proposed method demonstrates better model performance and superior noise resistance.
4.
K, Venkata Ratna Prabha; Bindu, Chinni Hima; Devi, K Rama
An interpretable deep learning approach for autism spectrum disorder detection in children using NASNet-mobile Journal Article
In: Biomedical Physics & Engineering Express, vol. 11, no. 4, pp. 045006, 2025.
@article{K_2025,
title = {An interpretable deep learning approach for autism spectrum disorder detection in children using NASNet-mobile},
author = {Venkata Ratna Prabha K and Chinni Hima Bindu and K Rama Devi},
url = {https://dx.doi.org/10.1088/2057-1976/addbe7},
doi = {10.1088/2057-1976/addbe7},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-01},
journal = {Biomedical Physics & Engineering Express},
volume = {11},
number = {4},
pages = {045006},
publisher = {IOP Publishing},
abstract = {Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental disorder featuring impaired social interactions and communication abilities engaging the individuals in a restrictive or repetitive behaviour. Though incurable early detection and intervention can reduce the severity of symptoms. Structural magnetic resonance imaging (sMRI) can improve diagnostic accuracy, facilitating early diagnosis to offer more tailored care. With the emergence of deep learning (DL), neuroimaging-based approaches for ASD diagnosis have been focused. However, many existing models lack interpretability of their decisions for diagnosis. The prime objective of this work is to perform ASD classification precisely and to interpret the classification process in a better way so as to discern the major features that are appropriate for the prediction of disorder. The proposed model employs neural architecture search network - mobile(NASNet-Mobile) model for ASD detection, which is integrated with an explainable artificial intelligence (XAI) technique called local interpretable model-agnostic explanations (LIME) for increased transparency of ASD classification. The model is trained on sMRI images of two age groups taken from autism brain imaging data exchange-I (ABIDE-I) dataset. The proposed model yielded accuracy of 0.9607, F1-score of 0.9614, specificity of 0.9774, sensitivity of 0.9451, negative predicted value (NPV) of 0.9429, positive predicted value (PPV) of 0.9783 and the diagnostic odds ratio of 745.59 for 2 to 11 years age group compared to 12 to 18 years group. These results are superior compared to other state of the art models Inception v3 and SqueezeNet.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental disorder featuring impaired social interactions and communication abilities engaging the individuals in a restrictive or repetitive behaviour. Though incurable early detection and intervention can reduce the severity of symptoms. Structural magnetic resonance imaging (sMRI) can improve diagnostic accuracy, facilitating early diagnosis to offer more tailored care. With the emergence of deep learning (DL), neuroimaging-based approaches for ASD diagnosis have been focused. However, many existing models lack interpretability of their decisions for diagnosis. The prime objective of this work is to perform ASD classification precisely and to interpret the classification process in a better way so as to discern the major features that are appropriate for the prediction of disorder. The proposed model employs neural architecture search network - mobile(NASNet-Mobile) model for ASD detection, which is integrated with an explainable artificial intelligence (XAI) technique called local interpretable model-agnostic explanations (LIME) for increased transparency of ASD classification. The model is trained on sMRI images of two age groups taken from autism brain imaging data exchange-I (ABIDE-I) dataset. The proposed model yielded accuracy of 0.9607, F1-score of 0.9614, specificity of 0.9774, sensitivity of 0.9451, negative predicted value (NPV) of 0.9429, positive predicted value (PPV) of 0.9783 and the diagnostic odds ratio of 745.59 for 2 to 11 years age group compared to 12 to 18 years group. These results are superior compared to other state of the art models Inception v3 and SqueezeNet.
5.
Wang, Chen; Long, Kaifang; Guo, Tiezheng; Yang, Qingwen; Liu, Yanyi; Li, Pan; Li, Zhi; Wang, Huihan; Ma, Lianbo; Wen, Yingyou
Enhancing Result Interpretability of Neural Architecture Search-Assisted Medical AI via Large Language Model Journal Article
In: IEEE Transactions on Emerging Topics in Computational Intelligence, pp. 1-11, 2025.
@article{11168895,
title = {Enhancing Result Interpretability of Neural Architecture Search-Assisted Medical AI via Large Language Model},
author = {Chen Wang and Kaifang Long and Tiezheng Guo and Qingwen Yang and Yanyi Liu and Pan Li and Zhi Li and Huihan Wang and Lianbo Ma and Yingyou Wen},
url = {https://ieeexplore.ieee.org/abstract/document/11168895},
doi = {10.1109/TETCI.2025.3607385},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {IEEE Transactions on Emerging Topics in Computational Intelligence},
pages = {1-11},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
6.
Zhang, Zijun Frank; Zhan, Huixin; Wu, Tinghui; Burns, Robert; Hundal, Jasreet; Costa, Helio A.
The Expanding Landscape of Neural Architectures and Their Impact in Biomedicine Journal Article
In: Annual Review of Biomedical Data Science, vol. 8, no. Volume 8, 2025, pp. 101-124, 2025, ISSN: 2574-3414.
@article{annurev:/content/journals/10.1146/annurev-biodatasci-103023-050856,
title = {The Expanding Landscape of Neural Architectures and Their Impact in Biomedicine},
author = {Zijun Frank Zhang and Huixin Zhan and Tinghui Wu and Robert Burns and Jasreet Hundal and Helio A. Costa},
url = {https://www.annualreviews.org/content/journals/10.1146/annurev-biodatasci-103023-050856},
doi = {https://doi.org/10.1146/annurev-biodatasci-103023-050856},
issn = {2574-3414},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Annual Review of Biomedical Data Science},
volume = {8},
number = {Volume 8, 2025},
pages = {101-124},
publisher = {Annual Reviews},
abstract = {Deep learning and artificial intelligence (AI) have seen explosive growth and success in biomedical applications in the last decade, largely due to the rapid development of deep neural networks and their underlying neural network (NN) architectures. Here, we explore biomedical deep learning and AI from the specific perspective of NN architectures. We discuss widely varying design principles of NN architectures, their use in particular biomedical applications, and the assumptions (often hidden) built into them. We explore neural architecture search techniques that automate the design of NN topology to optimize task performance. Advanced neural architectures are being developed for both molecular and healthcare applications, employing elements of graph networks, transformers, and interpretable NNs, and we discuss and summarize the design considerations and unique advantages of each architecture. Future advances will include the employment of multimodal language models and smaller highly focused mechanistic models that build on the success of today's large models."},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep learning and artificial intelligence (AI) have seen explosive growth and success in biomedical applications in the last decade, largely due to the rapid development of deep neural networks and their underlying neural network (NN) architectures. Here, we explore biomedical deep learning and AI from the specific perspective of NN architectures. We discuss widely varying design principles of NN architectures, their use in particular biomedical applications, and the assumptions (often hidden) built into them. We explore neural architecture search techniques that automate the design of NN topology to optimize task performance. Advanced neural architectures are being developed for both molecular and healthcare applications, employing elements of graph networks, transformers, and interpretable NNs, and we discuss and summarize the design considerations and unique advantages of each architecture. Future advances will include the employment of multimodal language models and smaller highly focused mechanistic models that build on the success of today's large models."
7.
Zhang, Hanyu; Tang, Lixin; Song, Xiangman
EM-NAS: A Cell-Based Evolutionary Multi-scale Neural Architecture Search for Medical Image Segmentation Proceedings Article
In: Proceedings of the Genetic and Evolutionary Computation Conference Companion, pp. 755–758, Association for Computing Machinery, NH Malaga Hotel, Malaga, Spain, 2025, ISBN: 9798400714641.
@inproceedings{10.1145/3712255.3726641,
title = {EM-NAS: A Cell-Based Evolutionary Multi-scale Neural Architecture Search for Medical Image Segmentation},
author = {Hanyu Zhang and Lixin Tang and Xiangman Song},
url = {https://doi.org/10.1145/3712255.3726641},
doi = {10.1145/3712255.3726641},
isbn = {9798400714641},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference Companion},
pages = {755–758},
publisher = {Association for Computing Machinery},
address = {NH Malaga Hotel, Malaga, Spain},
series = {GECCO '25 Companion},
abstract = {Accurate segmentation of medical images is a crucial part of pathology research and clinical practice. Currently, deep neural network (DNN)-based models are commonly used for medicad image segmentation. However, most DNN-based models are manually designed by experts which is a time-consuming and labor-intensive process as well as poorly generalized for different applications. Therefore, we propose EM-NAS, a novel evolutionary multi-scale NAS method with a cell-based design for medical image segmentation. Specifically, the structures of encoder and decoder cells are represented by directed acyclic graphs (DAGs). Moreover, some non-subsampling multi-scale decomposition operations are introduced into the candidate operation set to capture multi-scale, multi-directional and multi-frequency features of medical images. Subsequently, we specify the cell of the network through a well-designed discrete coding strategy, and a GA is used to search for the optimal network architecture. The experimental results demonstrate the effectiveness of the proposed model on two public datasets.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Accurate segmentation of medical images is a crucial part of pathology research and clinical practice. Currently, deep neural network (DNN)-based models are commonly used for medicad image segmentation. However, most DNN-based models are manually designed by experts which is a time-consuming and labor-intensive process as well as poorly generalized for different applications. Therefore, we propose EM-NAS, a novel evolutionary multi-scale NAS method with a cell-based design for medical image segmentation. Specifically, the structures of encoder and decoder cells are represented by directed acyclic graphs (DAGs). Moreover, some non-subsampling multi-scale decomposition operations are introduced into the candidate operation set to capture multi-scale, multi-directional and multi-frequency features of medical images. Subsequently, we specify the cell of the network through a well-designed discrete coding strategy, and a GA is used to search for the optimal network architecture. The experimental results demonstrate the effectiveness of the proposed model on two public datasets.
8.
Babuc, Diogen; Fortiş, Teodor-Florin
AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases Proceedings Article
In: Barolli, Leonard; Enokido, Tomoya; Woungang, Isaac (Ed.): Complex, Intelligent and Software Intensive Systems, pp. 24–35, Springer Nature Switzerland, Cham, 2025, ISBN: 978-3-031-96099-4.
@inproceedings{10.1007/978-3-031-96099-4_3b,
title = {AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases},
author = {Diogen Babuc and Teodor-Florin Fortiş},
editor = {Leonard Barolli and Tomoya Enokido and Isaac Woungang},
isbn = {978-3-031-96099-4},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Complex, Intelligent and Software Intensive Systems},
pages = {24–35},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.
9.
Chen, Xi; Lv, Jiahuan; Wang, Zeyu; Qin, Genggeng; Zhou, Zhiguo
In: Journal of Biomedical Informatics, vol. 169, pp. 104869, 2025, ISSN: 1532-0464.
@article{CHEN2025104869,
title = {Adaptive-AutoMO: A domain adaptive automated multiobjective neural network for reliable lesion malignancy prediction via digital breast tomosynthesis},
author = {Xi Chen and Jiahuan Lv and Zeyu Wang and Genggeng Qin and Zhiguo Zhou},
url = {https://www.sciencedirect.com/science/article/pii/S153204642500098X},
doi = {https://doi.org/10.1016/j.jbi.2025.104869},
issn = {1532-0464},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Journal of Biomedical Informatics},
volume = {169},
pages = {104869},
abstract = {Early diagnosis of breast cancer remains a significant global health challenge, and the potential use of deep learning in Digital Breast Tomosynthesis (DBT) based breast cancer diagnosis is a promising avenue. To address data scarcity and domain shift problems in building a lesion malignancy predictive model, we proposed a domain adaptive automated multiobjective neural network (Adaptive-AutoMO) for reliable lesion malignancy prediction via DBT. Adaptive-AutoMO addresses three key challenges simultaneously, they are: privacy preserving, credibility measurement, and balance, which consists of training, adaptation and testing stages. In the training stage, we developed a multiobjective immune neural architecture search algorithm (MINAS) to generate a Pareto-optimal model set with balanced sensitivity and specificity and introduced a Bayesian optimization algorithm to optimize the hyperparameters. In the adaptation stage, a semi-supervised domain adaptive feature network based on maximum mean discrepancy (MMD-SSDAF) was designed, which can make the balanced models adaptable to the target domain and preserve the data privacy in the source domain. In the testing stage, we proposed an evidence reasoning method based on entropy (ERE) that can fuse multiple adapted models and estimate uncertainty to improve the model credibility. The experiments on two DBT image datasets (source and target domain datasets) revealed that Adaptive-AutoMO outperformed ResNet-18, DenseNet-121, and other available domain adaptive models. Meanwhile, the removal of high uncertainty samples resulted in a performance improvement in the target domain. These experiments affirmed that Adaptive-AutoMO can not only enhance model’s performance, but also preserve privacy in the source domain data, boost model credibility, and achieve a balance between sensitivity and specificity.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Early diagnosis of breast cancer remains a significant global health challenge, and the potential use of deep learning in Digital Breast Tomosynthesis (DBT) based breast cancer diagnosis is a promising avenue. To address data scarcity and domain shift problems in building a lesion malignancy predictive model, we proposed a domain adaptive automated multiobjective neural network (Adaptive-AutoMO) for reliable lesion malignancy prediction via DBT. Adaptive-AutoMO addresses three key challenges simultaneously, they are: privacy preserving, credibility measurement, and balance, which consists of training, adaptation and testing stages. In the training stage, we developed a multiobjective immune neural architecture search algorithm (MINAS) to generate a Pareto-optimal model set with balanced sensitivity and specificity and introduced a Bayesian optimization algorithm to optimize the hyperparameters. In the adaptation stage, a semi-supervised domain adaptive feature network based on maximum mean discrepancy (MMD-SSDAF) was designed, which can make the balanced models adaptable to the target domain and preserve the data privacy in the source domain. In the testing stage, we proposed an evidence reasoning method based on entropy (ERE) that can fuse multiple adapted models and estimate uncertainty to improve the model credibility. The experiments on two DBT image datasets (source and target domain datasets) revealed that Adaptive-AutoMO outperformed ResNet-18, DenseNet-121, and other available domain adaptive models. Meanwhile, the removal of high uncertainty samples resulted in a performance improvement in the target domain. These experiments affirmed that Adaptive-AutoMO can not only enhance model’s performance, but also preserve privacy in the source domain data, boost model credibility, and achieve a balance between sensitivity and specificity.
10.
Babuc, Diogen; Fortiş, Teodor-Florin
AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases Proceedings Article
In: Barolli, Leonard; Enokido, Tomoya; Woungang, Isaac (Ed.): Complex, Intelligent and Software Intensive Systems, pp. 24–35, Springer Nature Switzerland, Cham, 2025, ISBN: 978-3-031-96099-4.
@inproceedings{10.1007/978-3-031-96099-4_3,
title = {AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases},
author = {Diogen Babuc and Teodor-Florin Fortiş},
editor = {Leonard Barolli and Tomoya Enokido and Isaac Woungang},
url = {https://link.springer.com/chapter/10.1007/978-3-031-96099-4_3},
isbn = {978-3-031-96099-4},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Complex, Intelligent and Software Intensive Systems},
pages = {24–35},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.
11.
Benmeziane, Hadjer; GACEM, Abderaouf; Maghraoui, Kaoutar El; Benmeziane, Sara
Efficient Graph Neural Architecture Search for Medical Imaging in Real-World Clinical Settings Proceedings Article
In: 4th Muslims in ML Workshop co-located with ICML 2025, 2025.
@inproceedings{<LineBreak>benmeziane2025efficient,
title = {Efficient Graph Neural Architecture Search for Medical Imaging in Real-World Clinical Settings},
author = {Hadjer Benmeziane and Abderaouf GACEM and Kaoutar El Maghraoui and Sara Benmeziane},
url = {https://openreview.net/forum?id=Em7jjcxHAR},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {4th Muslims in ML Workshop co-located with ICML 2025},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
12.
Shawi, Radwa El
AgingFedNAS: Aging Evolution Federated Deep Learning for Architecture and Hyperparameter Search Proceedings Article
In: Wu, Xintao; Spiliopoulou, Myra; Wang, Can; Kumar, Vipin; Cao, Longbing; Wu, Yanqiu; Yao, Yu; Wu, Zhangkai (Ed.): Advances in Knowledge Discovery and Data Mining, pp. 107–119, Springer Nature Singapore, Singapore, 2025, ISBN: 978-981-96-8173-0.
@inproceedings{10.1007/978-981-96-8173-0_9,
title = {AgingFedNAS: Aging Evolution Federated Deep Learning for Architecture and Hyperparameter Search},
author = {Radwa El Shawi},
editor = {Xintao Wu and Myra Spiliopoulou and Can Wang and Vipin Kumar and Longbing Cao and Yanqiu Wu and Yu Yao and Zhangkai Wu},
url = {https://link.springer.com/chapter/10.1007/978-981-96-8173-0_9},
isbn = {978-981-96-8173-0},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Advances in Knowledge Discovery and Data Mining},
pages = {107–119},
publisher = {Springer Nature Singapore},
address = {Singapore},
abstract = {Despite advancements in Automatic Machine Learning (AutoML), industries encounter challenges in implementation due to data privacy concerns and the costs of centralized data storage. Federated Learning (FL) provides a decentralized approach, allowing multiple clients to collaboratively train models without sharing their datasets. However, many existing FL techniques utilize pre-defined model architectures from centralized environments, which may not be optimal for the non-iid data distributions commonly found among FL clients. This paper introduces AgingFedNAS, a framework designed to automate model design in FL by employing an evolutionary approach to jointly optimize neural architectures and hyperparameters. Comprehensive experiments conducted on heterogeneous data splits from CIFAR-10, Shakespeare, FEMNIST, Tiny-ImageNet, and a medical breast density classification dataset demonstrate that AgingFedNAS outperforms state-of-the-art FL frameworks, including FedAvg, FEATHERS, FedEx, and FedNAS, particularly in non-iid conditions. Notably, AgingFedNAS achieves an accuracy of 89.8% on CIFAR-10, exceeding the best baseline, FEATHERS, by 3.78%. In the breast density classification task, it surpasses FedNAS by 1.3%, achieving up to 3.2% higher accuracy for specific clients under non-iid scenarios. Additionally, in highly heterogeneous data environments, AgingFedNAS shows a 2.3% accuracy improvement on CIFAR-10 compared to the top-performing baseline.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Despite advancements in Automatic Machine Learning (AutoML), industries encounter challenges in implementation due to data privacy concerns and the costs of centralized data storage. Federated Learning (FL) provides a decentralized approach, allowing multiple clients to collaboratively train models without sharing their datasets. However, many existing FL techniques utilize pre-defined model architectures from centralized environments, which may not be optimal for the non-iid data distributions commonly found among FL clients. This paper introduces AgingFedNAS, a framework designed to automate model design in FL by employing an evolutionary approach to jointly optimize neural architectures and hyperparameters. Comprehensive experiments conducted on heterogeneous data splits from CIFAR-10, Shakespeare, FEMNIST, Tiny-ImageNet, and a medical breast density classification dataset demonstrate that AgingFedNAS outperforms state-of-the-art FL frameworks, including FedAvg, FEATHERS, FedEx, and FedNAS, particularly in non-iid conditions. Notably, AgingFedNAS achieves an accuracy of 89.8% on CIFAR-10, exceeding the best baseline, FEATHERS, by 3.78%. In the breast density classification task, it surpasses FedNAS by 1.3%, achieving up to 3.2% higher accuracy for specific clients under non-iid scenarios. Additionally, in highly heterogeneous data environments, AgingFedNAS shows a 2.3% accuracy improvement on CIFAR-10 compared to the top-performing baseline.
