Automated Ischemic Stroke Classification from MRI Scans: Using a Vision Transformer Approach

Abstract

Background: This study evaluates the performance of a vision transformer (ViT) model, ViT-b16, in classifying ischemic stroke cases from Moroccan MRI scans and compares it to the Visual Geometry Group 16 (VGG-16) model used in a prior study. Methods: A dataset of 342 MRI scans, categorized into ‘Normal’ and ’Stroke’ classes, underwent preprocessing using TensorFlow’s tf.data API. Results: The ViT-b16 model was trained and evaluated, yielding an impressive accuracy of 97.59%, surpassing the VGG-16 model’s 90% accuracy. Conclusions: This research highlights the ViT-b16 model’s superior classification capabilities for ischemic stroke diagnosis, contributing to the field of medical image analysis. By showcasing the efficacy of advanced deep learning architectures, particularly in the context of Moroccan MRI scans, this study underscores the potential for real-world clinical applications. Ultimately, our findings emphasize the importance of further exploration into AI-based diagnostic tools for improving healthcare outcomes.