Post-Stroke Identification of EEG Signal Using Wavelet Filters and 2D-Convolutional Neural Networks

A stroke is a disruption of blood supply to the brain that leads to disability. Rehabilitation can reduce brain damage and other complications and also recover disability due to stroke. One instrument that can be used in post-stroke patients monitoring is Electroencephalogram (EEG). EEG signals are recorded from several channel pairs. So apart from maintaining signal sequences, handling time connectivity between channels is also essential. Therefore, the signal sequence and time connectivity between channels need to be handled simultaneously on two dimensions. The vertical dimension was the multi-channel dimension, and the horizontal dimension was the signal sequences of each channel concerning time. This research proposed the 2D-Convolutional Neural Networks method to identify post-stroke based on EEG signals. First, a wavelet filter was used to obtain a frequency range of 1–13 Hz to represent post-stroke patients. The identification results are one of the three levels of stroke, i.e., No Stroke, Minor Stroke, and Moderate Stroke. The experimental results show that 2D-CNN produced an accuracy of 94.99%, an increase of 8.32% from 1D-CNN, which obtained an accuracy of 86.67%.