Stroke Lesion Segmentation and Classification for Diffusion-Weighted Images using Machine Learning

Magnetic Resonance Imaging (MRI) plays a crucial role in diagnosing brain disorders, with stroke being a significant category among them. Recent studies emphasize the importance of swift treatment for stroke, known as "time is brain," as early intervention within six hours of stroke onset can save lives and improve outcomes. However, the conventional manual diagnosis of brain stroke by neuroradiologists is subjective and time-consuming. To address this issue, this study presents an automatic technique for detecting, segmenting, and classifying brain stroke lesions from MRI images. The technique utilizes machine learning methods, focusing on diffusion weighted imaging (DWI) sequences. The machine learning technique involves four stages: pre-processing, segmentation, feature extraction, and classification. In this paper, k-Means, is proposed to identify the stroke region. Statistical features extracted from these segments are then used for classification with linear discriminant analysis (LDA), support vector machine (SVM), weighted k-Nearest Neighbor (k-NN), and bagged tree classifier. The segmentation performance is assessed using Jaccard indices, Dice Coefficient, false positive, and false negative rates. Classification is evaluated based on accuracy, sensitivity, and specificity. The results show that k-Means performs best for stroke lesion segmentation, with sub-acute ischemic stroke achieving the highest Dice index of 0.85. For classification, support vector machine (SVM) demonstrates the highest accuracy of 98.5%, with an average training time of 1.8 seconds. In conclusion, this proposed stroke classification technique has promising potential for diagnosing and classifying brain stroke lesions, providing an efficient and automated approach to aid medical professionals in timely and accurate diagnoses.