A New Method of EEG Classification for Motor Impairment Neural Disorders Using Deep Learning

Abstract

AbstractThe research recent, convolutional neural networks (CNN) was usually used to automatically identify patterns on the medical signals. Among them, the EEG signal is receiving the most attention. The studies are looking for ways to represent the EEG signal to input the CNN model with the best results. Our study will present a method for analyzing EEG signals as time-frequency representations and will use it to execute end-to-end deep learning. Our proposed method will build a network without expert manipulation which automatically learns features from extracting the EEG signals as time-frequency representations. Our research propose methods of representing the EEG signals as time-frequency and it was used the input of the CNN to identify disorders. We have experimented with two time-frequency representations (Scalogram, Spectrogram) and two CNN (AlexNet, LeNet). After the experiment, we were very good results and evaluated them. The results show that the performance of the CNN was affected by the representation of the EEG signal and using AlexNet with the Scalogram images as input data is the most suitable with the accuracy 76.92%.KeywordsConvolutional neural networksAlexNetScalogramElectroencephalography