Study on Real-time Prediction Method of Seizures based on YOLOV3 for EEG Spike Wave Detection

Abstract

As an auxiliary examination method, electroencephalography (EEG) is widely used in the prediction and analysis of epileptic seizures. In this paper, the research aims to solve the difficulties of automatic prediction of epileptic seizures, the complexity of feature extraction in real-time prediction, and poor generality of the algorithm. We have proposed a target detection model (YOLOV3) in convolutional neural network (CNN) to detect spike waves in EEG to predict epileptic seizures in real time. Firstly, the low-complexity spike waves characteristics in the short-term scalp Bonn EEG database are extracted and labeled. Secondly, the YOLOV3 model is trained. Next, the trained model is used to verify the long-term CHB-MIT scalp EEG database. Four different preictal windows at 30 min, 60 min, 90 min and 120 min are used for real-time prediction of epileptic seizures. Finally, the experimental results show that the sensitivity of different preictal windows are 93.91%, 95.75%, 97.25% and 98.29% respectively, the average prediction time is 43.82 min, the average detection speed is 0.073 s per EEG and the false prediction rate(FPR) is 0.109 times/h. Compared with the traditional methods, the new method of epileptic seizures prediction based on YOLOV3 proposed in this paper can predict epileptic seizures accurately, efficiently and in real time, which has clinical application value.