Stroke severity classification based on EEG signals using 1D convolutional neural network

Abstract

Acute Ischemic Stroke (AIS) is one kind of stroke that occurs the most. Stroke itself is the number one cause of death that can reduce blood flow and deprive the oxygen into the brain. Early diagnosis can help patients getting faster medical treatment thus avoid unwanted damage to the brain. Electroencephalogram (EEG) is an alternative tool for diagnosing AIS to the standard tools as in MRI or CT-scan. In this research, we try to classify stroke severity with 1 dimensional CNN (Convolutional Neural Network). The proposed method calculates the power spectral density (PSD) of EEG recordings from normal and stroke subjects, as the model's inputs and extracts feature automatically using CNN. The final feature-maps were trained in fully connected layer to classify 4 classes: normal, mild, moderate and severe stroke. The research is conducted to reach the possible optimum computing time with accuracy reached to 97.3% for 64 s segmentation, and 50 convolutional filters with 1x120 kernel size. This result is obtained by using the EEG signal from 4 channels: C3, C4, O1, and O2.