Spatial temporal GRU convnets for vision-based real time epileptic seizure detection
Zhou Fang, Howan Leung, Chiu‐Sing Choy · 2018
Vision-based epileptic seizure detection has been used to assist clinicians to diagnose and supervise patients for a long time. This branch of medical discipline is called the analysis of semiology. Current automated methods for the analysis of semiology suffer from several drawbacks: a) contact markers are necessary in some of these methods; b) these existing methods are likely to miss short or unobvious epileptic seizure; c) they cannot work in complex or noisy situation. To overcome these drawbacks, we propose a novel deep convolutional neural network model which can not only detect epileptic seizure early, but also localize the onset point of time accurately. Evaluated with test videos that contain a large number of different situation, the novel deep learning model has shown satisfying performance with 3.41s detection latency and 0.214 false positive rate. This evaluation result proves that the novel model is accurate enough to detect and localize epileptic seizure.