Real-Time Epileptic Seizure Detection Based on Deep Learning

Epilepsy is one of the most common neurological diseases, and video EEG is the most commonly used examination method for epilepsy diagnosis. However, since the video EEG examination lasts for hours, the escort has a heavy burden, and the large amount of video EEG data needs to be visually checked by the doctor. The real-time detection of epileptic seizures can reduce the stress of the escort and provide a mark for the doctor to check the EEG efficiently. In this paper, we propose a deep neural network with specified signal representation for real-time seizure detection and add a smoothing filter on the model output to enhance performance. First, we compare the performance of real-time epileptic seizure detection model under different signal representations. Then we use the best signal representation for further analysis in real-time scenario. In the experiment, the EEG data of 9 patients in the CHB-MIT public data set was used, and a patient-specific neural network was trained for each individual. The recall was 97%, the false alarm was 0.219 times per hour, and the latency time was 3.4s for real-time seizure event detection. The results show that this method can realize the real-time detection of epileptic seizures, which is of great significance to the subsequent system design combined with actual scenes.