AUTOMATED DETECTION AND CLASSIFICATION OF BRAIN STROKE LESIONS IN MRI USING MACHINE LEARNING TECHNIQUES

Abstract

Magnetic Resonance Imaging (MRI) is crucial for diagnosing brain disorders, with strokes being particularly significant. Recent studies emphasize the importance of prompt stroke treatment, encapsulated in the adage "time is brain," which highlights that intervention within the first six hours can greatly improve outcomes and save lives. However, traditional manual stroke diagnosis by neuroradiologists is often subjective and time-consuming. To address this, our study presents an automated method for detecting, segmenting, and classifying brain stroke lesions from MRI scans. Utilizing machine learning, particularly diffusion-weighted imaging (DWI) sequences, our approach involves four main stages: pre-processing, segmentation, feature extraction, and classification. We employ k-Means for segmentation to identify stroke regions, and statistical features derived from these segments are used for classification with linear discriminant analysis (LDA), support vector machine (SVM), weighted k-Nearest Neighbor (k-NN), and a bagged tree classifier. Performance metrics include the Jaccard index, Dice coefficient, false positive and negative rates for segmentation, and accuracy, sensitivity, and specificity for classification. Our findings show that k-Means is optimal for stroke lesion segmentation, achieving a Dice index of 0.85, while SVM demonstrates the highest classification accuracy of 98.5% with an average training time of 1.8 seconds, suggesting an efficient automated solution for timely and accurate stroke diagnosis.