13.
Hu, Mengxiang; Li, Junchi; Dong, Yongquan; Zhang, Zichen; Liu, Weifan; Zhang, Peilin; Ping, Yuchao; Jiang, Le; Yu, Zekuan
Mixed-GGNAS: Mixed Search-space NAS based on genetic algorithm combined with gradient descent for medical image segmentation Journal Article
In: Expert Systems with Applications, vol. 289, pp. 128338, 2025, ISSN: 0957-4174.
@article{HU2025128338,
title = {Mixed-GGNAS: Mixed Search-space NAS based on genetic algorithm combined with gradient descent for medical image segmentation},
author = {Mengxiang Hu and Junchi Li and Yongquan Dong and Zichen Zhang and Weifan Liu and Peilin Zhang and Yuchao Ping and Le Jiang and Zekuan Yu},
url = {https://www.sciencedirect.com/science/article/pii/S0957417425019578},
doi = {https://doi.org/10.1016/j.eswa.2025.128338},
issn = {0957-4174},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Expert Systems with Applications},
volume = {289},
pages = {128338},
abstract = {Medical images segmentation is a pivotal procedure, playing a fundamental role in computer-assisted diagnosis and treatment. Despite the significant advancements in methods leveraging deep learning for this purpose, many networks still face challenges related to efficiency, often requiring substantial time and manual efforts. Neural architecture search (NAS) has gained considerable attention in the automated design of neural networks. This study introduces a new NAS method, Mixed-GGNAS, a Mixed Search-space NAS method based on Genetic algorithm combined with Gradient descent. Our approach creatively combines manually designed network blocks with DARTS blocks to construct a mixed search space. We then employ a method that integrates genetic algorithms and gradient descent to concurrently search for both block types and internal operations within the block. Within a U-shaped network framework, we propose a Multi-feature fusion strategy based on Vision Transformer (ViT) and search for hyperparameters of it. Additionally, we employ a Multi-scale mixed loss function to enhance the model’s ability to learn features at various scales. Experimental results demonstrate that the proposed approach outperforms or is comparable to the state-of-the-art NAS methods and manually designed Networks. Ablation studies conducted on two datasets further validate the method’s efficacy in enhancing model performance. The code is available at https://github.com/Hmxki/Mixed-GGNAS.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Medical images segmentation is a pivotal procedure, playing a fundamental role in computer-assisted diagnosis and treatment. Despite the significant advancements in methods leveraging deep learning for this purpose, many networks still face challenges related to efficiency, often requiring substantial time and manual efforts. Neural architecture search (NAS) has gained considerable attention in the automated design of neural networks. This study introduces a new NAS method, Mixed-GGNAS, a Mixed Search-space NAS method based on Genetic algorithm combined with Gradient descent. Our approach creatively combines manually designed network blocks with DARTS blocks to construct a mixed search space. We then employ a method that integrates genetic algorithms and gradient descent to concurrently search for both block types and internal operations within the block. Within a U-shaped network framework, we propose a Multi-feature fusion strategy based on Vision Transformer (ViT) and search for hyperparameters of it. Additionally, we employ a Multi-scale mixed loss function to enhance the model’s ability to learn features at various scales. Experimental results demonstrate that the proposed approach outperforms or is comparable to the state-of-the-art NAS methods and manually designed Networks. Ablation studies conducted on two datasets further validate the method’s efficacy in enhancing model performance. The code is available at https://github.com/Hmxki/Mixed-GGNAS.
14.
Sun, Yu; Zhang, Xianglin; Dong, Liang; Liu, Ning
In: Applied Soft Computing, vol. 179, pp. 113279, 2025, ISSN: 1568-4946.
@article{SUN2025113279,
title = {Multi-objective evolutionary neural architecture search for medical image analysis using transformer and large language models in advancing public health},
author = {Yu Sun and Xianglin Zhang and Liang Dong and Ning Liu},
url = {https://www.sciencedirect.com/science/article/pii/S1568494625005903},
doi = {https://doi.org/10.1016/j.asoc.2025.113279},
issn = {1568-4946},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Applied Soft Computing},
volume = {179},
pages = {113279},
abstract = {The rapid growth of medical imaging data in modern healthcare networks demands sophisticated automated analysis methods that can maintain high accuracy while operating efficiently at scale. Current approaches using transformers and large language models (LLMs) face challenges balancing computational requirements with diagnostic precision across diverse healthcare settings. This paper presents TransMed-NAS (transformer medical neural architecture search), a multi-objective evolutionary neural architecture search framework that automatically discovers efficient hybrid architectures by integrating transformers and LLMs for medical image segmentation. Our approach leverages evolutionary computation to optimize segmentation accuracy and computational efficiency while incorporating medical domain knowledge through LLM guidance. The framework introduces several innovations: a hierarchical channel selection strategy that preserves clinically relevant features, a weight entanglement mechanism that accelerates architecture search through intelligent knowledge transfer, and a surrogate model acceleration technique that reduces computational overhead while maintaining reliability. Experimental results on the ISIC 2020 dataset demonstrate TransMed-NAS’s superior performance compared to state-of-the-art methods. Our small model variant achieves competitive accuracy (0.934 Dice score) with only 0.82M parameters, while our large variant establishes new benchmarks (0.947 Dice score) with significantly reduced computational requirements. Ablation studies confirm the effectiveness of each component, particularly highlighting how LLM integration enhances architecture search efficiency and clinical relevance. These results demonstrate TransMed-NAS’s potential to advance automated medical image analysis in resource-diverse healthcare settings, making sophisticated diagnostic capabilities more accessible to underserved communities.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The rapid growth of medical imaging data in modern healthcare networks demands sophisticated automated analysis methods that can maintain high accuracy while operating efficiently at scale. Current approaches using transformers and large language models (LLMs) face challenges balancing computational requirements with diagnostic precision across diverse healthcare settings. This paper presents TransMed-NAS (transformer medical neural architecture search), a multi-objective evolutionary neural architecture search framework that automatically discovers efficient hybrid architectures by integrating transformers and LLMs for medical image segmentation. Our approach leverages evolutionary computation to optimize segmentation accuracy and computational efficiency while incorporating medical domain knowledge through LLM guidance. The framework introduces several innovations: a hierarchical channel selection strategy that preserves clinically relevant features, a weight entanglement mechanism that accelerates architecture search through intelligent knowledge transfer, and a surrogate model acceleration technique that reduces computational overhead while maintaining reliability. Experimental results on the ISIC 2020 dataset demonstrate TransMed-NAS’s superior performance compared to state-of-the-art methods. Our small model variant achieves competitive accuracy (0.934 Dice score) with only 0.82M parameters, while our large variant establishes new benchmarks (0.947 Dice score) with significantly reduced computational requirements. Ablation studies confirm the effectiveness of each component, particularly highlighting how LLM integration enhances architecture search efficiency and clinical relevance. These results demonstrate TransMed-NAS’s potential to advance automated medical image analysis in resource-diverse healthcare settings, making sophisticated diagnostic capabilities more accessible to underserved communities.
15.
Zechen, Zheng; Xuelei, He; Fengjun, Zhao; Xiaowei, He
PSNAS-Net: Hybrid gradient-physical optimizationfor efficient neural architecture search in customized medical imaging analysis Journal Article
In: Expert Systems with Applications, vol. 288, pp. 128155, 2025, ISSN: 0957-4174.
@article{ZECHEN2025128155,
title = {PSNAS-Net: Hybrid gradient-physical optimizationfor efficient neural architecture search in customized medical imaging analysis},
author = {Zheng Zechen and He Xuelei and Zhao Fengjun and He Xiaowei},
url = {https://www.sciencedirect.com/science/article/pii/S0957417425017750},
doi = {https://doi.org/10.1016/j.eswa.2025.128155},
issn = {0957-4174},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Expert Systems with Applications},
volume = {288},
pages = {128155},
abstract = {Neural architecture search (NAS) facilitates the automated construction of neural networks tailored to specific tasks and requirements, resulting in models that are more closely aligned with the target task’s demands. However, in many studies, the extensive design space, high search costs, and time-consuming evaluation calculations render NAS impractical for numerous medical data tasks. Addressing these challenges, this study introduces an efficient algorithm for searching deep learning architectures. Initially, we propose 19 fundamental rules to streamline the design space, thereby reducing its scale. To improve the efficiency of the algorithm, we designed a NAS framework (PSNAS-Net) for convolutional neural networks and VisionTransformer, which consists of two search stages: Firstly, the improved powell algorithm is used to determine the model range, and the population-based simulated annealing algorithm is utilized to expedite the search for the final model. During the neural architecture search process, we consider accuracy, parameters, FLOPs, and model stability as comprehensive evaluation objectives, we designed a robust, flexible, and comprehensive metric for model evaluation. The experimental results demonstrate that PSNAS-Net achieves significantly shorter search times (0.05-1.47 GPU Days) compared to 19 existing NAS methods, while discovering compact models (as small as 0.11M) with superior performance across five medical image benchmarks. This study offers a viable approach for model search that accommodates individualized requirements.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neural architecture search (NAS) facilitates the automated construction of neural networks tailored to specific tasks and requirements, resulting in models that are more closely aligned with the target task’s demands. However, in many studies, the extensive design space, high search costs, and time-consuming evaluation calculations render NAS impractical for numerous medical data tasks. Addressing these challenges, this study introduces an efficient algorithm for searching deep learning architectures. Initially, we propose 19 fundamental rules to streamline the design space, thereby reducing its scale. To improve the efficiency of the algorithm, we designed a NAS framework (PSNAS-Net) for convolutional neural networks and VisionTransformer, which consists of two search stages: Firstly, the improved powell algorithm is used to determine the model range, and the population-based simulated annealing algorithm is utilized to expedite the search for the final model. During the neural architecture search process, we consider accuracy, parameters, FLOPs, and model stability as comprehensive evaluation objectives, we designed a robust, flexible, and comprehensive metric for model evaluation. The experimental results demonstrate that PSNAS-Net achieves significantly shorter search times (0.05-1.47 GPU Days) compared to 19 existing NAS methods, while discovering compact models (as small as 0.11M) with superior performance across five medical image benchmarks. This study offers a viable approach for model search that accommodates individualized requirements.
16.
Becktepe, Jannis; Hennig, Leona; Oeltze-Jafra, Steffen; Lindauer, Marius
Auto-nnU-Net: Towards Automated Medical Image Segmentation Technical Report
2025.
@techreport{becktepe2025autonnunetautomatedmedicalimage,
title = {Auto-nnU-Net: Towards Automated Medical Image Segmentation},
author = {Jannis Becktepe and Leona Hennig and Steffen Oeltze-Jafra and Marius Lindauer},
url = {https://arxiv.org/abs/2505.16561},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
17.
Ardila, Diego Páez; Carvalho, Thiago; Saavedra, Santiago Vasquez; Niño, Cesar Valencia; Figueiredo, Karla; Vellasco, Marley
Quantum-Inspired NAS With Attention-Based Search Spaces in Medical Applications Proceedings Article
In: 2025 IEEE Symposium on Computational Intelligence in Health and Medicine Companion (CIHM Companion), pp. 1-5, 2025.
@inproceedings{11002695,
title = {Quantum-Inspired NAS With Attention-Based Search Spaces in Medical Applications},
author = {Diego Páez Ardila and Thiago Carvalho and Santiago Vasquez Saavedra and Cesar Valencia Niño and Karla Figueiredo and Marley Vellasco},
doi = {10.1109/CIHMCompanion65205.2025.11002695},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {2025 IEEE Symposium on Computational Intelligence in Health and Medicine Companion (CIHM Companion)},
pages = {1-5},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
18.
Cao, Bin; Deng, Huanyu; Hao, Yiming; Luo, Xiao
Multi-view information fusion based on federated multi-objective neural architecture search for MRI semantic segmentation Journal Article
In: Information Fusion, vol. 123, pp. 103301, 2025, ISSN: 1566-2535.
@article{CAO2025103301,
title = {Multi-view information fusion based on federated multi-objective neural architecture search for MRI semantic segmentation},
author = {Bin Cao and Huanyu Deng and Yiming Hao and Xiao Luo},
url = {https://www.sciencedirect.com/science/article/pii/S1566253525003744},
doi = {https://doi.org/10.1016/j.inffus.2025.103301},
issn = {1566-2535},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Information Fusion},
volume = {123},
pages = {103301},
abstract = {With the rapid development of artificial intelligence, medical image semantic segmentation is being used more widely. However, centralized training can lead to privacy risks. At the same time, MRI provides multiple views that together describe the anatomical structure of a lesion, but a single view may not fully capture all features. Therefore, integrating multi-view information in a federated learning setting is a key challenge for improving model generalization. This study combines federated learning, neural architecture search (NAS) and data fusion techniques to address issues related to data privacy, cross-institutional data distribution differences and multi-view information fusion in medical imaging. To achieve this, we propose the FL-MONAS framework, which leverages the advantages of NAS and federated learning. It uses a Pareto-frontier-based multi-objective optimization strategy to effectively combine 2D MRI with 3D anatomical structures, improving model performance while ensuring data privacy. Experimental results show that FL-MONAS maintains strong segmentation performance even in non-IID scenarios, providing an efficient and privacy-friendly solution for medical image analysis.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
With the rapid development of artificial intelligence, medical image semantic segmentation is being used more widely. However, centralized training can lead to privacy risks. At the same time, MRI provides multiple views that together describe the anatomical structure of a lesion, but a single view may not fully capture all features. Therefore, integrating multi-view information in a federated learning setting is a key challenge for improving model generalization. This study combines federated learning, neural architecture search (NAS) and data fusion techniques to address issues related to data privacy, cross-institutional data distribution differences and multi-view information fusion in medical imaging. To achieve this, we propose the FL-MONAS framework, which leverages the advantages of NAS and federated learning. It uses a Pareto-frontier-based multi-objective optimization strategy to effectively combine 2D MRI with 3D anatomical structures, improving model performance while ensuring data privacy. Experimental results show that FL-MONAS maintains strong segmentation performance even in non-IID scenarios, providing an efficient and privacy-friendly solution for medical image analysis.
19.
Mecharbat, Lotfi Abdelkrim; Almakky, Ibrahim; Takac, Martin; Yaqub, Mohammad
MedNNS: Supernet-based Medical Task-Adaptive Neural Network Search Technical Report
2025.
@techreport{mecharbat2025mednnssupernetbasedmedicaltaskadaptiveb,
title = {MedNNS: Supernet-based Medical Task-Adaptive Neural Network Search},
author = {Lotfi Abdelkrim Mecharbat and Ibrahim Almakky and Martin Takac and Mohammad Yaqub},
url = {https://arxiv.org/abs/2504.15865},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
20.
Wani, M. Arif; Sultan, Bisma; Ali, Sarwat; Sofi, Mukhtar Ahmad
Introduction to Deep Learning Applications Book Chapter
In: Advances in Deep Learning, Volume 2, pp. 1–14, Springer Nature Singapore, Singapore, 2025, ISBN: 978-981-96-3498-9.
@inbook{Wani2025c,
title = {Introduction to Deep Learning Applications},
author = {M. Arif Wani and Bisma Sultan and Sarwat Ali and Mukhtar Ahmad Sofi},
url = {https://doi.org/10.1007/978-981-96-3498-9_1},
doi = {10.1007/978-981-96-3498-9_1},
isbn = {978-981-96-3498-9},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Advances in Deep Learning, Volume 2},
pages = {1–14},
publisher = {Springer Nature Singapore},
address = {Singapore},
abstract = {Deep learning has emerged as a cornerstone of modern Artificial Intelligence (AI), offering unparalleled performance across a variety of complex tasks. Its ability to automatically learn features from vast amounts of data has transformed industries ranging from computer vision to natural language processing. Among the numerous applications of deep learning, three areas have witnessed significant progress: Neural Architecture Search (NAS), steganography, and medical applications. This chapter explores application of deep learning architectures in these three areas.},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Deep learning has emerged as a cornerstone of modern Artificial Intelligence (AI), offering unparalleled performance across a variety of complex tasks. Its ability to automatically learn features from vast amounts of data has transformed industries ranging from computer vision to natural language processing. Among the numerous applications of deep learning, three areas have witnessed significant progress: Neural Architecture Search (NAS), steganography, and medical applications. This chapter explores application of deep learning architectures in these three areas.
2026
159.
Sun, Xiaoxue; Wang, Hongpeng; Song, Pei-Cheng
Compact Training-Free NAS with Alternating Evolution Game for Medical Image Segmentation Proceedings Article
In: Gee, James C.; Alexander, Daniel C.; Hong, Jaesung; Iglesias, Juan Eugenio; Sudre, Carole H.; Venkataraman, Archana; Golland, Polina; Kim, Jong Hyo; Park, Jinah (Ed.): Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, pp. 108–118, Springer Nature Switzerland, Cham, 2026, ISBN: 978-3-032-05325-1.
