A U-Net Based Machine Learning Approach with Augmentation for Enhanced Precision in Multiple Sclerosis Lesion Segmentation from Multi-Modal MRI

Motivation: Early-stage diagnosis of Multiple Sclerosis (MS) is crucial for initiating prompt treatment. MRI can play a vital role in this process. Manual detection of MS lesions from MRI is time-consuming and subjective process, prone to human errors. Automated MS lesion segmentation techniques are needed; multi-modal MRI can be utilized for this purpose. Goal(s): The goal of our work was accurate determination of MS lesions from multi-modal MRI data. Approach: We developed a fully-automated approach to MS lesion segmentation using a U-Net based deep CNN. Results: Our model achieved a Dice Similarity Coefficient score of 0.79 for MS lesion segmentation from multi-modal MRI data. Impact: This study's robust MS lesion segmentation model could complement and improve diagnostic precision and monitoring for clinicians, leading to personalized treatment insights. It enables researchers to explore further multi-modal MRI benefits and model optimizations, ultimately enhancing patient care and outcomes.