MRI-Based Brain Stroke Classification using HOG Features and SVM with SMOTE Balancing

¹ Department of Computer Science, Kuvempu University, Shivamogga, India
² Department of Computer Science and Engineering, BGS Institute of Technology, Mandya, India

^* Corresponding author: vkb2231@gmail.com

Abstract

Brain MRI scans are vital for identifying stroke types, which is essential for quick and effective treatment. This study introduces a simple yet powerful machine learning method to categorize MRI images as Normal, Haemorrhagic (bleeding), or Ischemic (clot-related) strokes. The approach uses a technique called Histogram of Oriented Gradients (HOG) to identify key features in the images. To address an imbalance where some stroke types appear less frequently, a method called SMOTE was used to balance the data. Finally, a Support Vector Machine (SVM), adjusted for class importance, was employed for classification. The model performed exceptionally well, achieving an overall accuracy of 97%. It also showed strong performance across all stroke types. This method works well even with limited data and doesn't require extensive computing power, making it ideal for clinical settings, especially those with fewer resources. This highlights the ongoing value of traditional machine learning for medical image analysis.