@inproceedings{10.1007/978-3-032-05325-1_11,
title = {Compact Training-Free NAS with Alternating Evolution Game for Medical Image Segmentation},
author = {Xiaoxue Sun and Hongpeng Wang and Pei-Cheng Song},
editor = {James C. Gee and Daniel C. Alexander and Jaesung Hong and Juan Eugenio Iglesias and Carole H. Sudre and Archana Venkataraman and Polina Golland and Jong Hyo Kim and Jinah Park},
url = {https://link.springer.com/chapter/10.1007/978-3-032-05325-1_11},
isbn = {978-3-032-05325-1},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
booktitle = {Medical Image Computing and Computer Assisted Intervention – MICCAI 2025},
pages = {108–118},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) has shown significant potential in designing deep neural networks for medical image segmentation. However, even emerging training-free NAS frameworks often incur substantial computational costs and lengthy search times. To address the critical challenges of computational efficiency and architecture interpretability, the paper proposes a compact training-free NAS framework based on an Alternating Evolution Game (AEG-cTFNAS). The proposed method alternates the search and contribution evaluation of the encoder and decoder within the UNet architecture via alternating games. It employs a truncated normal distribution for compact encoding, sampling, and updating to minimize computational overhead, while Bayesian inference is utilized to estimate the contribution of each block, adaptively adjusting the search strategy and facilitating process visualization. Experimental results on two benchmark datasets reveal that AEG-cTFNAS outperforms both manually designed architectures and NAS-based algorithms, underscoring its efficacy and potential on medical image segmentation. Code is available at https://github.com/spcity/AEG-cTFNAS.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) has shown significant potential in designing deep neural networks for medical image segmentation. However, even emerging training-free NAS frameworks often incur substantial computational costs and lengthy search times. To address the critical challenges of computational efficiency and architecture interpretability, the paper proposes a compact training-free NAS framework based on an Alternating Evolution Game (AEG-cTFNAS). The proposed method alternates the search and contribution evaluation of the encoder and decoder within the UNet architecture via alternating games. It employs a truncated normal distribution for compact encoding, sampling, and updating to minimize computational overhead, while Bayesian inference is utilized to estimate the contribution of each block, adaptively adjusting the search strategy and facilitating process visualization. Experimental results on two benchmark datasets reveal that AEG-cTFNAS outperforms both manually designed architectures and NAS-based algorithms, underscoring its efficacy and potential on medical image segmentation. Code is available at https://github.com/spcity/AEG-cTFNAS.
158.
Pandala, Madhavi Latha; Periyanayagi, S.
Optimal explainable vision transformer framework for skin cancer diagnosis with neural architecture search feature learning Journal Article
In: Biomedical Signal Processing and Control, vol. 112, pp. 108723, 2026, ISSN: 1746-8094.
@article{PANDALA2026108723,
title = {Optimal explainable vision transformer framework for skin cancer diagnosis with neural architecture search feature learning},
author = {Madhavi Latha Pandala and S. Periyanayagi},
url = {https://www.sciencedirect.com/science/article/pii/S1746809425012340},
doi = {https://doi.org/10.1016/j.bspc.2025.108723},
issn = {1746-8094},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
journal = {Biomedical Signal Processing and Control},
volume = {112},
pages = {108723},
abstract = {Skin cancer is one of the most prevalent and life-threatening diseases worldwide, making early and accurate detection crucial for improving patient survival rates. Traditional diagnostic methods rely on manual examination by dermatologists, which is subjective and time-consuming. To address these challenges, this research presents an advanced Optimal XAI based Skin Cancer Classification Network (OXAI-SCC-Net) framework for automated skin cancer detection and classification. The proposed methodology integrates multiple novel techniques to enhance accuracy and robustness. Initially, a Neural Architecture Search-Large Network (NASL-Net) is employed for feature extraction, leveraging automated deep-learning architecture search to optimize feature learning. To tackle class imbalance, a Support Vector Machine-Adopted Synthetic Oversampling Technique (SSOT) is utilized, which improves upon Synthetic Minority Oversampling Technic (SMOTE) by generating synthetic samples based on Support Vector Machine (SVM) decision boundaries, ensuring a balanced dataset. Further, Hippopotamus Optimization Algorithm with Explainable Artificial Intelligence (HOA-XAI) is applied for feature selection, reducing computational complexity by selecting the most informative features while minimizing redundant ones. Finally, a Vison Transformer Convolutional Neural Network (VT-CNN) classifier is trained on the optimized dataset to classify skin lesions into different categories. The proposed OXAI-SCC-Net method achieved an accuracy of 99.12%, with precision, recall, and F1-score each at 99.13% on ISIC-2019 dataset. This indicates highly consistent and reliable performance across all evaluation metrics compared to state-of-art approaches.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Skin cancer is one of the most prevalent and life-threatening diseases worldwide, making early and accurate detection crucial for improving patient survival rates. Traditional diagnostic methods rely on manual examination by dermatologists, which is subjective and time-consuming. To address these challenges, this research presents an advanced Optimal XAI based Skin Cancer Classification Network (OXAI-SCC-Net) framework for automated skin cancer detection and classification. The proposed methodology integrates multiple novel techniques to enhance accuracy and robustness. Initially, a Neural Architecture Search-Large Network (NASL-Net) is employed for feature extraction, leveraging automated deep-learning architecture search to optimize feature learning. To tackle class imbalance, a Support Vector Machine-Adopted Synthetic Oversampling Technique (SSOT) is utilized, which improves upon Synthetic Minority Oversampling Technic (SMOTE) by generating synthetic samples based on Support Vector Machine (SVM) decision boundaries, ensuring a balanced dataset. Further, Hippopotamus Optimization Algorithm with Explainable Artificial Intelligence (HOA-XAI) is applied for feature selection, reducing computational complexity by selecting the most informative features while minimizing redundant ones. Finally, a Vison Transformer Convolutional Neural Network (VT-CNN) classifier is trained on the optimized dataset to classify skin lesions into different categories. The proposed OXAI-SCC-Net method achieved an accuracy of 99.12%, with precision, recall, and F1-score each at 99.13% on ISIC-2019 dataset. This indicates highly consistent and reliable performance across all evaluation metrics compared to state-of-art approaches.
157.
Wang, Xingyu; Ji, Junzhong; Liu, Gan; Xiao, Yadong
PE-RBNAS: A robust neural architecture search with progressive-enhanced strategies for brain network classification Journal Article
In: Medical Image Analysis, vol. 107, pp. 103813, 2026, ISSN: 1361-8415.
@article{WANG2026103813,
title = {PE-RBNAS: A robust neural architecture search with progressive-enhanced strategies for brain network classification},
author = {Xingyu Wang and Junzhong Ji and Gan Liu and Yadong Xiao},
url = {https://www.sciencedirect.com/science/article/pii/S1361841525003597},
doi = {https://doi.org/10.1016/j.media.2025.103813},
issn = {1361-8415},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
journal = {Medical Image Analysis},
volume = {107},
pages = {103813},
abstract = {Functional Brain Network (FBN) classification methods based on Neural Architecture Search (NAS) have been increasingly emerging, with their core advantage being the ability to automatically construct high-quality network architectures. However, existing methods exhibit poor robustness when dealing with FBNs that have inherent high-noise characteristics. To address these issues, we propose a robust NAS with progressive-enhanced strategies for FBN classification. Specifically, this method adopts Particle Swarm Optimization as the search method, while treating candidate architectures as individuals, and proposes two progressive-enhanced (PE) strategies to optimize the critical stages of population sampling and fitness evaluation. In the population sampling stage, we first utilize Latin Hypercube Sampling to initialize a small-scale population, ensuring a broad search range. Subsequently, to reduce random fluctuations in searches, we propose a PE supplementary sampling strategy that identifies advantageous regions of the solution space, and performs precise supplementary sampling of the population. In the fitness evaluation stage, to enhance the noise resistance of the searched architectures, we propose a PE fitness evaluation strategy. This strategy first evaluates individual fitness separately using both original data and artificially constructed noise-augmented data, then combines the two fitness scores through a novel progressive formula to determine the final individual fitness. Experiments were conducted on two public datasets: the ABIDE I dataset (1,112 subjects, 17 sites), and ADHD-200 (776 subjects, 8 sites), using AAL/CC200 atlases. Results demonstrate that PE-RBNAS achieves state-of-the-art performance, with 72.61% accuracy on clean ABIDE I data (vs. 71.05% for MC-APSONAS) and 71.82% accuracy under 0.2 noise (vs. 68.15% for PSO-BNAS). The results indicate that, compared to other methods, the proposed method demonstrates better model performance and superior noise resistance.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Functional Brain Network (FBN) classification methods based on Neural Architecture Search (NAS) have been increasingly emerging, with their core advantage being the ability to automatically construct high-quality network architectures. However, existing methods exhibit poor robustness when dealing with FBNs that have inherent high-noise characteristics. To address these issues, we propose a robust NAS with progressive-enhanced strategies for FBN classification. Specifically, this method adopts Particle Swarm Optimization as the search method, while treating candidate architectures as individuals, and proposes two progressive-enhanced (PE) strategies to optimize the critical stages of population sampling and fitness evaluation. In the population sampling stage, we first utilize Latin Hypercube Sampling to initialize a small-scale population, ensuring a broad search range. Subsequently, to reduce random fluctuations in searches, we propose a PE supplementary sampling strategy that identifies advantageous regions of the solution space, and performs precise supplementary sampling of the population. In the fitness evaluation stage, to enhance the noise resistance of the searched architectures, we propose a PE fitness evaluation strategy. This strategy first evaluates individual fitness separately using both original data and artificially constructed noise-augmented data, then combines the two fitness scores through a novel progressive formula to determine the final individual fitness. Experiments were conducted on two public datasets: the ABIDE I dataset (1,112 subjects, 17 sites), and ADHD-200 (776 subjects, 8 sites), using AAL/CC200 atlases. Results demonstrate that PE-RBNAS achieves state-of-the-art performance, with 72.61% accuracy on clean ABIDE I data (vs. 71.05% for MC-APSONAS) and 71.82% accuracy under 0.2 noise (vs. 68.15% for PSO-BNAS). The results indicate that, compared to other methods, the proposed method demonstrates better model performance and superior noise resistance.
2025
156.
K, Venkata Ratna Prabha; Bindu, Chinni Hima; Devi, K Rama
An interpretable deep learning approach for autism spectrum disorder detection in children using NASNet-mobile Journal Article
In: Biomedical Physics & Engineering Express, vol. 11, no. 4, pp. 045006, 2025.
@article{K_2025,
title = {An interpretable deep learning approach for autism spectrum disorder detection in children using NASNet-mobile},
author = {Venkata Ratna Prabha K and Chinni Hima Bindu and K Rama Devi},
url = {https://dx.doi.org/10.1088/2057-1976/addbe7},
doi = {10.1088/2057-1976/addbe7},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-01},
journal = {Biomedical Physics & Engineering Express},
volume = {11},
number = {4},
pages = {045006},
publisher = {IOP Publishing},
abstract = {Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental disorder featuring impaired social interactions and communication abilities engaging the individuals in a restrictive or repetitive behaviour. Though incurable early detection and intervention can reduce the severity of symptoms. Structural magnetic resonance imaging (sMRI) can improve diagnostic accuracy, facilitating early diagnosis to offer more tailored care. With the emergence of deep learning (DL), neuroimaging-based approaches for ASD diagnosis have been focused. However, many existing models lack interpretability of their decisions for diagnosis. The prime objective of this work is to perform ASD classification precisely and to interpret the classification process in a better way so as to discern the major features that are appropriate for the prediction of disorder. The proposed model employs neural architecture search network - mobile(NASNet-Mobile) model for ASD detection, which is integrated with an explainable artificial intelligence (XAI) technique called local interpretable model-agnostic explanations (LIME) for increased transparency of ASD classification. The model is trained on sMRI images of two age groups taken from autism brain imaging data exchange-I (ABIDE-I) dataset. The proposed model yielded accuracy of 0.9607, F1-score of 0.9614, specificity of 0.9774, sensitivity of 0.9451, negative predicted value (NPV) of 0.9429, positive predicted value (PPV) of 0.9783 and the diagnostic odds ratio of 745.59 for 2 to 11 years age group compared to 12 to 18 years group. These results are superior compared to other state of the art models Inception v3 and SqueezeNet.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental disorder featuring impaired social interactions and communication abilities engaging the individuals in a restrictive or repetitive behaviour. Though incurable early detection and intervention can reduce the severity of symptoms. Structural magnetic resonance imaging (sMRI) can improve diagnostic accuracy, facilitating early diagnosis to offer more tailored care. With the emergence of deep learning (DL), neuroimaging-based approaches for ASD diagnosis have been focused. However, many existing models lack interpretability of their decisions for diagnosis. The prime objective of this work is to perform ASD classification precisely and to interpret the classification process in a better way so as to discern the major features that are appropriate for the prediction of disorder. The proposed model employs neural architecture search network - mobile(NASNet-Mobile) model for ASD detection, which is integrated with an explainable artificial intelligence (XAI) technique called local interpretable model-agnostic explanations (LIME) for increased transparency of ASD classification. The model is trained on sMRI images of two age groups taken from autism brain imaging data exchange-I (ABIDE-I) dataset. The proposed model yielded accuracy of 0.9607, F1-score of 0.9614, specificity of 0.9774, sensitivity of 0.9451, negative predicted value (NPV) of 0.9429, positive predicted value (PPV) of 0.9783 and the diagnostic odds ratio of 745.59 for 2 to 11 years age group compared to 12 to 18 years group. These results are superior compared to other state of the art models Inception v3 and SqueezeNet.
155.
Wang, Chen; Long, Kaifang; Guo, Tiezheng; Yang, Qingwen; Liu, Yanyi; Li, Pan; Li, Zhi; Wang, Huihan; Ma, Lianbo; Wen, Yingyou
Enhancing Result Interpretability of Neural Architecture Search-Assisted Medical AI via Large Language Model Journal Article
In: IEEE Transactions on Emerging Topics in Computational Intelligence, pp. 1-11, 2025.
@article{11168895,
title = {Enhancing Result Interpretability of Neural Architecture Search-Assisted Medical AI via Large Language Model},
author = {Chen Wang and Kaifang Long and Tiezheng Guo and Qingwen Yang and Yanyi Liu and Pan Li and Zhi Li and Huihan Wang and Lianbo Ma and Yingyou Wen},
url = {https://ieeexplore.ieee.org/abstract/document/11168895},
doi = {10.1109/TETCI.2025.3607385},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {IEEE Transactions on Emerging Topics in Computational Intelligence},
pages = {1-11},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
154.
Zhang, Zijun Frank; Zhan, Huixin; Wu, Tinghui; Burns, Robert; Hundal, Jasreet; Costa, Helio A.
The Expanding Landscape of Neural Architectures and Their Impact in Biomedicine Journal Article
In: Annual Review of Biomedical Data Science, vol. 8, no. Volume 8, 2025, pp. 101-124, 2025, ISSN: 2574-3414.
@article{annurev:/content/journals/10.1146/annurev-biodatasci-103023-050856,
title = {The Expanding Landscape of Neural Architectures and Their Impact in Biomedicine},
author = {Zijun Frank Zhang and Huixin Zhan and Tinghui Wu and Robert Burns and Jasreet Hundal and Helio A. Costa},
url = {https://www.annualreviews.org/content/journals/10.1146/annurev-biodatasci-103023-050856},
doi = {https://doi.org/10.1146/annurev-biodatasci-103023-050856},
issn = {2574-3414},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Annual Review of Biomedical Data Science},
volume = {8},
number = {Volume 8, 2025},
pages = {101-124},
publisher = {Annual Reviews},
abstract = {Deep learning and artificial intelligence (AI) have seen explosive growth and success in biomedical applications in the last decade, largely due to the rapid development of deep neural networks and their underlying neural network (NN) architectures. Here, we explore biomedical deep learning and AI from the specific perspective of NN architectures. We discuss widely varying design principles of NN architectures, their use in particular biomedical applications, and the assumptions (often hidden) built into them. We explore neural architecture search techniques that automate the design of NN topology to optimize task performance. Advanced neural architectures are being developed for both molecular and healthcare applications, employing elements of graph networks, transformers, and interpretable NNs, and we discuss and summarize the design considerations and unique advantages of each architecture. Future advances will include the employment of multimodal language models and smaller highly focused mechanistic models that build on the success of today's large models."},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep learning and artificial intelligence (AI) have seen explosive growth and success in biomedical applications in the last decade, largely due to the rapid development of deep neural networks and their underlying neural network (NN) architectures. Here, we explore biomedical deep learning and AI from the specific perspective of NN architectures. We discuss widely varying design principles of NN architectures, their use in particular biomedical applications, and the assumptions (often hidden) built into them. We explore neural architecture search techniques that automate the design of NN topology to optimize task performance. Advanced neural architectures are being developed for both molecular and healthcare applications, employing elements of graph networks, transformers, and interpretable NNs, and we discuss and summarize the design considerations and unique advantages of each architecture. Future advances will include the employment of multimodal language models and smaller highly focused mechanistic models that build on the success of today's large models."
153.
Zhang, Hanyu; Tang, Lixin; Song, Xiangman
EM-NAS: A Cell-Based Evolutionary Multi-scale Neural Architecture Search for Medical Image Segmentation Proceedings Article
In: Proceedings of the Genetic and Evolutionary Computation Conference Companion, pp. 755–758, Association for Computing Machinery, NH Malaga Hotel, Malaga, Spain, 2025, ISBN: 9798400714641.
@inproceedings{10.1145/3712255.3726641,
title = {EM-NAS: A Cell-Based Evolutionary Multi-scale Neural Architecture Search for Medical Image Segmentation},
author = {Hanyu Zhang and Lixin Tang and Xiangman Song},
url = {https://doi.org/10.1145/3712255.3726641},
doi = {10.1145/3712255.3726641},
isbn = {9798400714641},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference Companion},
pages = {755–758},
publisher = {Association for Computing Machinery},
address = {NH Malaga Hotel, Malaga, Spain},
series = {GECCO '25 Companion},
abstract = {Accurate segmentation of medical images is a crucial part of pathology research and clinical practice. Currently, deep neural network (DNN)-based models are commonly used for medicad image segmentation. However, most DNN-based models are manually designed by experts which is a time-consuming and labor-intensive process as well as poorly generalized for different applications. Therefore, we propose EM-NAS, a novel evolutionary multi-scale NAS method with a cell-based design for medical image segmentation. Specifically, the structures of encoder and decoder cells are represented by directed acyclic graphs (DAGs). Moreover, some non-subsampling multi-scale decomposition operations are introduced into the candidate operation set to capture multi-scale, multi-directional and multi-frequency features of medical images. Subsequently, we specify the cell of the network through a well-designed discrete coding strategy, and a GA is used to search for the optimal network architecture. The experimental results demonstrate the effectiveness of the proposed model on two public datasets.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Accurate segmentation of medical images is a crucial part of pathology research and clinical practice. Currently, deep neural network (DNN)-based models are commonly used for medicad image segmentation. However, most DNN-based models are manually designed by experts which is a time-consuming and labor-intensive process as well as poorly generalized for different applications. Therefore, we propose EM-NAS, a novel evolutionary multi-scale NAS method with a cell-based design for medical image segmentation. Specifically, the structures of encoder and decoder cells are represented by directed acyclic graphs (DAGs). Moreover, some non-subsampling multi-scale decomposition operations are introduced into the candidate operation set to capture multi-scale, multi-directional and multi-frequency features of medical images. Subsequently, we specify the cell of the network through a well-designed discrete coding strategy, and a GA is used to search for the optimal network architecture. The experimental results demonstrate the effectiveness of the proposed model on two public datasets.
152.
