Epileptic seizure auto-detection using deep learning method

Traditional method of epileptic seizure detection could not avoid the process of manually selecting the features. Recently, the development of deep learning technology has provided a new direction. This paper introduces a new method of the seizure detection based on EEG signal using the short time Fourier transform(STFT) and convolution neural network(CNN). And the paper verifies the feasibility of this method through the actual research data and parameter setting. Afterwards, the method of single threshold is adopted to combine the multi-channel results. Then, the comparison with the classical method using the support vector machine(SVM) has been done, which shows that the approach presented in this paper is better. And the experimental result on single channel is that the average accuracy is 86%. In addition, the method of the multi-channel could increase the average accuracy to 90% and the average true positive rate(TPR) to 96.5% while decrease the average false positive rate(FPR) to 7%. All of those indexes reveal the high performance and stability of the approach for the epileptic seizure detection.