Research on Early Warning Technology of Epilepsy Based on Deep Learning

Epilepsy refers to the sudden abnormal discharge of brain neurons. As the second major neurological disease, epilepsy has brought trouble to many people’s lives. It is very important for patients to predict the onset of epilepsy in advance, because it can not only reduce the troubles in life, but also avoid physical side effects caused by excessive medication. This paper uses short-time Fourier transform on a 30-second EEG window to extract time-domain and frequency-domain information, and puts the generated time-frequency graph into the constructed network for training. The neural network designed in this paper uses feature extraction module and classification module to extract and classify time-frequency images. In addition, this paper constructs an attention module for electrode dimension to enhance the attention between electrodes. The experimental results on the CHB-MIT data set show that the accuracy of the algorithm in this paper has reached 90%, the false alarm rate is as low as 0.096/h, and it has high visibility, which meets the requirements of high accuracy and high robustness in the medical field.