Babuc, Diogen; Fortiş, Teodor-Florin
AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases Proceedings Article
In: Barolli, Leonard; Enokido, Tomoya; Woungang, Isaac (Ed.): Complex, Intelligent and Software Intensive Systems, pp. 24–35, Springer Nature Switzerland, Cham, 2025, ISBN: 978-3-031-96099-4.
@inproceedings{10.1007/978-3-031-96099-4_3b,
title = {AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases},
author = {Diogen Babuc and Teodor-Florin Fortiş},
editor = {Leonard Barolli and Tomoya Enokido and Isaac Woungang},
isbn = {978-3-031-96099-4},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Complex, Intelligent and Software Intensive Systems},
pages = {24–35},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.
151.
Chen, Xi; Lv, Jiahuan; Wang, Zeyu; Qin, Genggeng; Zhou, Zhiguo
In: Journal of Biomedical Informatics, vol. 169, pp. 104869, 2025, ISSN: 1532-0464.
@article{CHEN2025104869,
title = {Adaptive-AutoMO: A domain adaptive automated multiobjective neural network for reliable lesion malignancy prediction via digital breast tomosynthesis},
author = {Xi Chen and Jiahuan Lv and Zeyu Wang and Genggeng Qin and Zhiguo Zhou},
url = {https://www.sciencedirect.com/science/article/pii/S153204642500098X},
doi = {https://doi.org/10.1016/j.jbi.2025.104869},
issn = {1532-0464},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Journal of Biomedical Informatics},
volume = {169},
pages = {104869},
abstract = {Early diagnosis of breast cancer remains a significant global health challenge, and the potential use of deep learning in Digital Breast Tomosynthesis (DBT) based breast cancer diagnosis is a promising avenue. To address data scarcity and domain shift problems in building a lesion malignancy predictive model, we proposed a domain adaptive automated multiobjective neural network (Adaptive-AutoMO) for reliable lesion malignancy prediction via DBT. Adaptive-AutoMO addresses three key challenges simultaneously, they are: privacy preserving, credibility measurement, and balance, which consists of training, adaptation and testing stages. In the training stage, we developed a multiobjective immune neural architecture search algorithm (MINAS) to generate a Pareto-optimal model set with balanced sensitivity and specificity and introduced a Bayesian optimization algorithm to optimize the hyperparameters. In the adaptation stage, a semi-supervised domain adaptive feature network based on maximum mean discrepancy (MMD-SSDAF) was designed, which can make the balanced models adaptable to the target domain and preserve the data privacy in the source domain. In the testing stage, we proposed an evidence reasoning method based on entropy (ERE) that can fuse multiple adapted models and estimate uncertainty to improve the model credibility. The experiments on two DBT image datasets (source and target domain datasets) revealed that Adaptive-AutoMO outperformed ResNet-18, DenseNet-121, and other available domain adaptive models. Meanwhile, the removal of high uncertainty samples resulted in a performance improvement in the target domain. These experiments affirmed that Adaptive-AutoMO can not only enhance model’s performance, but also preserve privacy in the source domain data, boost model credibility, and achieve a balance between sensitivity and specificity.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Early diagnosis of breast cancer remains a significant global health challenge, and the potential use of deep learning in Digital Breast Tomosynthesis (DBT) based breast cancer diagnosis is a promising avenue. To address data scarcity and domain shift problems in building a lesion malignancy predictive model, we proposed a domain adaptive automated multiobjective neural network (Adaptive-AutoMO) for reliable lesion malignancy prediction via DBT. Adaptive-AutoMO addresses three key challenges simultaneously, they are: privacy preserving, credibility measurement, and balance, which consists of training, adaptation and testing stages. In the training stage, we developed a multiobjective immune neural architecture search algorithm (MINAS) to generate a Pareto-optimal model set with balanced sensitivity and specificity and introduced a Bayesian optimization algorithm to optimize the hyperparameters. In the adaptation stage, a semi-supervised domain adaptive feature network based on maximum mean discrepancy (MMD-SSDAF) was designed, which can make the balanced models adaptable to the target domain and preserve the data privacy in the source domain. In the testing stage, we proposed an evidence reasoning method based on entropy (ERE) that can fuse multiple adapted models and estimate uncertainty to improve the model credibility. The experiments on two DBT image datasets (source and target domain datasets) revealed that Adaptive-AutoMO outperformed ResNet-18, DenseNet-121, and other available domain adaptive models. Meanwhile, the removal of high uncertainty samples resulted in a performance improvement in the target domain. These experiments affirmed that Adaptive-AutoMO can not only enhance model’s performance, but also preserve privacy in the source domain data, boost model credibility, and achieve a balance between sensitivity and specificity.
150.
Babuc, Diogen; Fortiş, Teodor-Florin
AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases Proceedings Article
In: Barolli, Leonard; Enokido, Tomoya; Woungang, Isaac (Ed.): Complex, Intelligent and Software Intensive Systems, pp. 24–35, Springer Nature Switzerland, Cham, 2025, ISBN: 978-3-031-96099-4.
@inproceedings{10.1007/978-3-031-96099-4_3,
title = {AD-ZeroNAS: Zero-Shot Proxies for Efficient Neural Architecture Search via Activation Diversity Function on Histopathological Use Cases},
author = {Diogen Babuc and Teodor-Florin Fortiş},
editor = {Leonard Barolli and Tomoya Enokido and Isaac Woungang},
url = {https://link.springer.com/chapter/10.1007/978-3-031-96099-4_3},
isbn = {978-3-031-96099-4},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Complex, Intelligent and Software Intensive Systems},
pages = {24–35},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) is a technique for automating deep learning model design, but its high computational cost remains a significant challenge. This paper introduces the activation diversity function score (ADFS) as a zero-shot proxy for evaluating candidate convolutional neural architectures without requiring full training on histopathological datasets, namely colorectal polyps and cervical cells. ADFS measures activation diversity across layers, weights, and updates, favoring architectures that learn rich, non-redundant feature representations. We apply ADFS to a stochastic multi-layer perceptron generation process, varying the number of layers, neurons per layer, activation functions, and skip connections. Architectures are ranked based on ADFS and undergo constrained training to validate performance. Experiments on structured medical histopathological datasets show that ADFS-selected models achieve high accuracy with significantly reduced training costs, aligning with Green AI initiatives. This work demonstrates the feasibility of efficient zero-shot NAS for deep learning. This makes model selection scalable and accessible.
149.
Benmeziane, Hadjer; GACEM, Abderaouf; Maghraoui, Kaoutar El; Benmeziane, Sara
Efficient Graph Neural Architecture Search for Medical Imaging in Real-World Clinical Settings Proceedings Article
In: 4th Muslims in ML Workshop co-located with ICML 2025, 2025.
@inproceedings{<LineBreak>benmeziane2025efficient,
title = {Efficient Graph Neural Architecture Search for Medical Imaging in Real-World Clinical Settings},
author = {Hadjer Benmeziane and Abderaouf GACEM and Kaoutar El Maghraoui and Sara Benmeziane},
url = {https://openreview.net/forum?id=Em7jjcxHAR},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {4th Muslims in ML Workshop co-located with ICML 2025},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
148.
Shawi, Radwa El
AgingFedNAS: Aging Evolution Federated Deep Learning for Architecture and Hyperparameter Search Proceedings Article
In: Wu, Xintao; Spiliopoulou, Myra; Wang, Can; Kumar, Vipin; Cao, Longbing; Wu, Yanqiu; Yao, Yu; Wu, Zhangkai (Ed.): Advances in Knowledge Discovery and Data Mining, pp. 107–119, Springer Nature Singapore, Singapore, 2025, ISBN: 978-981-96-8173-0.
@inproceedings{10.1007/978-981-96-8173-0_9,
title = {AgingFedNAS: Aging Evolution Federated Deep Learning for Architecture and Hyperparameter Search},
author = {Radwa El Shawi},
editor = {Xintao Wu and Myra Spiliopoulou and Can Wang and Vipin Kumar and Longbing Cao and Yanqiu Wu and Yu Yao and Zhangkai Wu},
url = {https://link.springer.com/chapter/10.1007/978-981-96-8173-0_9},
isbn = {978-981-96-8173-0},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Advances in Knowledge Discovery and Data Mining},
pages = {107–119},
publisher = {Springer Nature Singapore},
address = {Singapore},
abstract = {Despite advancements in Automatic Machine Learning (AutoML), industries encounter challenges in implementation due to data privacy concerns and the costs of centralized data storage. Federated Learning (FL) provides a decentralized approach, allowing multiple clients to collaboratively train models without sharing their datasets. However, many existing FL techniques utilize pre-defined model architectures from centralized environments, which may not be optimal for the non-iid data distributions commonly found among FL clients. This paper introduces AgingFedNAS, a framework designed to automate model design in FL by employing an evolutionary approach to jointly optimize neural architectures and hyperparameters. Comprehensive experiments conducted on heterogeneous data splits from CIFAR-10, Shakespeare, FEMNIST, Tiny-ImageNet, and a medical breast density classification dataset demonstrate that AgingFedNAS outperforms state-of-the-art FL frameworks, including FedAvg, FEATHERS, FedEx, and FedNAS, particularly in non-iid conditions. Notably, AgingFedNAS achieves an accuracy of 89.8% on CIFAR-10, exceeding the best baseline, FEATHERS, by 3.78%. In the breast density classification task, it surpasses FedNAS by 1.3%, achieving up to 3.2% higher accuracy for specific clients under non-iid scenarios. Additionally, in highly heterogeneous data environments, AgingFedNAS shows a 2.3% accuracy improvement on CIFAR-10 compared to the top-performing baseline.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Despite advancements in Automatic Machine Learning (AutoML), industries encounter challenges in implementation due to data privacy concerns and the costs of centralized data storage. Federated Learning (FL) provides a decentralized approach, allowing multiple clients to collaboratively train models without sharing their datasets. However, many existing FL techniques utilize pre-defined model architectures from centralized environments, which may not be optimal for the non-iid data distributions commonly found among FL clients. This paper introduces AgingFedNAS, a framework designed to automate model design in FL by employing an evolutionary approach to jointly optimize neural architectures and hyperparameters. Comprehensive experiments conducted on heterogeneous data splits from CIFAR-10, Shakespeare, FEMNIST, Tiny-ImageNet, and a medical breast density classification dataset demonstrate that AgingFedNAS outperforms state-of-the-art FL frameworks, including FedAvg, FEATHERS, FedEx, and FedNAS, particularly in non-iid conditions. Notably, AgingFedNAS achieves an accuracy of 89.8% on CIFAR-10, exceeding the best baseline, FEATHERS, by 3.78%. In the breast density classification task, it surpasses FedNAS by 1.3%, achieving up to 3.2% higher accuracy for specific clients under non-iid scenarios. Additionally, in highly heterogeneous data environments, AgingFedNAS shows a 2.3% accuracy improvement on CIFAR-10 compared to the top-performing baseline.
147.
Hu, Mengxiang; Li, Junchi; Dong, Yongquan; Zhang, Zichen; Liu, Weifan; Zhang, Peilin; Ping, Yuchao; Jiang, Le; Yu, Zekuan
Mixed-GGNAS: Mixed Search-space NAS based on genetic algorithm combined with gradient descent for medical image segmentation Journal Article
In: Expert Systems with Applications, vol. 289, pp. 128338, 2025, ISSN: 0957-4174.
@article{HU2025128338,
title = {Mixed-GGNAS: Mixed Search-space NAS based on genetic algorithm combined with gradient descent for medical image segmentation},
author = {Mengxiang Hu and Junchi Li and Yongquan Dong and Zichen Zhang and Weifan Liu and Peilin Zhang and Yuchao Ping and Le Jiang and Zekuan Yu},
url = {https://www.sciencedirect.com/science/article/pii/S0957417425019578},
doi = {https://doi.org/10.1016/j.eswa.2025.128338},
issn = {0957-4174},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Expert Systems with Applications},
volume = {289},
pages = {128338},
abstract = {Medical images segmentation is a pivotal procedure, playing a fundamental role in computer-assisted diagnosis and treatment. Despite the significant advancements in methods leveraging deep learning for this purpose, many networks still face challenges related to efficiency, often requiring substantial time and manual efforts. Neural architecture search (NAS) has gained considerable attention in the automated design of neural networks. This study introduces a new NAS method, Mixed-GGNAS, a Mixed Search-space NAS method based on Genetic algorithm combined with Gradient descent. Our approach creatively combines manually designed network blocks with DARTS blocks to construct a mixed search space. We then employ a method that integrates genetic algorithms and gradient descent to concurrently search for both block types and internal operations within the block. Within a U-shaped network framework, we propose a Multi-feature fusion strategy based on Vision Transformer (ViT) and search for hyperparameters of it. Additionally, we employ a Multi-scale mixed loss function to enhance the model’s ability to learn features at various scales. Experimental results demonstrate that the proposed approach outperforms or is comparable to the state-of-the-art NAS methods and manually designed Networks. Ablation studies conducted on two datasets further validate the method’s efficacy in enhancing model performance. The code is available at https://github.com/Hmxki/Mixed-GGNAS.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Medical images segmentation is a pivotal procedure, playing a fundamental role in computer-assisted diagnosis and treatment. Despite the significant advancements in methods leveraging deep learning for this purpose, many networks still face challenges related to efficiency, often requiring substantial time and manual efforts. Neural architecture search (NAS) has gained considerable attention in the automated design of neural networks. This study introduces a new NAS method, Mixed-GGNAS, a Mixed Search-space NAS method based on Genetic algorithm combined with Gradient descent. Our approach creatively combines manually designed network blocks with DARTS blocks to construct a mixed search space. We then employ a method that integrates genetic algorithms and gradient descent to concurrently search for both block types and internal operations within the block. Within a U-shaped network framework, we propose a Multi-feature fusion strategy based on Vision Transformer (ViT) and search for hyperparameters of it. Additionally, we employ a Multi-scale mixed loss function to enhance the model’s ability to learn features at various scales. Experimental results demonstrate that the proposed approach outperforms or is comparable to the state-of-the-art NAS methods and manually designed Networks. Ablation studies conducted on two datasets further validate the method’s efficacy in enhancing model performance. The code is available at https://github.com/Hmxki/Mixed-GGNAS.
146.
Sun, Yu; Zhang, Xianglin; Dong, Liang; Liu, Ning
In: Applied Soft Computing, vol. 179, pp. 113279, 2025, ISSN: 1568-4946.
@article{SUN2025113279,
title = {Multi-objective evolutionary neural architecture search for medical image analysis using transformer and large language models in advancing public health},
author = {Yu Sun and Xianglin Zhang and Liang Dong and Ning Liu},
url = {https://www.sciencedirect.com/science/article/pii/S1568494625005903},
doi = {https://doi.org/10.1016/j.asoc.2025.113279},
issn = {1568-4946},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Applied Soft Computing},
volume = {179},
pages = {113279},
abstract = {The rapid growth of medical imaging data in modern healthcare networks demands sophisticated automated analysis methods that can maintain high accuracy while operating efficiently at scale. Current approaches using transformers and large language models (LLMs) face challenges balancing computational requirements with diagnostic precision across diverse healthcare settings. This paper presents TransMed-NAS (transformer medical neural architecture search), a multi-objective evolutionary neural architecture search framework that automatically discovers efficient hybrid architectures by integrating transformers and LLMs for medical image segmentation. Our approach leverages evolutionary computation to optimize segmentation accuracy and computational efficiency while incorporating medical domain knowledge through LLM guidance. The framework introduces several innovations: a hierarchical channel selection strategy that preserves clinically relevant features, a weight entanglement mechanism that accelerates architecture search through intelligent knowledge transfer, and a surrogate model acceleration technique that reduces computational overhead while maintaining reliability. Experimental results on the ISIC 2020 dataset demonstrate TransMed-NAS’s superior performance compared to state-of-the-art methods. Our small model variant achieves competitive accuracy (0.934 Dice score) with only 0.82M parameters, while our large variant establishes new benchmarks (0.947 Dice score) with significantly reduced computational requirements. Ablation studies confirm the effectiveness of each component, particularly highlighting how LLM integration enhances architecture search efficiency and clinical relevance. These results demonstrate TransMed-NAS’s potential to advance automated medical image analysis in resource-diverse healthcare settings, making sophisticated diagnostic capabilities more accessible to underserved communities.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The rapid growth of medical imaging data in modern healthcare networks demands sophisticated automated analysis methods that can maintain high accuracy while operating efficiently at scale. Current approaches using transformers and large language models (LLMs) face challenges balancing computational requirements with diagnostic precision across diverse healthcare settings. This paper presents TransMed-NAS (transformer medical neural architecture search), a multi-objective evolutionary neural architecture search framework that automatically discovers efficient hybrid architectures by integrating transformers and LLMs for medical image segmentation. Our approach leverages evolutionary computation to optimize segmentation accuracy and computational efficiency while incorporating medical domain knowledge through LLM guidance. The framework introduces several innovations: a hierarchical channel selection strategy that preserves clinically relevant features, a weight entanglement mechanism that accelerates architecture search through intelligent knowledge transfer, and a surrogate model acceleration technique that reduces computational overhead while maintaining reliability. Experimental results on the ISIC 2020 dataset demonstrate TransMed-NAS’s superior performance compared to state-of-the-art methods. Our small model variant achieves competitive accuracy (0.934 Dice score) with only 0.82M parameters, while our large variant establishes new benchmarks (0.947 Dice score) with significantly reduced computational requirements. Ablation studies confirm the effectiveness of each component, particularly highlighting how LLM integration enhances architecture search efficiency and clinical relevance. These results demonstrate TransMed-NAS’s potential to advance automated medical image analysis in resource-diverse healthcare settings, making sophisticated diagnostic capabilities more accessible to underserved communities.
145.
Zechen, Zheng; Xuelei, He; Fengjun, Zhao; Xiaowei, He
PSNAS-Net: Hybrid gradient-physical optimizationfor efficient neural architecture search in customized medical imaging analysis Journal Article
In: Expert Systems with Applications, vol. 288, pp. 128155, 2025, ISSN: 0957-4174.
