Wenwen Chang, Bingyang Ji, Dandan Li, Lei Zhen, Yaxuan Wei, Xuan Liu, Guanghui Yan
{"title":"Epilepsy Prediction via Time-Frequency Features and Multi-Scale Hybrid Neural Networks.","authors":"Wenwen Chang, Bingyang Ji, Dandan Li, Lei Zhen, Yaxuan Wei, Xuan Liu, Guanghui Yan","doi":"10.1007/s10916-025-02224-w","DOIUrl":null,"url":null,"abstract":"<p><p>The prediction of epileptic seizures heavily depends on the precise embedding and classification of complex, multi-dimensional electroencephalogram (EEG) signals. Due to individual variability and the dynamic non-linear nature of EEG signals, extracting highly discriminative spatiotemporal features is a core challenge in this field. In this study, to address this issue, we proposed a novel architecture based on the Epilepsy Prediction using Multi-Scale Hybrid Neural Network (EPM-HNN), which integrates adaptive channel weighting, multi-scale spatial feature extraction, and bidirectional temporal dependency modeling. Specifically, we incorporated a sliding window mechanism with spatiotemporal resolution into the feature extraction process, enhancing the model's sensitivity to neural dynamics across frequency bands and improving its ability to capture micro-patterns. We used the Res2Net-50 multi-scale feature extractor to enhance the convolutional neural network's capacity to process complex local micro-features, such as polyspike-and-slow-wave complexes. Additionally, we introduced Squeeze-and-Excitation Networks (SENet) to adaptively capture potential effective features between different EEG channels. This dynamic weighting mechanism based on adaptive attention demonstrates strong robustness and high generalization across individual subject data. Furthermore, we proposed a non-single-subject, non-specific cross-subject training and testing method, demonstrating its ability to combat overfitting when addressing differences in data distribution. Experiments on the CHB-MIT scalp EEG dataset achieved an overall prediction accuracy of 97.7%, validating the effectiveness of the proposed EPM-HNN architecture.</p>","PeriodicalId":16338,"journal":{"name":"Journal of Medical Systems","volume":"49 1","pages":"90"},"PeriodicalIF":3.5000,"publicationDate":"2025-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Medical Systems","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1007/s10916-025-02224-w","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"HEALTH CARE SCIENCES & SERVICES","Score":null,"Total":0}
引用次数: 0
Abstract
The prediction of epileptic seizures heavily depends on the precise embedding and classification of complex, multi-dimensional electroencephalogram (EEG) signals. Due to individual variability and the dynamic non-linear nature of EEG signals, extracting highly discriminative spatiotemporal features is a core challenge in this field. In this study, to address this issue, we proposed a novel architecture based on the Epilepsy Prediction using Multi-Scale Hybrid Neural Network (EPM-HNN), which integrates adaptive channel weighting, multi-scale spatial feature extraction, and bidirectional temporal dependency modeling. Specifically, we incorporated a sliding window mechanism with spatiotemporal resolution into the feature extraction process, enhancing the model's sensitivity to neural dynamics across frequency bands and improving its ability to capture micro-patterns. We used the Res2Net-50 multi-scale feature extractor to enhance the convolutional neural network's capacity to process complex local micro-features, such as polyspike-and-slow-wave complexes. Additionally, we introduced Squeeze-and-Excitation Networks (SENet) to adaptively capture potential effective features between different EEG channels. This dynamic weighting mechanism based on adaptive attention demonstrates strong robustness and high generalization across individual subject data. Furthermore, we proposed a non-single-subject, non-specific cross-subject training and testing method, demonstrating its ability to combat overfitting when addressing differences in data distribution. Experiments on the CHB-MIT scalp EEG dataset achieved an overall prediction accuracy of 97.7%, validating the effectiveness of the proposed EPM-HNN architecture.
期刊介绍:
Journal of Medical Systems provides a forum for the presentation and discussion of the increasingly extensive applications of new systems techniques and methods in hospital clinic and physician''s office administration; pathology radiology and pharmaceutical delivery systems; medical records storage and retrieval; and ancillary patient-support systems. The journal publishes informative articles essays and studies across the entire scale of medical systems from large hospital programs to novel small-scale medical services. Education is an integral part of this amalgamation of sciences and selected articles are published in this area. Since existing medical systems are constantly being modified to fit particular circumstances and to solve specific problems the journal includes a special section devoted to status reports on current installations.