Epilepsy Analysis with Portable Device based Multi-modal Physiological Signals

Portable devices have been widely studied in epilepsy detection, but few epilepsy analyses using wristband based multi-modal physiological signals are reported. In this paper, multimodal physiological signals of acceleration (ACC), gyroscope (GYR), electromyography (EMG), and electrodermal skin (EDA) recorded by a wistband from 36 childhood epilepsy subjects from the Children's Hospital, Zhejiang University School of Medicine (CHZU) are involved for study. The significance of multi-modal physiological signal features in characterizing epilepsy seizure onset is carried out based on statistical analysis and feature correlation. Discriminative feature selection based on the Statistical Analysis is applied for feature reduction. Studies reveal that: 1) using single-modal physiological signal either by ACC, GYR ,EMG, or EDA can offer a satisfactory seizure detection performance, where the highest sensitivity by singlemodal physiological signal is 78.22%, 2) using multi-modal physiological signals can improve the detection sensitivity to 84.16%, 3) using discriminative feature selection on multi-modal physiological signals can further enhance the seizure detection accuracy to 88.16%.