EEG Classification For Stroke Detection Using Deep Learning Networks

Abstract

Stroke is currently a major public health concern. Hence more accurate and objective methods for diagnosis and prognosis are required to enable better clinical decision making. Electroencephalogram (EEG) is a non-invasive, low-cost method that can provide information regarding changes in the cerebral cortex throughout the recovery process following a stroke. EEG gives information on the progression of brain activity patterns. Many strategies have recently been developed to improve detection accuracy such as Support Vector Machine(SVM), Artificial Neural Network (RNN), Logistic Regression (LR), etc. VGG-16 and RESNET-50 are two non-invasive, low- cost transfer learning methods compared in this study. The results show that the proposed models can correctly classify EEG signals as stroke or not-stroke with 90% accuracy and 100% sensitivity for RESNET-50 while VGG-16 has a 90% accuracy, 100% specificity, and 100% precision. The work also compares other parameter i.e., F1-score between VGG-16 and RESNET-50 for this purpose. RESNET-50 is a major improvement over VGG-16 in terms of speed. Based on the results, this work appears to have been a success in terms of deep learning. Automation and great accuracy are achievable with this technique, which may be used in instances where Computed Tomography (CT) scans or Magnetic Resonance Imaging (MRI) examinations are not accessible.