Early stroke disease prediction with facial features using convolutional neural network model

Abstract

Past researcher has proposed computed tomography (CT) and magnetic resonance image (MRI) scan images as the most efficient ways to diagnose stroke disease. These methods are not only hectic and take much time but are also costly. This paper proposes a new approach to diagnosing this disease and gives a time and cost-efficient solution. We have offered a two-step solution to diagnose stroke disease in a patient using only the patient’s facial image. In the first step, we gathered a dataset of several stroke patients and normal persons. Then we applied several pre-processing operations, including red, green and blue (RGB) to grayscale conversion, scaling/ resizing, and normalization on dataset images before training them. In the second step, we trained the cropped images of their face regions and trained them using a convolutional neural network (CNN). We have successfully achieved an efficiency of 98%. The accuracy, precision, recall, and f-measure of the results were measured at 98%, 97%, 99%, and 98% respectively.