Modeling Clinical Decision Variability in Explainable Multimodal Seizure Detection

Asfandyar Azhar, Amulyal Mathur, Sahil Jain, James Emilian, Shaurjya Mandal, Nidhish Shah, Yongjie Jessica Zhang
Proceedings of the 4th Machine Learning for Health Symposium, PMLR 259:61-72, 2025.

Abstract

Electroencephalography (EEG) plays a critical role in the monitoring and diagnosis of neurological disorders, particularly in detecting seizures and other harmful brain activities. However, interpreting EEG signals is a complex task that often suffers from high variability and subjectivity among clinical experts. This study introduces the BiG-WaR architecture, a comprehensive multimodal framework designed to classify harmful brain activity using EEG signals. BiG-WaR combines several neural network models, including BiLSTM, GNN, WaveNet, and ResNet, to effectively leverage spatial and temporal dynamics inherent in EEG data. Our approach integrates curriculum learning and appropriate data preprocessing to address the challenges of EEG analysis, such as high variability and the need for robust feature extraction. Initial results demonstrate that BiG-WaR framework is a robust benchmark, enhancing reliability and interpretability—critical factors for clinical adoption—by integrating attention mechanisms and gradient-weighted class activation mappings to provide insights into model decisions.

Cite this Paper

BibTeX
@InProceedings{pmlr-v259-azhar25b, title = {Modeling Clinical Decision Variability in Explainable Multimodal Seizure Detection}, author = {Azhar, Asfandyar and Mathur, Amulyal and Jain, Sahil and Emilian, James and Mandal, Shaurjya and Shah, Nidhish and Zhang, Yongjie Jessica}, booktitle = {Proceedings of the 4th Machine Learning for Health Symposium}, pages = {61--72}, year = {2025}, editor = {Hegselmann, Stefan and Zhou, Helen and Healey, Elizabeth and Chang, Trenton and Ellington, Caleb and Mhasawade, Vishwali and Tonekaboni, Sana and Argaw, Peniel and Zhang, Haoran}, volume = {259}, series = {Proceedings of Machine Learning Research}, month = {15--16 Dec}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v259/main/assets/azhar25b/azhar25b.pdf}, url = {https://proceedings.mlr.press/v259/azhar25b.html}, abstract = {Electroencephalography (EEG) plays a critical role in the monitoring and diagnosis of neurological disorders, particularly in detecting seizures and other harmful brain activities. However, interpreting EEG signals is a complex task that often suffers from high variability and subjectivity among clinical experts. This study introduces the BiG-WaR architecture, a comprehensive multimodal framework designed to classify harmful brain activity using EEG signals. BiG-WaR combines several neural network models, including BiLSTM, GNN, WaveNet, and ResNet, to effectively leverage spatial and temporal dynamics inherent in EEG data. Our approach integrates curriculum learning and appropriate data preprocessing to address the challenges of EEG analysis, such as high variability and the need for robust feature extraction. Initial results demonstrate that BiG-WaR framework is a robust benchmark, enhancing reliability and interpretability—critical factors for clinical adoption—by integrating attention mechanisms and gradient-weighted class activation mappings to provide insights into model decisions.} }
Endnote
%0 Conference Paper %T Modeling Clinical Decision Variability in Explainable Multimodal Seizure Detection %A Asfandyar Azhar %A Amulyal Mathur %A Sahil Jain %A James Emilian %A Shaurjya Mandal %A Nidhish Shah %A Yongjie Jessica Zhang %B Proceedings of the 4th Machine Learning for Health Symposium %C Proceedings of Machine Learning Research %D 2025 %E Stefan Hegselmann %E Helen Zhou %E Elizabeth Healey %E Trenton Chang %E Caleb Ellington %E Vishwali Mhasawade %E Sana Tonekaboni %E Peniel Argaw %E Haoran Zhang %F pmlr-v259-azhar25b %I PMLR %P 61--72 %U https://proceedings.mlr.press/v259/azhar25b.html %V 259 %X Electroencephalography (EEG) plays a critical role in the monitoring and diagnosis of neurological disorders, particularly in detecting seizures and other harmful brain activities. However, interpreting EEG signals is a complex task that often suffers from high variability and subjectivity among clinical experts. This study introduces the BiG-WaR architecture, a comprehensive multimodal framework designed to classify harmful brain activity using EEG signals. BiG-WaR combines several neural network models, including BiLSTM, GNN, WaveNet, and ResNet, to effectively leverage spatial and temporal dynamics inherent in EEG data. Our approach integrates curriculum learning and appropriate data preprocessing to address the challenges of EEG analysis, such as high variability and the need for robust feature extraction. Initial results demonstrate that BiG-WaR framework is a robust benchmark, enhancing reliability and interpretability—critical factors for clinical adoption—by integrating attention mechanisms and gradient-weighted class activation mappings to provide insights into model decisions.
APA
Azhar, A., Mathur, A., Jain, S., Emilian, J., Mandal, S., Shah, N. & Zhang, Y.J.. (2025). Modeling Clinical Decision Variability in Explainable Multimodal Seizure Detection. Proceedings of the 4th Machine Learning for Health Symposium, in Proceedings of Machine Learning Research 259:61-72 Available from https://proceedings.mlr.press/v259/azhar25b.html.

Related Material