An Explainable Transfer Learning Method for EEG-based Seizure Type Classification.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy diagnosis relies on electroencephalograms (EEGs), but manual analysis is time-consuming.
• Automated seizure classification methods can significantly aid clinicians in epilepsy analysis.

Purpose of the Study:

• To develop an explainable transfer learning method for classifying seizure types in EEG recordings.
• To improve the efficiency and accuracy of epilepsy diagnosis and monitoring.

Main Methods:

• Utilized spectrograms derived from 19 EEG channels capturing seizure events.
• Employed four pre-trained transfer learning models: Inception, ResNet, DenseNet, and VGG16.
• Integrated the LIME technique for explainable AI, generating heatmap visualizations with colorbars to enhance clinician understanding.

Main Results:

• The transfer learning model demonstrated high accuracy in classifying seizure types on an independent test dataset.
• Explainable AI visualizations (heatmaps) effectively highlighted important EEG features for seizure classification.
• The colorbar aided clinicians in comprehending model predictions and identifying seizure events.

Conclusions:

• The developed explainable transfer learning method shows significant promise for accurate epilepsy seizure classification.
• This approach can serve as a valuable tool for neurologists, supporting comprehensive epilepsy case analysis.
• Enhanced clinician trust through explainable AI facilitates the adoption of automated EEG analysis tools.