Multimodal Wearable Device Signal Based Epilepsy Detection with Multi-scale Convolutional Neural Network

Seizure detection based on wearable devices has gradually become a popular research direction. The ability of wearable devices to capture signals is also improving, and a variety of physiological signals can be collected. However, current models for wearable devices focus on single-scale analysis and cannot adapt to current multi-modal signals. In this paper, an attention module-based convolutional neural network multi-scale model based on a novel wearable device is proposed to recognize epileptic seizures. The network extracts feature at different scales from multimodal physiological signals, supplemented by an attention module to retain valuable information. Experiments on multimodal physiological data from 13 typical epilepsy patients demonstrated that the proposed model achieves 93.5% sensitivity and 97.3% specificity.