Summary
International Conference on Emerging Technologies for Communications
2020
Session Number:SA1
Session:
Number:SA1-6
Explainable Deep Convolutional Neural Network in EEG Seizure Onset Prediction
Punnawish THUWAJIT, Nannapas BANLUESOMBATKUL, Phattarapong SAWANGJAI, Payongkit LAKHAN, Theerawit WILAIPRASITPORN,
pp.-
Publication Date:2020/12/2
Online ISSN:2188-5079
DOI:10.34385/proc.63.SA1-6
PDF download
Summary:
Epilepsy is a common neurological disorder that affects millions of people worldwide. It can be classified by frequent seizure onsets. Deep Convolutional Neural Network (DCNN) has been used in electroencephalography (EEG) analysis in seizure onset prediction, achiev- ing over 90% accuracy. Despite the high accuracy, thought process behind the DCNN model’s decision wasn’t explained which limits the model’s po- tential in clinical usage. In this study, Layer-wise Relevance Propagation (LRP) was applied on a DCNN model, classifying normal, pre-seizure and seizure EEG signal, in order to plot the relevance of each data point in the form of heatmap. The result showed that the DCNN achieving 96.47% accuracy selects the features related to in the classification of these three classes, with LRP explanation revealing that the DCNN can select the fea- tures related to the distictive features of each signal type. In conclusion, these findings substantiate the reliability of the DCNN model in clinical usage.