@article{ZECHEN2025128155,
title = {PSNAS-Net: Hybrid gradient-physical optimizationfor efficient neural architecture search in customized medical imaging analysis},
author = {Zheng Zechen and He Xuelei and Zhao Fengjun and He Xiaowei},
url = {https://www.sciencedirect.com/science/article/pii/S0957417425017750},
doi = {https://doi.org/10.1016/j.eswa.2025.128155},
issn = {0957-4174},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Expert Systems with Applications},
volume = {288},
pages = {128155},
abstract = {Neural architecture search (NAS) facilitates the automated construction of neural networks tailored to specific tasks and requirements, resulting in models that are more closely aligned with the target task’s demands. However, in many studies, the extensive design space, high search costs, and time-consuming evaluation calculations render NAS impractical for numerous medical data tasks. Addressing these challenges, this study introduces an efficient algorithm for searching deep learning architectures. Initially, we propose 19 fundamental rules to streamline the design space, thereby reducing its scale. To improve the efficiency of the algorithm, we designed a NAS framework (PSNAS-Net) for convolutional neural networks and VisionTransformer, which consists of two search stages: Firstly, the improved powell algorithm is used to determine the model range, and the population-based simulated annealing algorithm is utilized to expedite the search for the final model. During the neural architecture search process, we consider accuracy, parameters, FLOPs, and model stability as comprehensive evaluation objectives, we designed a robust, flexible, and comprehensive metric for model evaluation. The experimental results demonstrate that PSNAS-Net achieves significantly shorter search times (0.05-1.47 GPU Days) compared to 19 existing NAS methods, while discovering compact models (as small as 0.11M) with superior performance across five medical image benchmarks. This study offers a viable approach for model search that accommodates individualized requirements.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neural architecture search (NAS) facilitates the automated construction of neural networks tailored to specific tasks and requirements, resulting in models that are more closely aligned with the target task’s demands. However, in many studies, the extensive design space, high search costs, and time-consuming evaluation calculations render NAS impractical for numerous medical data tasks. Addressing these challenges, this study introduces an efficient algorithm for searching deep learning architectures. Initially, we propose 19 fundamental rules to streamline the design space, thereby reducing its scale. To improve the efficiency of the algorithm, we designed a NAS framework (PSNAS-Net) for convolutional neural networks and VisionTransformer, which consists of two search stages: Firstly, the improved powell algorithm is used to determine the model range, and the population-based simulated annealing algorithm is utilized to expedite the search for the final model. During the neural architecture search process, we consider accuracy, parameters, FLOPs, and model stability as comprehensive evaluation objectives, we designed a robust, flexible, and comprehensive metric for model evaluation. The experimental results demonstrate that PSNAS-Net achieves significantly shorter search times (0.05-1.47 GPU Days) compared to 19 existing NAS methods, while discovering compact models (as small as 0.11M) with superior performance across five medical image benchmarks. This study offers a viable approach for model search that accommodates individualized requirements.
144.
Becktepe, Jannis; Hennig, Leona; Oeltze-Jafra, Steffen; Lindauer, Marius
Auto-nnU-Net: Towards Automated Medical Image Segmentation Technical Report
2025.
@techreport{becktepe2025autonnunetautomatedmedicalimage,
title = {Auto-nnU-Net: Towards Automated Medical Image Segmentation},
author = {Jannis Becktepe and Leona Hennig and Steffen Oeltze-Jafra and Marius Lindauer},
url = {https://arxiv.org/abs/2505.16561},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
143.
Ardila, Diego Páez; Carvalho, Thiago; Saavedra, Santiago Vasquez; Niño, Cesar Valencia; Figueiredo, Karla; Vellasco, Marley
Quantum-Inspired NAS With Attention-Based Search Spaces in Medical Applications Proceedings Article
In: 2025 IEEE Symposium on Computational Intelligence in Health and Medicine Companion (CIHM Companion), pp. 1-5, 2025.
@inproceedings{11002695,
title = {Quantum-Inspired NAS With Attention-Based Search Spaces in Medical Applications},
author = {Diego Páez Ardila and Thiago Carvalho and Santiago Vasquez Saavedra and Cesar Valencia Niño and Karla Figueiredo and Marley Vellasco},
doi = {10.1109/CIHMCompanion65205.2025.11002695},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {2025 IEEE Symposium on Computational Intelligence in Health and Medicine Companion (CIHM Companion)},
pages = {1-5},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
142.
Cao, Bin; Deng, Huanyu; Hao, Yiming; Luo, Xiao
Multi-view information fusion based on federated multi-objective neural architecture search for MRI semantic segmentation Journal Article
In: Information Fusion, vol. 123, pp. 103301, 2025, ISSN: 1566-2535.
@article{CAO2025103301,
title = {Multi-view information fusion based on federated multi-objective neural architecture search for MRI semantic segmentation},
author = {Bin Cao and Huanyu Deng and Yiming Hao and Xiao Luo},
url = {https://www.sciencedirect.com/science/article/pii/S1566253525003744},
doi = {https://doi.org/10.1016/j.inffus.2025.103301},
issn = {1566-2535},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Information Fusion},
volume = {123},
pages = {103301},
abstract = {With the rapid development of artificial intelligence, medical image semantic segmentation is being used more widely. However, centralized training can lead to privacy risks. At the same time, MRI provides multiple views that together describe the anatomical structure of a lesion, but a single view may not fully capture all features. Therefore, integrating multi-view information in a federated learning setting is a key challenge for improving model generalization. This study combines federated learning, neural architecture search (NAS) and data fusion techniques to address issues related to data privacy, cross-institutional data distribution differences and multi-view information fusion in medical imaging. To achieve this, we propose the FL-MONAS framework, which leverages the advantages of NAS and federated learning. It uses a Pareto-frontier-based multi-objective optimization strategy to effectively combine 2D MRI with 3D anatomical structures, improving model performance while ensuring data privacy. Experimental results show that FL-MONAS maintains strong segmentation performance even in non-IID scenarios, providing an efficient and privacy-friendly solution for medical image analysis.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
With the rapid development of artificial intelligence, medical image semantic segmentation is being used more widely. However, centralized training can lead to privacy risks. At the same time, MRI provides multiple views that together describe the anatomical structure of a lesion, but a single view may not fully capture all features. Therefore, integrating multi-view information in a federated learning setting is a key challenge for improving model generalization. This study combines federated learning, neural architecture search (NAS) and data fusion techniques to address issues related to data privacy, cross-institutional data distribution differences and multi-view information fusion in medical imaging. To achieve this, we propose the FL-MONAS framework, which leverages the advantages of NAS and federated learning. It uses a Pareto-frontier-based multi-objective optimization strategy to effectively combine 2D MRI with 3D anatomical structures, improving model performance while ensuring data privacy. Experimental results show that FL-MONAS maintains strong segmentation performance even in non-IID scenarios, providing an efficient and privacy-friendly solution for medical image analysis.
141.
Mecharbat, Lotfi Abdelkrim; Almakky, Ibrahim; Takac, Martin; Yaqub, Mohammad
MedNNS: Supernet-based Medical Task-Adaptive Neural Network Search Technical Report
2025.
@techreport{mecharbat2025mednnssupernetbasedmedicaltaskadaptiveb,
title = {MedNNS: Supernet-based Medical Task-Adaptive Neural Network Search},
author = {Lotfi Abdelkrim Mecharbat and Ibrahim Almakky and Martin Takac and Mohammad Yaqub},
url = {https://arxiv.org/abs/2504.15865},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
140.
Wani, M. Arif; Sultan, Bisma; Ali, Sarwat; Sofi, Mukhtar Ahmad
Introduction to Deep Learning Applications Book Chapter
In: Advances in Deep Learning, Volume 2, pp. 1–14, Springer Nature Singapore, Singapore, 2025, ISBN: 978-981-96-3498-9.
@inbook{Wani2025c,
title = {Introduction to Deep Learning Applications},
author = {M. Arif Wani and Bisma Sultan and Sarwat Ali and Mukhtar Ahmad Sofi},
url = {https://doi.org/10.1007/978-981-96-3498-9_1},
doi = {10.1007/978-981-96-3498-9_1},
isbn = {978-981-96-3498-9},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
booktitle = {Advances in Deep Learning, Volume 2},
pages = {1–14},
publisher = {Springer Nature Singapore},
address = {Singapore},
abstract = {Deep learning has emerged as a cornerstone of modern Artificial Intelligence (AI), offering unparalleled performance across a variety of complex tasks. Its ability to automatically learn features from vast amounts of data has transformed industries ranging from computer vision to natural language processing. Among the numerous applications of deep learning, three areas have witnessed significant progress: Neural Architecture Search (NAS), steganography, and medical applications. This chapter explores application of deep learning architectures in these three areas.},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Deep learning has emerged as a cornerstone of modern Artificial Intelligence (AI), offering unparalleled performance across a variety of complex tasks. Its ability to automatically learn features from vast amounts of data has transformed industries ranging from computer vision to natural language processing. Among the numerous applications of deep learning, three areas have witnessed significant progress: Neural Architecture Search (NAS), steganography, and medical applications. This chapter explores application of deep learning architectures in these three areas.
139.
Guarrasi, Valerio; Mogensen, Klara; Tassinari, Sara; Qvarlander, Sara; Soda, Paolo
Timing Is Everything: Finding the Optimal Fusion Points in Multimodal Medical Imaging Technical Report
2025.
@techreport{guarrasi2025timingeverythingfindingoptimal,
title = {Timing Is Everything: Finding the Optimal Fusion Points in Multimodal Medical Imaging},
author = {Valerio Guarrasi and Klara Mogensen and Sara Tassinari and Sara Qvarlander and Paolo Soda},
url = {https://arxiv.org/abs/2505.02467},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
138.
Mecharbat, Lotfi Abdelkrim; Almakky, Ibrahim; Takac, Martin; Yaqub, Mohammad
MedNNS: Supernet-based Medical Task-Adaptive Neural Network Search Technical Report
2025.
@techreport{mecharbat2025mednnssupernetbasedmedicaltaskadaptive,
title = {MedNNS: Supernet-based Medical Task-Adaptive Neural Network Search},
author = {Lotfi Abdelkrim Mecharbat and Ibrahim Almakky and Martin Takac and Mohammad Yaqub},
url = {https://arxiv.org/abs/2504.15865},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
137.
Yu, Jiandong; Li, Tongtong; Shi, Xuerong; Zhao, Ziyang; Chen, Miao; Zhang, Yu; Wang, Junyu; Yao, Zhijun; Fang, Lei; Hu, Bin
ETMO-NAS: An efficient two-step multimodal one-shot NAS for lung nodules classification Journal Article
In: Biomedical Signal Processing and Control, vol. 104, pp. 107479, 2025, ISSN: 1746-8094.
@article{YU2025107479,
title = {ETMO-NAS: An efficient two-step multimodal one-shot NAS for lung nodules classification},
author = {Jiandong Yu and Tongtong Li and Xuerong Shi and Ziyang Zhao and Miao Chen and Yu Zhang and Junyu Wang and Zhijun Yao and Lei Fang and Bin Hu},
url = {https://www.sciencedirect.com/science/article/pii/S1746809424015374},
doi = {https://doi.org/10.1016/j.bspc.2024.107479},
issn = {1746-8094},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Biomedical Signal Processing and Control},
volume = {104},
pages = {107479},
abstract = {Malignant lung nodules are the initial diagnostic manifestation of lung cancer. Accurate predictive classification of malignant from benign lung nodules can improve treatment efficacy and survival rate of lung cancer patients. Since current deep learning-based PET/CT pulmonary nodule-assisted diagnosis models typically rely on network architectures carefully designed by researchers, which require professional expertise and extensive prior knowledge. To combat these challenges, in this paper, we propose an efficient two-step multimodal one-shot NAS (ETMO-NAS) for searching high-performance network architectures for reliable and accurate classification of lung nodules for multimodal PET/CT data. Specifically, the step I focuses on fully training the performance of all candidate architectures in the search space using the sandwich rule and in-place distillation strategy. The step II aims to split the search space into multiple non-overlapping subsupernets by parallel operation edge decomposition strategy and then fine-tune the subsupernets further improve performance. Finally, the performance of ETMO-NAS was validated on a set of real clinical data. The experimental results show that the classification architecture searched by ETMO-NAS achieves the best performance with accuracy, precision, sensitivity, specificity, and F-1 score of 94.23%, 92.10%, 95.83%, 92.86% and 0.9388, respectively. In addition, compared with the classical CNN model and NAS model, ETMO-NAS performs better with the same inputs, but with only 1/33–1/5 of the parameters. This provides substantial evidence for the competitiveness of the model in classification tasks and presents a new approach for automated diagnosis of PET/CT pulmonary nodules. Code and models will be available at: https://github.com/yujiandong0002/ETMO-NAS.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Malignant lung nodules are the initial diagnostic manifestation of lung cancer. Accurate predictive classification of malignant from benign lung nodules can improve treatment efficacy and survival rate of lung cancer patients. Since current deep learning-based PET/CT pulmonary nodule-assisted diagnosis models typically rely on network architectures carefully designed by researchers, which require professional expertise and extensive prior knowledge. To combat these challenges, in this paper, we propose an efficient two-step multimodal one-shot NAS (ETMO-NAS) for searching high-performance network architectures for reliable and accurate classification of lung nodules for multimodal PET/CT data. Specifically, the step I focuses on fully training the performance of all candidate architectures in the search space using the sandwich rule and in-place distillation strategy. The step II aims to split the search space into multiple non-overlapping subsupernets by parallel operation edge decomposition strategy and then fine-tune the subsupernets further improve performance. Finally, the performance of ETMO-NAS was validated on a set of real clinical data. The experimental results show that the classification architecture searched by ETMO-NAS achieves the best performance with accuracy, precision, sensitivity, specificity, and F-1 score of 94.23%, 92.10%, 95.83%, 92.86% and 0.9388, respectively. In addition, compared with the classical CNN model and NAS model, ETMO-NAS performs better with the same inputs, but with only 1/33–1/5 of the parameters. This provides substantial evidence for the competitiveness of the model in classification tasks and presents a new approach for automated diagnosis of PET/CT pulmonary nodules. Code and models will be available at: https://github.com/yujiandong0002/ETMO-NAS.
136.
Wang, Weibo; Li, Hua
NAS FD Lung: A novel lung assist diagnostic system based on neural architecture search Journal Article
In: Biomedical Signal Processing and Control, vol. 100, pp. 107022, 2025, ISSN: 1746-8094.
@article{WANG2025107022,
title = {NAS FD Lung: A novel lung assist diagnostic system based on neural architecture search},
author = {Weibo Wang and Hua Li},
url = {https://www.sciencedirect.com/science/article/pii/S1746809424010802},
doi = {https://doi.org/10.1016/j.bspc.2024.107022},
issn = {1746-8094},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {Biomedical Signal Processing and Control},
volume = {100},
pages = {107022},
abstract = {In the detection and recognition of lung nodules, pulmonary nodules vary in size and shape and contain many similar tissues and organs around them, leading to the problems of both missed detection and false detection in existing detection algorithms. Designing proprietary detection and recognition networks manually requires substantial professional expertise. This process is time-consuming and labour-intensive and leads to issues like parameter redundancy and improper feature selection. Therefore, this paper proposes a new pulmonary CAD (computer-aided diagnosis) system for pulmonary nodules, NAS FD Lung (Using the NAS approach to search deep FPN and DPN networks), that can automatically learn and generate a deep learning network tailored to pulmonary nodule detection and recognition task requirements. NAS FD Lung aims to use automatic search to generate deep learning networks in the auxiliary diagnosis of pulmonary nodules to replace the manual design of deep learning networks. NAS FD Lung comprises two automatic search networks: BM NAS-FPN (Using NAS methods to search for deep FPN structures with Binary operation and Matrix multiplication fusion methods) network for nodule detection and NAS-A-DPN (Using the NAS approach to search deep DPN networks with attention mechanism) for nodule identification. The proposed technique is tested on the LUNA16 dataset, and the experimental results show that the model is superior to many existing state-of-the-art approaches. The detection accuracy of lung nodules is 98.23%. Regarding the lung nodules classification, the accuracy, specificity, sensitivity, and AUC values achieved 96.32%,97.14%,95.82%, and 98.33%, respectively.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In the detection and recognition of lung nodules, pulmonary nodules vary in size and shape and contain many similar tissues and organs around them, leading to the problems of both missed detection and false detection in existing detection algorithms. Designing proprietary detection and recognition networks manually requires substantial professional expertise. This process is time-consuming and labour-intensive and leads to issues like parameter redundancy and improper feature selection. Therefore, this paper proposes a new pulmonary CAD (computer-aided diagnosis) system for pulmonary nodules, NAS FD Lung (Using the NAS approach to search deep FPN and DPN networks), that can automatically learn and generate a deep learning network tailored to pulmonary nodule detection and recognition task requirements. NAS FD Lung aims to use automatic search to generate deep learning networks in the auxiliary diagnosis of pulmonary nodules to replace the manual design of deep learning networks. NAS FD Lung comprises two automatic search networks: BM NAS-FPN (Using NAS methods to search for deep FPN structures with Binary operation and Matrix multiplication fusion methods) network for nodule detection and NAS-A-DPN (Using the NAS approach to search deep DPN networks with attention mechanism) for nodule identification. The proposed technique is tested on the LUNA16 dataset, and the experimental results show that the model is superior to many existing state-of-the-art approaches. The detection accuracy of lung nodules is 98.23%. Regarding the lung nodules classification, the accuracy, specificity, sensitivity, and AUC values achieved 96.32%,97.14%,95.82%, and 98.33%, respectively.
2024
135.
AL-Sabri, Raeed; Gao, Jianliang; Chen, Jiamin; Oloulade, Babatounde Moctard; Wu, Zhenpeng; Abdullah, Monir; Hu, Xiaohua
M3GNAS: Multi-modal Multi-view Graph Neural Architecture Search for Medical Outcome Predictions Proceedings Article
In: 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 1783-1788, IEEE Computer Society, Los Alamitos, CA, USA, 2024.
