Identification of Post-Stroke Patients from Significant Variable Electroencephalogram Signals Using Multiple 2D-Convolutional Neural Networks

Data from the World Stroke Organization provides information that 13.7 million people experience a stroke and 5.5 million die. To deal with this stroke, a rehabilitation process is needed, handled by a neurologist using a device called an Electroencephalogram (EEG). This EEG device detects electrical activity in the scalp area. Several studies have extracted EEG signals with Wavelets for Alpha, Gamma, Beta, Mu, and Theta waves. The EEG signal results do not directly tell the state of the post-stroke patient's signal because this EEG has high complexity, low amplitude, and does not have a specific pattern that is easy to analyze. In addition, there is the same inter- channel signal as the multi-channel signal. Research has overcome multi-channel problems with several methods such as 2D-Convolutional Neural Network (2D-CNN), Principal Component Analysis (PCA), and Multiple RNN. However, in the case of stroke identification, it is necessary to review the combination of features to find the combination of features that represent the condition of post-stroke patients, including variables such as Theta, Alpha, Mu, Alpha Asymmetric, Amplitude, Delta and Theta. So in this study using the Multiple 2D-CNN methods. Results of the study indicate that the use of the Multiple 2D- CNN methods with the use of 100 epochs and the Adam optimizer has an accuracy of 93.88% on the combination of Alpha and Theta features, increasing accuracy by 7.22% of the overall network display.