Childhood Epilepsy Syndrome Classification through a Deep Learning Network with Clinical History Integration

Abstract

In this conference paper, we present TSA3-D, a novel 
two-stream 3-D attention module-based deep network aimed 
at classifying childhood epilepsy syndromes using multichannel 
EEG data. Unlike existing research primarily focusing on seizure 
detection, we emphasize syndrome classification, integrating clin- 
ical history such as age of onset, family history, and treatment 
responses as predictive features to enhance precision. We optimize 
EEG features through multichannel montage transforms to min- 
imize artifact interference. TSA3-D incorporates channel-wise 
and dual spatial attention modules to improve feature learning. 
With data from 115 subjects covering seven epilepsy syndromes 
and a control group, our results demonstrate an outstanding 
accuracy of 99.52, surpassing existing state-of-the-art methods. 
This amalgamation of advanced deep learning and clinical history 
offers a promising avenue for precise syndrome classification, 
thereby facilitating improved diagnosis and tailored treatment 
strategies.