@inproceedings{10821927,
title = { M3GNAS: Multi-modal Multi-view Graph Neural Architecture Search for Medical Outcome Predictions },
author = {Raeed AL-Sabri and Jianliang Gao and Jiamin Chen and Babatounde Moctard Oloulade and Zhenpeng Wu and Monir Abdullah and Xiaohua Hu},
url = {https://doi.ieeecomputersociety.org/10.1109/BIBM62325.2024.10821927},
doi = {10.1109/BIBM62325.2024.10821927},
year = {2024},
date = {2024-12-01},
urldate = {2024-12-01},
booktitle = {2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},
pages = {1783-1788},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA, USA},
abstract = {Multi-modal multi-view graph learning models have achieved significant success in medical outcome prediction, combining various modalities to enhance the performance of various medical tasks. However, current architectures for multi-modal multi-view graph learning (M3GL) models heavily depend on manual design, demanding significant effort and expert experience. Meanwhile, significant advancements have been achieved in the field of graph neural architecture search (GNAS), contributing to the automated design of learning architectures based on graphs. However, GNAS faces challenges in automating multimodal multi-view graph learning (M3GL) models, as existing frameworks cannot handle M3GL architecture topology, and current search spaces do not consider M3GL models. To address the above challenges, we propose, for the first time, a multi-modal multi-view graph neural architecture search (M3GNAS) framework that automates the construction of the optimal M3GL models, enabling the integration of multi-modal features from different views. We also design an effective multi-modal multi-view learning (M3L) search space to develop inner-view and outer-view graph representation learning in the context of graph learning, obtaining a latent graph representation tailored to the specific requirements of downstream tasks. To examine the effectiveness of M3GNAS, it is evaluated on medical outcome prediction tasks. The experimental findings demonstrate our proposed framework’s superior performance compared to state-of-the-art models.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Multi-modal multi-view graph learning models have achieved significant success in medical outcome prediction, combining various modalities to enhance the performance of various medical tasks. However, current architectures for multi-modal multi-view graph learning (M3GL) models heavily depend on manual design, demanding significant effort and expert experience. Meanwhile, significant advancements have been achieved in the field of graph neural architecture search (GNAS), contributing to the automated design of learning architectures based on graphs. However, GNAS faces challenges in automating multimodal multi-view graph learning (M3GL) models, as existing frameworks cannot handle M3GL architecture topology, and current search spaces do not consider M3GL models. To address the above challenges, we propose, for the first time, a multi-modal multi-view graph neural architecture search (M3GNAS) framework that automates the construction of the optimal M3GL models, enabling the integration of multi-modal features from different views. We also design an effective multi-modal multi-view learning (M3L) search space to develop inner-view and outer-view graph representation learning in the context of graph learning, obtaining a latent graph representation tailored to the specific requirements of downstream tasks. To examine the effectiveness of M3GNAS, it is evaluated on medical outcome prediction tasks. The experimental findings demonstrate our proposed framework’s superior performance compared to state-of-the-art models.
134.
(Ed.)
Medical Neural Architecture Search: Survey and Taxonomy Collection
2024.
@collection{Benmeziane-ijcai24a,
title = {Medical Neural Architecture Search: Survey and Taxonomy},
author = {Hadjer Benmeziane and Imane Hamzaoui and Kaoutar El Maghraoui and Kaoutar El Maghraoui},
url = {https://www.ijcai.org/proceedings/2024/0878.pdf},
year = {2024},
date = {2024-08-03},
urldate = {2024-08-03},
booktitle = {IJCAI 2024},
keywords = {},
pubstate = {published},
tppubtype = {collection}
}
133.
Zhang, Jinnian; Chen, Weijie; Joshi, Tanmayee; Uyanik, Meltem; Zhang, Xiaomin; Loh, Po-Ling; Jog, Varun; Bruce, Richard; Garrett, John; McMillan, Alan
RobMedNAS: searching robust neural network architectures for medical image synthesis Journal Article
In: Biomedical Physics & Engineering Express, vol. 10, no. 5, pp. 055029, 2024.
@article{Zhang_2024,
title = {RobMedNAS: searching robust neural network architectures for medical image synthesis},
author = {Jinnian Zhang and Weijie Chen and Tanmayee Joshi and Meltem Uyanik and Xiaomin Zhang and Po-Ling Loh and Varun Jog and Richard Bruce and John Garrett and Alan McMillan},
url = {https://dx.doi.org/10.1088/2057-1976/ad6e87},
doi = {10.1088/2057-1976/ad6e87},
year = {2024},
date = {2024-08-01},
urldate = {2024-08-01},
journal = {Biomedical Physics & Engineering Express},
volume = {10},
number = {5},
pages = {055029},
publisher = {IOP Publishing},
abstract = {Investigating U-Net model robustness in medical image synthesis against adversarial perturbations, this study introduces RobMedNAS, a neural architecture search strategy for identifying resilient U-Net configurations. Through retrospective analysis of synthesized CT from MRI data, employing Dice coefficient and mean absolute error metrics across critical anatomical areas, the study evaluates traditional U-Net models and RobMedNAS-optimized models under adversarial attacks. Findings demonstrate RobMedNAS’s efficacy in enhancing U-Net resilience without compromising on accuracy, proposing a novel pathway for robust medical image processing.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Investigating U-Net model robustness in medical image synthesis against adversarial perturbations, this study introduces RobMedNAS, a neural architecture search strategy for identifying resilient U-Net configurations. Through retrospective analysis of synthesized CT from MRI data, employing Dice coefficient and mean absolute error metrics across critical anatomical areas, the study evaluates traditional U-Net models and RobMedNAS-optimized models under adversarial attacks. Findings demonstrate RobMedNAS’s efficacy in enhancing U-Net resilience without compromising on accuracy, proposing a novel pathway for robust medical image processing.
132.
Asiimwe, Arnold; Das, William; Benmeziane, Hadjer; Maghraoui, Kaoutar El
EDGE2024, 2024.
@conference{Asiimwe-edge24a,
title = {EfficientMedSAM: Accelerating Medical Image Segmentation via Neural Architecture Search and Knowledge Distillation},
author = {Arnold Asiimwe and William Das and Hadjer Benmeziane and Kaoutar El Maghraoui},
url = {https://research.ibm.com/publications/efficientmedsam-accelerating-medical-image-segmentation-via-neural-architecture-search-and-knowledge-distillation},
year = {2024},
date = {2024-07-07},
urldate = {2024-07-07},
booktitle = {EDGE2024},
journal = {EDGE 2024},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
131.
Berezsky, O. M.; Liashchynskyi, P. B.
METHOD OF GENERATIVE-ADVERSARIAL NETWORKS SEARCHING ARCHITECTURES FOR BIOMEDICAL IMAGES SYNTHESIS Journal Article
In: Radio Electronics, Computer Science, Control, no. 1, pp. 104, 2024.
@article{Berezsky_Liashchynskyi_2024,
title = {METHOD OF GENERATIVE-ADVERSARIAL NETWORKS SEARCHING ARCHITECTURES FOR BIOMEDICAL IMAGES SYNTHESIS},
author = {O. M. Berezsky and P. B. Liashchynskyi},
url = {http://ric.zntu.edu.ua/article/view/300976},
doi = {10.15588/1607-3274-2024-1-10},
year = {2024},
date = {2024-04-01},
urldate = {2024-04-01},
journal = {Radio Electronics, Computer Science, Control},
number = {1},
pages = {104},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
130.
Rajesh, Chilukamari; Sadam, Ravichandra; Kumar, Sushil
Automated Deep Learning Models for Medical Image Segmentation and Denoising Proceedings Article
In: 2024 17th International Conference on Signal Processing and Communication System (ICSPCS), pp. 1-7, 2024.
@inproceedings{10815837,
title = {Automated Deep Learning Models for Medical Image Segmentation and Denoising},
author = {Chilukamari Rajesh and Ravichandra Sadam and Sushil Kumar},
url = {https://ieeexplore.ieee.org/abstract/document/10815837},
doi = {10.1109/ICSPCS63175.2024.10815837},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {2024 17th International Conference on Signal Processing and Communication System (ICSPCS)},
pages = {1-7},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
129.
Franco-Gaona, Erick; Avila-Garcia, Maria-Susana; Cruz-Aceves, Ivan; Orocio-Garcia, Hiram-Efrain; Escobedo-Gordillo, Andres; Brieva, Jorge
Neural Architecture Search Using Trajectory Metaheuristics to Classify Coronary Stenosis Proceedings Article
In: 2024 20th International Symposium on Medical Information Processing and Analysis (SIPAIM), pp. 1-4, 2024.
@inproceedings{10783513,
title = {Neural Architecture Search Using Trajectory Metaheuristics to Classify Coronary Stenosis},
author = {Erick Franco-Gaona and Maria-Susana Avila-Garcia and Ivan Cruz-Aceves and Hiram-Efrain Orocio-Garcia and Andres Escobedo-Gordillo and Jorge Brieva},
url = {https://ieeexplore.ieee.org/abstract/document/10783513},
doi = {10.1109/SIPAIM62974.2024.10783513},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {2024 20th International Symposium on Medical Information Processing and Analysis (SIPAIM)},
pages = {1-4},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
128.
Kuş, Zeki; Kiraz, Berna; Aydin, Musa; Kiraz, Alper
BioNAS: Neural Architecture Search for Multiple Data Modalities in Biomedical Image Classification Proceedings Article
In: Garcia, Fausto P.; Jamil, Akhtar; Hameed, Alaa Ali; Ortis, Alessandro; Ramirez, Isaac Segovia (Ed.): Recent Trends and Advances in Artificial Intelligence, pp. 539–550, Springer Nature Switzerland, Cham, 2024, ISBN: 978-3-031-70924-1.
@inproceedings{10.1007/978-3-031-70924-1_41,
title = {BioNAS: Neural Architecture Search for Multiple Data Modalities in Biomedical Image Classification},
author = {Zeki Kuş and Berna Kiraz and Musa Aydin and Alper Kiraz},
editor = {Fausto P. Garcia and Akhtar Jamil and Alaa Ali Hameed and Alessandro Ortis and Isaac Segovia Ramirez},
isbn = {978-3-031-70924-1},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Recent Trends and Advances in Artificial Intelligence},
pages = {539–550},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Neural Architecture Search (NAS) for biomedical image classification has the potential to design highly efficient and accurate networks automatically for tasks from different modalities. This paper presents BioNAS, a new NAS approach designed for multi-modal biomedical image classification. Unlike other methods, BioNAS dynamically adjusts the number of stacks, modules, and feature maps in the network to improve both performance and complexity. The proposed approach utilizes an opposition-based differential evolution optimization technique to identify the optimal network structure. We have compared our methods on two public multi-class classification datasets with different data modalities: DermaMNIST and OrganCMNIST. BioNAS outperforms hand-designed networks, automatic machine learning frameworks, and most NAS studies in terms of accuracy (ACC) and area under the curve (AUC) on the OrganCMNIST and DermaMNIST datasets. The proposed networks significantly outperform all other methods on the DermaMNIST dataset, achieving accuracy improvements of up to 4.4 points and AUC improvements of up to 2.6 points, and also surpass other studies by up to 5.4 points in accuracy and 0.6 points in AUC on OrganCMNIST. Moreover, the proposed networks have fewer parameters than hand-designed architectures like ResNet-18 and ResNet-50. The results indicate that BioNAS has the potential to be an effective alternative to hand-designed networks and automatic frameworks, offering a competitive solution in the classification of biomedical images.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Neural Architecture Search (NAS) for biomedical image classification has the potential to design highly efficient and accurate networks automatically for tasks from different modalities. This paper presents BioNAS, a new NAS approach designed for multi-modal biomedical image classification. Unlike other methods, BioNAS dynamically adjusts the number of stacks, modules, and feature maps in the network to improve both performance and complexity. The proposed approach utilizes an opposition-based differential evolution optimization technique to identify the optimal network structure. We have compared our methods on two public multi-class classification datasets with different data modalities: DermaMNIST and OrganCMNIST. BioNAS outperforms hand-designed networks, automatic machine learning frameworks, and most NAS studies in terms of accuracy (ACC) and area under the curve (AUC) on the OrganCMNIST and DermaMNIST datasets. The proposed networks significantly outperform all other methods on the DermaMNIST dataset, achieving accuracy improvements of up to 4.4 points and AUC improvements of up to 2.6 points, and also surpass other studies by up to 5.4 points in accuracy and 0.6 points in AUC on OrganCMNIST. Moreover, the proposed networks have fewer parameters than hand-designed architectures like ResNet-18 and ResNet-50. The results indicate that BioNAS has the potential to be an effective alternative to hand-designed networks and automatic frameworks, offering a competitive solution in the classification of biomedical images.
127.
Ji, Yuanfeng; 纪源丰,
Towards efficient deep learning for medical image analysis Bachelor Thesis
2024.
@bachelorthesis{HKUHUB_10722_350322,
title = {Towards efficient deep learning for medical image analysis},
author = {Yuanfeng Ji and 纪源丰},
url = {http://hdl.handle.net/10722/350322},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {HKU Theses Online (HKUTO)},
abstract = {The advancement of deep learning in medical image analysis has revolutionized the field of medical diagnostics, improving both the accuracy and efficiency of computer- aided diagnostic systems. Despite substantial progress, significant challenges remain, mainly due to the diversity of medical tasks and the scarcity of high-quality annotated data. This thesis addresses these challenges by proposing efficient deep learning meth- ods that improve the development and evaluation of medical imaging models, ensuring their reliability and effectiveness in clinical settings. First, the thesis presents the Ab- dominal Multi-Organ Segmentation (AMOS) dataset, a robust collection of annotated medical images from different demographics and imaging modalities. AMOS utilizes a semi-automated annotation process powered by a pre-trained model, which not only accelerates annotation, but also improves the accuracy and consistency of the data. This approach helps curate a comprehensive benchmark that reflects the real-world complexity and variability of clinical environments, facilitating rigorous testing and evaluation of medical deep learning applications. Then, to address the diversity of imaging tasks, this thesis presents UXNet, a novel application of Neural Architecture Search (NAS) technology designed to adapt neural network architectures specifically for different medical image analysis tasks. UXNet dynamically adapts to the specifics of the input data and output tasks, optimizing model structures to achieve high levels of accuracy and efficiency in different settings. This reduces the reliance on manual tuning and expert knowledge, streamlining the development process of deep learn- ing solutions for medical imaging. Moreover, recognizing the increasing complexity of deep learning models, the thesis introduces AutoBench, an automated tool for the assessment and governance of these models. Leveraging large language models, Au- toBench automates the creation of evaluation standards and conducts comprehensive performance evaluations, facilitating continuous monitoring and adaptation of model performance in medical applications. Finally, I discuss some future work towards in developing efficient and effective deep learning medical applications.},
keywords = {},
pubstate = {published},
tppubtype = {bachelorthesis}
}
The advancement of deep learning in medical image analysis has revolutionized the field of medical diagnostics, improving both the accuracy and efficiency of computer- aided diagnostic systems. Despite substantial progress, significant challenges remain, mainly due to the diversity of medical tasks and the scarcity of high-quality annotated data. This thesis addresses these challenges by proposing efficient deep learning meth- ods that improve the development and evaluation of medical imaging models, ensuring their reliability and effectiveness in clinical settings. First, the thesis presents the Ab- dominal Multi-Organ Segmentation (AMOS) dataset, a robust collection of annotated medical images from different demographics and imaging modalities. AMOS utilizes a semi-automated annotation process powered by a pre-trained model, which not only accelerates annotation, but also improves the accuracy and consistency of the data. This approach helps curate a comprehensive benchmark that reflects the real-world complexity and variability of clinical environments, facilitating rigorous testing and evaluation of medical deep learning applications. Then, to address the diversity of imaging tasks, this thesis presents UXNet, a novel application of Neural Architecture Search (NAS) technology designed to adapt neural network architectures specifically for different medical image analysis tasks. UXNet dynamically adapts to the specifics of the input data and output tasks, optimizing model structures to achieve high levels of accuracy and efficiency in different settings. This reduces the reliance on manual tuning and expert knowledge, streamlining the development process of deep learn- ing solutions for medical imaging. Moreover, recognizing the increasing complexity of deep learning models, the thesis introduces AutoBench, an automated tool for the assessment and governance of these models. Leveraging large language models, Au- toBench automates the creation of evaluation standards and conducts comprehensive performance evaluations, facilitating continuous monitoring and adaptation of model performance in medical applications. Finally, I discuss some future work towards in developing efficient and effective deep learning medical applications.
126.
Chen, Xi; Lv, Jiahuan; Wang, Zeyu; Qin, Genggeng; Zhou, Zhiguo
In: Computers in Biology and Medicine, vol. 183, pp. 109299, 2024, ISSN: 0010-4825.
@article{CHEN2024109299,
title = {Deep-AutoMO: Deep automated multiobjective neural network for trustworthy lesion malignancy diagnosis in the early stage via digital breast tomosynthesis},
author = {Xi Chen and Jiahuan Lv and Zeyu Wang and Genggeng Qin and Zhiguo Zhou},
url = {https://www.sciencedirect.com/science/article/pii/S0010482524013842},
doi = {https://doi.org/10.1016/j.compbiomed.2024.109299},
issn = {0010-4825},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {Computers in Biology and Medicine},
volume = {183},
pages = {109299},
abstract = {Breast cancer is the most prevalent cancer in women, and early diagnosis of malignant lesions is crucial for developing treatment plans. Digital breast tomosynthesis (DBT) has emerged as a valuable tool for early breast cancer detection, as it can identify more lesions and improve the early detection rate. Deep learning has shown great potential in medical image-based cancer diagnosis, including DBT. However, deploying these models in clinical practice may be challenging due to concerns about reliability and robustness. In this study, we developed a novel deep automated multiobjective neural network (Deep-AutoMO) to build a trustworthy model and achieve balance, safety and robustness in a unified way. During the training stage, we introduced a multiobjective immune neural architecture search (MINAS) that simultaneously considers sensitivity and specificity as objective functions, aiming to strike a balance between the two. Each neural network in Deep-AutoMO comprises a combination of a ResNet block, a DenseNet block and a pooling layer. We employ Bayesian optimization to optimize the hyperparameters in the MINAS, enhancing the efficiency of the model training process. In the testing stage, evidential reasoning based on entropy (ERE) approach is proposed to build a safe and robust model. The experimental study on DBT images demonstrated that Deep-AutoMO achieves promising performance with a well-balanced trade-off between sensitivity and specificity, outperforming currently available methods. Moreover, the model's safety is ensured through uncertainty estimation, and its robustness is improved, making it a trustworthy tool for breast cancer diagnosis in clinical settings. We have shared the code on GitHub for other researchers to use. The code can be found at https://github.com/ChaoyangZhang-XJTU/Deep-AutoMO.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Breast cancer is the most prevalent cancer in women, and early diagnosis of malignant lesions is crucial for developing treatment plans. Digital breast tomosynthesis (DBT) has emerged as a valuable tool for early breast cancer detection, as it can identify more lesions and improve the early detection rate. Deep learning has shown great potential in medical image-based cancer diagnosis, including DBT. However, deploying these models in clinical practice may be challenging due to concerns about reliability and robustness. In this study, we developed a novel deep automated multiobjective neural network (Deep-AutoMO) to build a trustworthy model and achieve balance, safety and robustness in a unified way. During the training stage, we introduced a multiobjective immune neural architecture search (MINAS) that simultaneously considers sensitivity and specificity as objective functions, aiming to strike a balance between the two. Each neural network in Deep-AutoMO comprises a combination of a ResNet block, a DenseNet block and a pooling layer. We employ Bayesian optimization to optimize the hyperparameters in the MINAS, enhancing the efficiency of the model training process. In the testing stage, evidential reasoning based on entropy (ERE) approach is proposed to build a safe and robust model. The experimental study on DBT images demonstrated that Deep-AutoMO achieves promising performance with a well-balanced trade-off between sensitivity and specificity, outperforming currently available methods. Moreover, the model's safety is ensured through uncertainty estimation, and its robustness is improved, making it a trustworthy tool for breast cancer diagnosis in clinical settings. We have shared the code on GitHub for other researchers to use. The code can be found at https://github.com/ChaoyangZhang-XJTU/Deep-AutoMO.
