Optimization of Multi-Channel EEG Signal Using Genetic Algorithm in Post-Stroke Classification

Abstract

Stroke is a disease with the highest cause of disability in the world. Therefore, the post-stroke rehabilitation stage is crucial for patients to carry out daily activities as usual. Recording and processing Electroencephalogram (EEG) signals support to evaluate the development of post-stroke patients. EEG signal obtained from multi-channel is possible to become redundancy, which can affect processing time and computational time. The diminishing channel can reduce processing time, computational load, and the effects of overfitting due to excessive utilization of EEG channels. Some methods have been applied to cope with the problems. In this paper, the signal data used are those contained in the channel combination resulting from the channel optimization process. Wavelet transform is used for the extraction of EEG signals into Delta, Theta, Alpha, and Mu waves. The waves and amplitudes of each channel are extracted using a Genetic Algorithm (GA). GA reduced the channels from 14 channels to 12 channels. Then the channels optimized by GA are classified using Convolutional Neural Networks (CNN) into three classes, specifically “No Stroke”, “Minor Stroke”, and “Moderate Stroke”. The experiment showed that 12 channel combinations from GA output yield an accuracy of 93.33%, while classification using a complete channel produces an accuracy of 66.67%. The choice of optimization model also influences the accuracy where the study obtained SGD provides more accuracy in the long run (increased epoch). At the same time, Adam responds more quickly to improve accuracy at the beginning of training.