Analysis for Early Seizure Detection System Based on Deep Learning Algorithm

In recent years, deep learning technologies has developed rapidly and applies to different fields increasingly [1]. Based on the development of deep learning, Clinical Decision Support System(CDSS) is also developing rapidly. Clinical decision support systems (CDSS) can effectively reduce the misdiagnosis and missed diagnosis rate of doctors. In this situation, we can analyze the results for early seizure detection system based on deep learning algorithm to help doctors to diagnose.In this paper, the problems of seizure prediction based on electroencephalogram (EEG) dataset are studied in depth, especially the probabilistic problem of predicting pre-seizure fragments. We propose to transform time domain EEG data into frequency domain information by discrete Fourier transform. We also propose a method for extracting frequency domain and time series data features based on two-layer convolutional neural network (CNN). This method AUC index can reach 0.79. In addition, we also carry out the comparative analysis and systematic analysis of linear discriminant analysis (LDA), logistic regression (LR) and basic recurrent neural network (RNN) methods. The experimental results show that the prediction performance based on the CNN model is better than other models.