Enhanced and Interpretable Brain Stroke Detection with High Accuracy using ViT B-16 Model

Brain stroke is a type of medical emergency that requires prompt and appropriate diagnosis to prevent more brain damage and improve patient outcomes. Standard diagnostic approaches are generally CNN-based but lack interpretability, and still struggle to capture long-range dependencies in medical images, which has limited their application at the clinic. In this study, this challenge was addressed by using the B-16 ViT model for automated stroke detection in the brain. On the other hand, rare details of the subtle spatial relationships are captured very well in scans of the brain and later become useful in looking for faint abnormalities in the spread of strokes across regions. For more confident predictions, CAMs<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> are used, thereby visually explaining to the reader which regions of the image influence model decisions on classification. It was found that rigorous evaluation reported over 96% accuracy along with high precision, recall, and F1-score compared with the state-of-the-art CNNbased models. Besides this, transparency is another aspect for which this model is closer to clinical applications, using CAMs. The architecture of this very strong performing and, at the same time, interpretable ViT B-16 model is extremely promising to promote advancement in stroke diagnosis and better patient care in real-world settings.<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup>CAMs are visual explanations that highlight the regions of an image that the model focuses on when making a prediction, helping to improve interpretability.