125.
Park, Eunbin; Lee, Youngjoo
mDARTS: Searching ML-Based ECG Classifiers against Membership Inference Attacks Journal Article
In: IEEE Journal of Biomedical and Health Informatics, pp. 1-11, 2024.
@article{10720208,
title = {mDARTS: Searching ML-Based ECG Classifiers against Membership Inference Attacks},
author = {Eunbin Park and Youngjoo Lee},
url = {https://ieeexplore.ieee.org/abstract/document/10720208},
doi = {10.1109/JBHI.2024.3481505},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {IEEE Journal of Biomedical and Health Informatics},
pages = {1-11},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
124.
Zhang, Qing; Shao, Dan; Lin, Lin; Gong, Guoliang; Xu, Rui; Kido, Shoji; Cui, HongWei
Feature Separation in Diffuse Lung Disease Image Classification by Using Evolutionary Algorithm-Based NAS Journal Article
In: IEEE Journal of Biomedical and Health Informatics, pp. 1-12, 2024.
@article{10716760,
title = {Feature Separation in Diffuse Lung Disease Image Classification by Using Evolutionary Algorithm-Based NAS},
author = {Qing Zhang and Dan Shao and Lin Lin and Guoliang Gong and Rui Xu and Shoji Kido and HongWei Cui},
url = {https://ieeexplore.ieee.org/document/10716760},
doi = {10.1109/JBHI.2024.3481012},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {IEEE Journal of Biomedical and Health Informatics},
pages = {1-12},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
123.
Wu, Yu; Fan, Hailong; Ying, Weiqin; Zhou, Zekun; Zheng, Qiaoqiao; Zhang, Jiajian
Federated Neural Architecture Search with Hierarchical Progressive Acceleration for Medical Image Segmentation Proceedings Article
In: Tan, Ying; Shi, Yuhui (Ed.): Advances in Swarm Intelligence, pp. 112–123, Springer Nature Singapore, Singapore, 2024, ISBN: 978-981-97-7184-4.
@inproceedings{10.1007/978-981-97-7184-4_10,
title = {Federated Neural Architecture Search with Hierarchical Progressive Acceleration for Medical Image Segmentation},
author = {Yu Wu and Hailong Fan and Weiqin Ying and Zekun Zhou and Qiaoqiao Zheng and Jiajian Zhang},
editor = {Ying Tan and Yuhui Shi},
url = {https://link.springer.com/chapter/10.1007/978-981-97-7184-4_10},
isbn = {978-981-97-7184-4},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Advances in Swarm Intelligence},
pages = {112–123},
publisher = {Springer Nature Singapore},
address = {Singapore},
abstract = {Deep neural networks for medical image segmentation often require data from multiple medical institutions, but privacy concerns limit data sharing, making federated learning (FL) a viable alternative. However, predefined network architectures in FL are often suboptimal and need extensive manual tuning. Traditional neural architecture search (NAS) methods are unsuitable for FL due to high communication and evaluation costs. This paper presents an evolutionary NAS method (FS-ENAS) for federated medical image segmentation. FS-ENAS utilizes a U-Net++ based supernet with depthwise separable convolution and adaptable skip connections. It introduces a novel multi-stage, hierarchical progressive acceleration strategy tailored for federated neural architecture search to reduce communication and evaluation burdens. Experimental results on retinal blood vessel segmentation tasks show that FS-ENAS efficiently searches for suitable architectures with reduced communication and evaluation costs while protecting privacy.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Deep neural networks for medical image segmentation often require data from multiple medical institutions, but privacy concerns limit data sharing, making federated learning (FL) a viable alternative. However, predefined network architectures in FL are often suboptimal and need extensive manual tuning. Traditional neural architecture search (NAS) methods are unsuitable for FL due to high communication and evaluation costs. This paper presents an evolutionary NAS method (FS-ENAS) for federated medical image segmentation. FS-ENAS utilizes a U-Net++ based supernet with depthwise separable convolution and adaptable skip connections. It introduces a novel multi-stage, hierarchical progressive acceleration strategy tailored for federated neural architecture search to reduce communication and evaluation burdens. Experimental results on retinal blood vessel segmentation tasks show that FS-ENAS efficiently searches for suitable architectures with reduced communication and evaluation costs while protecting privacy.
122.
Domanski, Peter; Ray, Aritra; Lafata, Kyle; Firouzi, Farshad; Chakrabarty, Krishnendu; Pflüger, Dirk
Advancing blood glucose prediction with neural architecture search and deep reinforcement learning for type 1 diabetics Journal Article
In: Biocybernetics and Biomedical Engineering, vol. 44, no. 3, pp. 481-500, 2024, ISSN: 0208-5216.
@article{DOMANSKI2024481,
title = {Advancing blood glucose prediction with neural architecture search and deep reinforcement learning for type 1 diabetics},
author = {Peter Domanski and Aritra Ray and Kyle Lafata and Farshad Firouzi and Krishnendu Chakrabarty and Dirk Pflüger},
url = {https://www.sciencedirect.com/science/article/pii/S0208521624000536},
doi = {https://doi.org/10.1016/j.bbe.2024.07.006},
issn = {0208-5216},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {Biocybernetics and Biomedical Engineering},
volume = {44},
number = {3},
pages = {481-500},
abstract = {For individuals with Type-1 diabetes mellitus, accurate prediction of future blood glucose values is crucial to aid its regulation with insulin administration, tailored to the individual’s specific needs. The authors propose a novel approach for the integration of a neural architecture search framework with deep reinforcement learning to autonomously generate and train architectures, optimized for each subject over model size and analytical prediction performance, for the blood glucose prediction task in individuals with Type-1 diabetes. The authors evaluate the proposed approach on the OhioT1DM dataset, which includes blood glucose monitoring records at 5-min intervals over 8 weeks for 12 patients with Type-1 diabetes mellitus. Prior work focused on predicting blood glucose levels in 30 and 45-min prediction horizons, equivalent to 6 and 9 data points, respectively. Compared to the previously achieved best error, the proposed method demonstrates improvements of 18.4 % and 22.5 % on average for mean absolute error in the 30-min and 45-min prediction horizons, respectively, through the proposed deep reinforcement learning framework. Using the deep reinforcement learning framework, the best-case and worst-case analytical performance measured over root mean square error and mean absolute error was obtained for subject ID 570 and subject ID 584, respectively. Models optimized for performance on the prediction task and model size were obtained after implementing neural architecture search in conjunction with deep reinforcement learning on these two extreme cases. The authors demonstrate improvements of 4.8 % using Long Short Term Memory-based architectures and 5.7 % with Gated Recurrent Units-based architectures for patient ID 570 on the analytical prediction performance by integrating neural architecture search with deep reinforcement learning framework. The patient with the lowest performance (ID 584) on the deep reinforcement learning method had an even greater performance boost, with improvements of 10.0 % and 12.6 % observed for the Long Short-Term Memory and Gated Recurrent Units, respectively. The subject-specific optimized models over performance and model size from the neural architecture search in conjunction with deep reinforcement learning had a reduction in model size which ranged from 20 to 150 times compared to the model obtained using only the deep reinforcement learning method. The smaller size, indicating a reduction in model complexity in terms of the number of trainable network parameters, was achieved without a loss in the prediction performance.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
For individuals with Type-1 diabetes mellitus, accurate prediction of future blood glucose values is crucial to aid its regulation with insulin administration, tailored to the individual’s specific needs. The authors propose a novel approach for the integration of a neural architecture search framework with deep reinforcement learning to autonomously generate and train architectures, optimized for each subject over model size and analytical prediction performance, for the blood glucose prediction task in individuals with Type-1 diabetes. The authors evaluate the proposed approach on the OhioT1DM dataset, which includes blood glucose monitoring records at 5-min intervals over 8 weeks for 12 patients with Type-1 diabetes mellitus. Prior work focused on predicting blood glucose levels in 30 and 45-min prediction horizons, equivalent to 6 and 9 data points, respectively. Compared to the previously achieved best error, the proposed method demonstrates improvements of 18.4 % and 22.5 % on average for mean absolute error in the 30-min and 45-min prediction horizons, respectively, through the proposed deep reinforcement learning framework. Using the deep reinforcement learning framework, the best-case and worst-case analytical performance measured over root mean square error and mean absolute error was obtained for subject ID 570 and subject ID 584, respectively. Models optimized for performance on the prediction task and model size were obtained after implementing neural architecture search in conjunction with deep reinforcement learning on these two extreme cases. The authors demonstrate improvements of 4.8 % using Long Short Term Memory-based architectures and 5.7 % with Gated Recurrent Units-based architectures for patient ID 570 on the analytical prediction performance by integrating neural architecture search with deep reinforcement learning framework. The patient with the lowest performance (ID 584) on the deep reinforcement learning method had an even greater performance boost, with improvements of 10.0 % and 12.6 % observed for the Long Short-Term Memory and Gated Recurrent Units, respectively. The subject-specific optimized models over performance and model size from the neural architecture search in conjunction with deep reinforcement learning had a reduction in model size which ranged from 20 to 150 times compared to the model obtained using only the deep reinforcement learning method. The smaller size, indicating a reduction in model complexity in terms of the number of trainable network parameters, was achieved without a loss in the prediction performance.
121.
Chaiyarin, Sinee; Rojbundit, Napassorn; Piyabenjarad, Panichanok; Limpitigranon, Pimpattra; Wisitthipakdeekul, Siraprapa; Nonthasaen, Pawaree; Achararit, Paniti
Neural architecture search for medicine: A survey Journal Article
In: Informatics in Medicine Unlocked, vol. 50, pp. 101565, 2024, ISSN: 2352-9148.
@article{CHAIYARIN2024101565,
title = {Neural architecture search for medicine: A survey},
author = {Sinee Chaiyarin and Napassorn Rojbundit and Panichanok Piyabenjarad and Pimpattra Limpitigranon and Siraprapa Wisitthipakdeekul and Pawaree Nonthasaen and Paniti Achararit},
url = {https://www.sciencedirect.com/science/article/pii/S2352914824001217},
doi = {https://doi.org/10.1016/j.imu.2024.101565},
issn = {2352-9148},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {Informatics in Medicine Unlocked},
volume = {50},
pages = {101565},
abstract = {In this article we examined research on using neural architecture search (NAS) in medical applications, prompted by the current shortage of health care professionals relative to patient volumes. We explored the current state of NAS development in various medical fields, evaluated its performance, and examined potential future directions of NAS in medicine. We conducted a comprehensive search for articles published between 2019 and 2024, using the search string (Neural Architecture Search) OR (NAS) AND (medicine) OR (medical) OR (disease) OR (cardiovascular system) OR (MRI). We identified relevant studies published by Elsevier, IEEE, MDPI (IJERPH, Mathematics, Sensors), Nature, and SpringerLink, specifically focusing on experimental NAS applications in medical contexts. Data from 62 articles were collected, revealing a predominant use of NAS for image data classification, particularly in neurological research. Moreover, NAS demonstrated superior model performance compared with conventional deep learning methods. It is anticipated that future developments in NAS models for medical applications will lead to greater ease of use and enhanced efficacy as well as reduced computational resource consumption, thereby helping to mitigate health care workforce shortages and improve diagnostic accuracy. In addition to its application in diagnosis, NAS holds promise in everyday health monitoring, which could potentially enable the early detection of diseases, empowering people to receive the care that need and live healthier lives.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In this article we examined research on using neural architecture search (NAS) in medical applications, prompted by the current shortage of health care professionals relative to patient volumes. We explored the current state of NAS development in various medical fields, evaluated its performance, and examined potential future directions of NAS in medicine. We conducted a comprehensive search for articles published between 2019 and 2024, using the search string (Neural Architecture Search) OR (NAS) AND (medicine) OR (medical) OR (disease) OR (cardiovascular system) OR (MRI). We identified relevant studies published by Elsevier, IEEE, MDPI (IJERPH, Mathematics, Sensors), Nature, and SpringerLink, specifically focusing on experimental NAS applications in medical contexts. Data from 62 articles were collected, revealing a predominant use of NAS for image data classification, particularly in neurological research. Moreover, NAS demonstrated superior model performance compared with conventional deep learning methods. It is anticipated that future developments in NAS models for medical applications will lead to greater ease of use and enhanced efficacy as well as reduced computational resource consumption, thereby helping to mitigate health care workforce shortages and improve diagnostic accuracy. In addition to its application in diagnosis, NAS holds promise in everyday health monitoring, which could potentially enable the early detection of diseases, empowering people to receive the care that need and live healthier lives.
120.
Sheng, Yi; Yang, Junhuan; Li, Jinyang; Alaina, James; Xu, Xiaowei; Shi, Yiyu; Hu, Jingtong; Jiang, Weiwen; Yang, Lei
Data-Algorithm-Architecture Co-Optimization for Fair Neural Networks on Skin Lesion Dataset Proceedings Article
In: Linguraru, Marius George; Dou, Qi; Feragen, Aasa; Giannarou, Stamatia; Glocker, Ben; Lekadir, Karim; Schnabel, Julia A. (Ed.): Medical Image Computing and Computer Assisted Intervention - MICCAI 2024 - 27th International Conference, Marrakesh, Morocco, October 6-10, 2024, Proceedings, Part X, pp. 153–163, Springer, 2024.
@inproceedings{DBLP:conf/miccai/ShengYLAXSHJY24,
title = {Data-Algorithm-Architecture Co-Optimization for Fair Neural Networks on Skin Lesion Dataset},
author = {Yi Sheng and Junhuan Yang and Jinyang Li and James Alaina and Xiaowei Xu and Yiyu Shi and Jingtong Hu and Weiwen Jiang and Lei Yang},
editor = {Marius George Linguraru and Qi Dou and Aasa Feragen and Stamatia Giannarou and Ben Glocker and Karim Lekadir and Julia A. Schnabel},
url = {https://doi.org/10.1007/978-3-031-72117-5_15},
doi = {10.1007/978-3-031-72117-5_15},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Medical Image Computing and Computer Assisted Intervention - MICCAI
2024 - 27th International Conference, Marrakesh, Morocco, October
6-10, 2024, Proceedings, Part X},
volume = {15010},
pages = {153–163},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
119.
Wei, Jiahong; Xue, Bing; Zhang, Mengjie
EZUAS: Evolutionary Zero-shot U-shape Architecture Search for Medical Image Segmentation Proceedings Article
In: Proceedings of the Genetic and Evolutionary Computation Conference, pp. 422–430, Association for Computing Machinery, Melbourne, VIC, Australia, 2024, ISBN: 9798400704949.
@inproceedings{10.1145/3638529.3654041,
title = {EZUAS: Evolutionary Zero-shot U-shape Architecture Search for Medical Image Segmentation},
author = {Jiahong Wei and Bing Xue and Mengjie Zhang},
url = {https://doi.org/10.1145/3638529.3654041},
doi = {10.1145/3638529.3654041},
isbn = {9798400704949},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference},
pages = {422–430},
publisher = {Association for Computing Machinery},
address = {Melbourne, VIC, Australia},
series = {GECCO '24},
abstract = {Recently, deep learning-based methods have become the mainstream for medical image segmentation. Since manually designing deep neural networks (DNNs) is laborious and time-consuming, neural architecture search (NAS) becomes a popular stream for automatically designing DNNs for medical image segmentation. However, existing NAS work for medical image segmentation is still computationally expensive. Given the limited computation power, it is not always applicable to search for a well-performing model from an enlarged search space. In this paper, we propose EZUAS, a novel method of evolutionary zero-shot NAS for medical image segmentation, to address these issues. First, a new search space is designed for the automated design of DNNs. A genetic algorithm (GA) with an aligned crossover operation is then leveraged to search the network architectures under the model complexity constraints to get performant and lightweight models. In addition, a variable-length integer encoding scheme is devised to encode the candidate U-shaped DNNs with different stages. We conduct experiments on two commonly used medical image segmentation datasets to verify the effectiveness of the proposed EZUAS. Compared with the state-of-the-art methods, the proposed method can find a model much faster (about 0.04 GPU day) and achieve the best performance with lower computational complexity.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Recently, deep learning-based methods have become the mainstream for medical image segmentation. Since manually designing deep neural networks (DNNs) is laborious and time-consuming, neural architecture search (NAS) becomes a popular stream for automatically designing DNNs for medical image segmentation. However, existing NAS work for medical image segmentation is still computationally expensive. Given the limited computation power, it is not always applicable to search for a well-performing model from an enlarged search space. In this paper, we propose EZUAS, a novel method of evolutionary zero-shot NAS for medical image segmentation, to address these issues. First, a new search space is designed for the automated design of DNNs. A genetic algorithm (GA) with an aligned crossover operation is then leveraged to search the network architectures under the model complexity constraints to get performant and lightweight models. In addition, a variable-length integer encoding scheme is devised to encode the candidate U-shaped DNNs with different stages. We conduct experiments on two commonly used medical image segmentation datasets to verify the effectiveness of the proposed EZUAS. Compared with the state-of-the-art methods, the proposed method can find a model much faster (about 0.04 GPU day) and achieve the best performance with lower computational complexity.
118.
Ali, Muhammad Junaid; Moalic, Laurent; Essaid, Mokhtar; Idoumghar, Lhassane
Evolutionary Neural Architecture Search for 2D and 3D Medical Image Classification Proceedings Article
In: Franco, Leonardo; Mulatier, Clélia; Paszynski, Maciej; Krzhizhanovskaya, Valeria V.; Dongarra, Jack J.; Sloot, Peter M. A. (Ed.): Computational Science – ICCS 2024, pp. 131–146, Springer Nature Switzerland, Cham, 2024, ISBN: 978-3-031-63751-3.
