The Effects of Diverse Brain MRI Modalities on Multiple Sclerosis Lesions Segmentation: Comprehensive Analysis

Abstract

Multiple Sclerosis (MS) is a long-term neurological condition that significantly impacts the spinal cord and brain. Diagnosis of MS is primarily conducted through various methods, with Magnetic Resonance Imaging (MRI) being a key technique. MRI modalities are reliable and precise diagnostic tools that offer clinicians detailed images, revealing vital details about the brain's structure and function. MRI images can be obtained using MRI scanners, with different settings. This yields to the generation of different types of images which are called sequences/modalities. Some of the most popular modalities are Proton Density weighted (PDw), T1-weighted (T1w), T2-weighted (T2w), and Fluid Attenuated Inversion Recovery (FLAIR). Diagnosing MS from brain MRI is crucial because it is a hard and time-intensive process and is particularly susceptible to manual or human error. In this work, we conduct a comprehensive analysis of how the choice of MRI modality can significantly affect the process of MS lesion segmentation using the proposed segmentation methodology based on UNETR model. The study showed and proved that FLAIR modality shows up to be the best and the most efficient method to use where it significantly improves all evaluation metrics scores than Tlw and T2w modalities. Using FLAIR sequence the model could achieve an average Dice Similarity Coefficient (DSC), sensitivity, and precision scores of 78.1%, 56.9%, and 62.1 %, respectively.