Multiple Sclerosis Lesion Segmentation using MRI: U-Net Approach

This study presents an approach to multiple sclerosis (MS) lesion segmentation using a modified U-Net architecture. Leveraging the ISBI2015 dataset, a widely recognized benchmark dataset for MS lesion segmentation, our methodology effectively addresses the challenges of accurately delineating MS lesions from brain MRI. Through rigorous experimentation and evaluation, we demonstrate the efficacy and robustness of our approach, achieving the highest Dice similarity coefficient (DSC) among state-of-the-art methods. Specifically, our model attained a DSC of 77.56% on training set and 73.98% on test set, significantly surpassing previous benchmarks. Our training results show an accuracy of 99.3%, a DSC of 77.56%, precision of 73.60%, and recall of 93.56%. Similarly, our testing results indicate an accuracy of 99.29%, a DSC of 73.98%, precision of 70.84% and recall of 91.49%. These metrics consistently showcase the superior performance of the modified U-Net architecture in accurately detecting and delineating MS lesions across diverse patient data. The application of our modified U-Net architecture holds significant promise for improving the diagnosis and management of MS, ultimately advancing patient care and treatment outcomes.