@inproceedings{10.1007/978-3-031-63751-3_9,
title = {Evolutionary Neural Architecture Search for 2D and 3D Medical Image Classification},
author = {Muhammad Junaid Ali and Laurent Moalic and Mokhtar Essaid and Lhassane Idoumghar},
editor = {Leonardo Franco and Clélia Mulatier and Maciej Paszynski and Valeria V. Krzhizhanovskaya and Jack J. Dongarra and Peter M. A. Sloot},
url = {https://link.springer.com/chapter/10.1007/978-3-031-63751-3_9},
isbn = {978-3-031-63751-3},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Computational Science – ICCS 2024},
pages = {131–146},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Designing deep learning architectures is a challenging and time-consuming task. To address this problem, Neural Architecture Search (NAS) which automatically searches for a network topology is used. While existing NAS methods mainly focus on image classification tasks, particularly 2D medical images, this study presents an evolutionary NAS approach for 2D and 3D Medical image classification. We defined two different search spaces for 2D and 3D datasets and performed a comparative study of different meta-heuristics used in different NAS studies. Moreover, zero-cost proxies have been used to evaluate the performance of deep neural networks, which helps reduce the searching cost of the overall approach. Furthermore, recognizing the importance of Data Augmentation (DA) in model generalization, we propose a genetic algorithm based automatic DA strategy to find the optimal DA policy. Experiments on MedMNIST benchmark and BreakHIS dataset demonstrate the effectiveness of our approach, showcasing competitive results compared to existing AutoML approaches. The source code of our proposed approach is available at https://github.com/Junaid199f/evo_nas_med_2d_3d.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Designing deep learning architectures is a challenging and time-consuming task. To address this problem, Neural Architecture Search (NAS) which automatically searches for a network topology is used. While existing NAS methods mainly focus on image classification tasks, particularly 2D medical images, this study presents an evolutionary NAS approach for 2D and 3D Medical image classification. We defined two different search spaces for 2D and 3D datasets and performed a comparative study of different meta-heuristics used in different NAS studies. Moreover, zero-cost proxies have been used to evaluate the performance of deep neural networks, which helps reduce the searching cost of the overall approach. Furthermore, recognizing the importance of Data Augmentation (DA) in model generalization, we propose a genetic algorithm based automatic DA strategy to find the optimal DA policy. Experiments on MedMNIST benchmark and BreakHIS dataset demonstrate the effectiveness of our approach, showcasing competitive results compared to existing AutoML approaches. The source code of our proposed approach is available at https://github.com/Junaid199f/evo_nas_med_2d_3d.
117.
Cardoso, Fabio; Vellasco, Marley; Figueiredo, Karla
Comparative Study Between Q-NAS and Traditional CNNs for Brain Tumor Classification Proceedings Article
In: Iliadis, Lazaros; Maglogiannis, Ilias; Papaleonidas, Antonios; Pimenidis, Elias; Jayne, Chrisina (Ed.): Engineering Applications of Neural Networks, pp. 93–105, Springer Nature Switzerland, Cham, 2024, ISBN: 978-3-031-62495-7.
@inproceedings{10.1007/978-3-031-62495-7_8,
title = {Comparative Study Between Q-NAS and Traditional CNNs for Brain Tumor Classification},
author = {Fabio Cardoso and Marley Vellasco and Karla Figueiredo},
editor = {Lazaros Iliadis and Ilias Maglogiannis and Antonios Papaleonidas and Elias Pimenidis and Chrisina Jayne},
url = {https://link.springer.com/chapter/10.1007/978-3-031-62495-7_8},
isbn = {978-3-031-62495-7},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Engineering Applications of Neural Networks},
pages = {93–105},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Brain tumours caused approximately 251,329 deaths worldwide in 2020, with the primary diagnostic method for these tumours involving medical imaging. In recent years, many works and applications have observed the use of Artificial Intelligence-based models using Convolution Neural Networks (CNNs) to identify health problems using images. In our study, we searched for new architectures based on CNN using the Q-NAS algorithm. We compared its performance and number of parameters with traditional architectures such as VGG, ResNet, and MobileNet to classify types of brain tumors in MRI images. The best architecture found by Q-NAS achieved an accuracy of 92% on the test data set, with a model with less than one million parameters, which is much smaller than that found in the selected traditional architectures for this study. It shows the potential of the Q-NAS algorithm and highlights the importance of efficient model design in the context of accurate and feature-aware medical image analysis.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Brain tumours caused approximately 251,329 deaths worldwide in 2020, with the primary diagnostic method for these tumours involving medical imaging. In recent years, many works and applications have observed the use of Artificial Intelligence-based models using Convolution Neural Networks (CNNs) to identify health problems using images. In our study, we searched for new architectures based on CNN using the Q-NAS algorithm. We compared its performance and number of parameters with traditional architectures such as VGG, ResNet, and MobileNet to classify types of brain tumors in MRI images. The best architecture found by Q-NAS achieved an accuracy of 92% on the test data set, with a model with less than one million parameters, which is much smaller than that found in the selected traditional architectures for this study. It shows the potential of the Q-NAS algorithm and highlights the importance of efficient model design in the context of accurate and feature-aware medical image analysis.
116.
Kaur, Tanvir; Kamboj, Shivani; Singh, Lovedeep; Tamanna,
Advanced YOLO-NAS-Based Detection and Screening of Brain Tumors Using Medical Diagnostic Proceedings Article
In: 2024 2nd International Conference on Artificial Intelligence and Machine Learning Applications Theme: Healthcare and Internet of Things (AIMLA), pp. 1-6, 2024.
@inproceedings{10531625,
title = {Advanced YOLO-NAS-Based Detection and Screening of Brain Tumors Using Medical Diagnostic},
author = {Tanvir Kaur and Shivani Kamboj and Lovedeep Singh and Tamanna},
url = {https://ieeexplore.ieee.org/abstract/document/10531625},
doi = {10.1109/AIMLA59606.2024.10531625},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {2024 2nd International Conference on Artificial Intelligence and Machine Learning Applications Theme: Healthcare and Internet of Things (AIMLA)},
pages = {1-6},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
115.
Saeedizadeh, Narges; Jalali, Seyed Mohammad Jafar; Khan, Burhan; Kebria, Parham Mohsenzadeh; Mohamed, Shady
A new optimization approach based on neural architecture search to enhance deep U-Net for efficient road segmentation Journal Article
In: Knowledge-Based Systems, vol. 296, pp. 111966, 2024, ISSN: 0950-7051.
@article{SAEEDIZADEH2024111966,
title = {A new optimization approach based on neural architecture search to enhance deep U-Net for efficient road segmentation},
author = {Narges Saeedizadeh and Seyed Mohammad Jafar Jalali and Burhan Khan and Parham Mohsenzadeh Kebria and Shady Mohamed},
url = {https://www.sciencedirect.com/science/article/pii/S0950705124006002},
doi = {https://doi.org/10.1016/j.knosys.2024.111966},
issn = {0950-7051},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {Knowledge-Based Systems},
volume = {296},
pages = {111966},
abstract = {Neural Architecture Search (NAS) has significantly improved the accuracy of image classification and segmentation. However, these methods concentrate on finding segmentation structures for natural or medical applications. In this study, we introduce a NAS approach based on gradient optimization to identify ideal cell designs for road segmentation. To the best of our knowledge, this work represents the first application of gradient-based NAS to road extraction. Taking insight from the U-Net model and its successful variations in different image segmentation tasks, we propose NAS-enhanced U-Net, illustrated by an equal number of cells in both encoder and decoder levels. While cross-entropy combined with dice loss is commonly used in many segmentation tasks, road extraction brings up a unique challenge due to class imbalance. To address this, we introduce a combination of loss function. This function merges cross-entropy with weighted Dice loss, focusing on elevating the importance of the road class by assigning it a weight (⍵), while background Dice values are disregarded. The results indicate that the optimal weight for the proposed model equals 2. Additionally, our work challenges the assumption that increased model parameters or depth inherently leads to improved performance. Therefore, we establish search spaces 2,3,4,5,6,7 and 8 to automatically choose the optimal depth for model. We present promising segmentation results for our proposed method, achieved without any pretraining on the Massachusetts road dataset. Furthermore, these results are compared with those of 14 models categorized into four groups: U-Net, Segnet, FCN8, and Nas-U-Net.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neural Architecture Search (NAS) has significantly improved the accuracy of image classification and segmentation. However, these methods concentrate on finding segmentation structures for natural or medical applications. In this study, we introduce a NAS approach based on gradient optimization to identify ideal cell designs for road segmentation. To the best of our knowledge, this work represents the first application of gradient-based NAS to road extraction. Taking insight from the U-Net model and its successful variations in different image segmentation tasks, we propose NAS-enhanced U-Net, illustrated by an equal number of cells in both encoder and decoder levels. While cross-entropy combined with dice loss is commonly used in many segmentation tasks, road extraction brings up a unique challenge due to class imbalance. To address this, we introduce a combination of loss function. This function merges cross-entropy with weighted Dice loss, focusing on elevating the importance of the road class by assigning it a weight (⍵), while background Dice values are disregarded. The results indicate that the optimal weight for the proposed model equals 2. Additionally, our work challenges the assumption that increased model parameters or depth inherently leads to improved performance. Therefore, we establish search spaces 2,3,4,5,6,7 and 8 to automatically choose the optimal depth for model. We present promising segmentation results for our proposed method, achieved without any pretraining on the Massachusetts road dataset. Furthermore, these results are compared with those of 14 models categorized into four groups: U-Net, Segnet, FCN8, and Nas-U-Net.
114.
Xie, Lunchen; Lomurno, Eugenio; Gambella, Matteo; Ardagna, Danilo; Roveri, Manuel; Matteucci, Matteo; Shi, Qingjiang
A Lightweight Neural Architecture Search Model for Medical Image Classification Technical Report
2024.
@techreport{xie2024lightweight,
title = {A Lightweight Neural Architecture Search Model for Medical Image Classification},
author = {Lunchen Xie and Eugenio Lomurno and Matteo Gambella and Danilo Ardagna and Manuel Roveri and Matteo Matteucci and Qingjiang Shi},
url = {https://arxiv.org/abs/2405.03462},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
113.
Berezsky, Oleh; Liashchynskyi, Petro; Pitsun, Oleh; Izonin, Ivan
Synthesis of Convolutional Neural Network architectures for biomedical image classification Journal Article
In: Biomedical Signal Processing and Control, vol. 95, pp. 106325, 2024, ISSN: 1746-8094.
@article{BEREZSKY2024106325,
title = {Synthesis of Convolutional Neural Network architectures for biomedical image classification},
author = {Oleh Berezsky and Petro Liashchynskyi and Oleh Pitsun and Ivan Izonin},
url = {https://www.sciencedirect.com/science/article/pii/S1746809424003835},
doi = {https://doi.org/10.1016/j.bspc.2024.106325},
issn = {1746-8094},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {Biomedical Signal Processing and Control},
volume = {95},
pages = {106325},
abstract = {Convolutional Neural Networks (CNNs) are frequently used for image classification. This is crucial for the biomedical image classification used for automatic diagnosis in oncology. Designing optimal convolutional neural network architectures is a routine procedure that requires expert knowledge of computer vision and biomedical image features. To address this issue, we developed an automatic method for finding optimal CNN architectures. Our two-step method includes a genetic algorithm-based micro- and macro-search. Micro-search aims to find the optimal cell architecture based on the number of nodes and a set of predefined operations between nodes. Macro-search identifies the optimal number of cells and the operations between them to obtain the final optimal architecture. We obtained several optimal CNN architectures using the developed method of automatic architecture search. We conducted several computer experiments using cytological image classification as an example. The studies’ findings demonstrated that cytological image classification accuracy is higher compared to the classification accuracy of known CNN architectures (VGG-16, AlexNet, LeNet-5, ResNet-18, ResNet-50, MobileNetV3). The method is efficient because the search time for optimal architectures is short. Additionally, the method of optimal architecture search can be used for the synthesis of architectures used for the classification of other classes of biomedical images.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Convolutional Neural Networks (CNNs) are frequently used for image classification. This is crucial for the biomedical image classification used for automatic diagnosis in oncology. Designing optimal convolutional neural network architectures is a routine procedure that requires expert knowledge of computer vision and biomedical image features. To address this issue, we developed an automatic method for finding optimal CNN architectures. Our two-step method includes a genetic algorithm-based micro- and macro-search. Micro-search aims to find the optimal cell architecture based on the number of nodes and a set of predefined operations between nodes. Macro-search identifies the optimal number of cells and the operations between them to obtain the final optimal architecture. We obtained several optimal CNN architectures using the developed method of automatic architecture search. We conducted several computer experiments using cytological image classification as an example. The studies’ findings demonstrated that cytological image classification accuracy is higher compared to the classification accuracy of known CNN architectures (VGG-16, AlexNet, LeNet-5, ResNet-18, ResNet-50, MobileNetV3). The method is efficient because the search time for optimal architectures is short. Additionally, the method of optimal architecture search can be used for the synthesis of architectures used for the classification of other classes of biomedical images.
112.
Ali, Muhammad Junaid; Moalic, Laurent; Essaid, Mokhtar; Idoumghar, Lhassane
Robust Neural Architecture Search Using Differential Evolution for Medical Images Proceedings Article
In: Smith, Stephen; Correia, João; Cintrano, Christian (Ed.): Applications of Evolutionary Computation, pp. 163–179, Springer Nature Switzerland, Cham, 2024, ISBN: 978-3-031-56855-8.
@inproceedings{10.1007/978-3-031-56855-8_10,
title = {Robust Neural Architecture Search Using Differential Evolution for Medical Images},
author = {Muhammad Junaid Ali and Laurent Moalic and Mokhtar Essaid and Lhassane Idoumghar},
editor = {Stephen Smith and João Correia and Christian Cintrano},
url = {https://link.springer.com/chapter/10.1007/978-3-031-56855-8_10},
isbn = {978-3-031-56855-8},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
booktitle = {Applications of Evolutionary Computation},
pages = {163–179},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Recent studies have demonstrated that Convolutional Neural Network (CNN) architectures are sensitive to adversarial attacks with imperceptible permutations. Adversarial attacks on medical images may cause manipulated decisions and decrease the performance of the diagnosis system. The robustness of medical systems is crucial, as it assures an improved healthcare system and assists medical professionals in making decisions. Various studies have been proposed to secure medical systems against adversarial attacks, but they have used handcrafted architectures. This study proposes an evolutionary Neural Architecture Search (NAS) approach for searching robust architectures for medical image classification. The Differential Evolution (DE) algorithm is used as a search algorithm. Furthermore, we utilize an attention-based search space consisting of five different attention layers and sixteen convolution and pooling operations. Experiments on multiple MedMNIST datasets show that the proposed approach has achieved better results than deep learning architectures and a robust NAS approach.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Recent studies have demonstrated that Convolutional Neural Network (CNN) architectures are sensitive to adversarial attacks with imperceptible permutations. Adversarial attacks on medical images may cause manipulated decisions and decrease the performance of the diagnosis system. The robustness of medical systems is crucial, as it assures an improved healthcare system and assists medical professionals in making decisions. Various studies have been proposed to secure medical systems against adversarial attacks, but they have used handcrafted architectures. This study proposes an evolutionary Neural Architecture Search (NAS) approach for searching robust architectures for medical image classification. The Differential Evolution (DE) algorithm is used as a search algorithm. Furthermore, we utilize an attention-based search space consisting of five different attention layers and sixteen convolution and pooling operations. Experiments on multiple MedMNIST datasets show that the proposed approach has achieved better results than deep learning architectures and a robust NAS approach.
111.
Liu, Xin; Tian, Jie; Duan, Peiyong; Yu, Qian; Wang, Gaige; Wang, Yingjie
GrMoNAS: A granularity-based multi-objective NAS framework for efficient medical diagnosis Journal Article
In: Computers in Biology and Medicine, vol. 171, pp. 108118, 2024, ISSN: 0010-4825.
@article{LIU2024108118,
title = {GrMoNAS: A granularity-based multi-objective NAS framework for efficient medical diagnosis},
author = {Xin Liu and Jie Tian and Peiyong Duan and Qian Yu and Gaige Wang and Yingjie Wang},
url = {https://www.sciencedirect.com/science/article/pii/S0010482524002026},
doi = {https://doi.org/10.1016/j.compbiomed.2024.108118},
issn = {0010-4825},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {Computers in Biology and Medicine},
volume = {171},
pages = {108118},
abstract = {Neural Architecture Search (NAS) has been widely applied to automate medical image diagnostics. However, traditional NAS methods require significant computational resources and time for performance evaluation. To address this, we introduce the GrMoNAS framework, designed to balance diagnostic accuracy and efficiency using proxy datasets for granularity transformation and multi-objective optimization algorithms. The approach initiates with a coarse granularity phase, wherein diverse candidate neural architectures undergo evaluation utilizing a reduced proxy dataset. This initial phase facilitates the swift and effective identification of architectures exhibiting promise. Subsequently, in the fine granularity phase, a comprehensive validation and optimization process is undertaken for these identified architectures. Concurrently, employing multi-objective optimization and Pareto frontier sorting aims to enhance both accuracy and computational efficiency simultaneously. Importantly, the GrMoNAS framework is particularly suitable for hospitals with limited computational resources. We evaluated GrMoNAS in a range of medical scenarios, such as COVID-19, Skin cancer, Lung, Colon, and Acute Lymphoblastic Leukemia diseases, comparing it against traditional models like VGG16, VGG19, and recent NAS approaches including GA-CNN, EBNAS, NEXception, and CovNAS. The results show that GrMoNAS achieves comparable or superior diagnostic precision, significantly enhancing diagnostic efficiency. Moreover, GrMoNAS effectively avoids local optima, indicating its significant potential for precision medical diagnosis.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neural Architecture Search (NAS) has been widely applied to automate medical image diagnostics. However, traditional NAS methods require significant computational resources and time for performance evaluation. To address this, we introduce the GrMoNAS framework, designed to balance diagnostic accuracy and efficiency using proxy datasets for granularity transformation and multi-objective optimization algorithms. The approach initiates with a coarse granularity phase, wherein diverse candidate neural architectures undergo evaluation utilizing a reduced proxy dataset. This initial phase facilitates the swift and effective identification of architectures exhibiting promise. Subsequently, in the fine granularity phase, a comprehensive validation and optimization process is undertaken for these identified architectures. Concurrently, employing multi-objective optimization and Pareto frontier sorting aims to enhance both accuracy and computational efficiency simultaneously. Importantly, the GrMoNAS framework is particularly suitable for hospitals with limited computational resources. We evaluated GrMoNAS in a range of medical scenarios, such as COVID-19, Skin cancer, Lung, Colon, and Acute Lymphoblastic Leukemia diseases, comparing it against traditional models like VGG16, VGG19, and recent NAS approaches including GA-CNN, EBNAS, NEXception, and CovNAS. The results show that GrMoNAS achieves comparable or superior diagnostic precision, significantly enhancing diagnostic efficiency. Moreover, GrMoNAS effectively avoids local optima, indicating its significant potential for precision medical diagnosis.
110.
Cui, Suhan; Wang, Jiaqi; Zhong, Yuan; Liu, Han; Wang, Ting; Ma, Fenglong
Automated Fusion of Multimodal Electronic Health Records for Better Medical Predictions Technical Report
2024.
@techreport{cui2024automated,
title = {Automated Fusion of Multimodal Electronic Health Records for Better Medical Predictions},
author = {Suhan Cui and Jiaqi Wang and Yuan Zhong and Han Liu and Ting Wang and Fenglong Ma},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}