[Paper Review] A Bibliography of Multiple Sclerosis Lesions Detection Methods using Brain MRIs

Atif Shah, Maged S. Al-Shaibani+2 more|arXiv (Cornell University)|Feb 19, 2023

Brain Tumor Detection and Classification4 citations

TL;DR

This paper presents a comprehensive bibliographic review of 241 studies on multiple sclerosis (MS) lesion detection in brain MRIs, categorizing methods into six groups: data-driven, statistical, supervised machine learning, unsupervised learning, fuzzy, and deep learning. It finds that deep learning methods now outperform others, with Dice scores exceeding 0.8 in top-performing models, and identifies key research gaps for future work.

ABSTRACT

Introduction: Multiple Sclerosis (MS) is a chronic disease that affects millions of people across the globe. MS can critically affect different organs of the central nervous system such as the eyes, the spinal cord, and the brain. Background: To help physicians in diagnosing MS lesions, computer-aided methods are widely used. In this regard, a considerable research has been carried out in the area of automatic detection and segmentation of MS lesions in magnetic resonance images (MRIs). Methodology: In this study, we review the different approaches that have been used in computer-aided detection and segmentation of MS lesions. Our review resulted in categorizing MS lesion segmentation approaches into six broad categories: data-driven, statistical, supervised machine learning, unsupervised machine learning, fuzzy, and deep learning-based techniques. We critically analyze the different techniques under these approaches and highlight their strengths and weaknesses. Results: From the study, we observe that a considerable amount of work, around 25% of related literature, is focused on statistical-based MS lesion segmentation techniques, followed by 21.15% for data-driven based methods, 19.23% for deep learning and 15.38% for supervised methods. Implication: The study points out the challenges/gaps to be addressed in future research. The study shows the work which has been done in last one decade in detection and segmentation of MS lesions. The results show that, in recent years, deep learning methods are outperforming all the others methods.

Motivation & Objective

• To systematically review and categorize existing computer-aided diagnosis (CAD) methods for MS lesion detection in brain MRIs.
• To identify the dominant methodological trends in MS lesion segmentation from 2005 to 2020.
• To evaluate the performance of different approaches—especially deep learning—using standardized metrics like Dice Similarity Coefficient (DSC).
• To highlight limitations and research gaps in current methodologies for improved future development of MS-CAD systems.
• To provide a structured framework for evaluating and comparing MS lesion detection techniques across diverse datasets and clinical settings.

Proposed method

• The study conducted a systematic literature review using databases including IEEE Xplore, PubMed, Google Scholar, and Scopus, covering publications from January 2005 to December 2020.
• Methods were classified into six categories: data-driven, statistical, supervised machine learning, unsupervised machine learning, fuzzy logic, and deep learning-based techniques.
• Performance evaluation focused on standard segmentation metrics: Dice Similarity Coefficient (DSC), sensitivity (Sen), specificity (Spec), positive predictive value (PPV), and false positive rate (FPR).
• The review included 241 studies, with detailed analysis of 24 representative papers reporting quantitative results on public (e.g., ISBI 2015) and private datasets.
• Deep learning models, particularly 3D-CNNs, U-Net architectures, and GAN-based frameworks, were analyzed for their ability to detect lesions at voxel and lesion levels.
• Statistical and data-driven methods were evaluated for their reliance on intensity modeling and spatial priors, while fuzzy and supervised methods were assessed for robustness and generalization.

Figure 1: MS papers statistics for period of 2001 - 2020

Experimental results

Research questions

• RQ1 What are the dominant methodological approaches used in automatic MS lesion detection from brain MRI between 2005 and 2020?
• RQ2 How do different categories of methods—especially deep learning—compare in terms of performance metrics like DSC, sensitivity, and PPV?
• RQ3 What are the most commonly used datasets in MS lesion segmentation research, and how do model performances vary across them?
• RQ4 What are the key limitations and research gaps in current MS lesion detection systems, particularly in clinical applicability and generalization?
• RQ5 To what extent have deep learning models outperformed traditional methods in detecting MS lesions in brain MRI?

Key findings

• Deep learning methods accounted for 19.23% of the reviewed literature and have emerged as the most effective approach, with top models achieving a DSC of 0.83 on private datasets.
• The highest-performing model (Salem et al., 2020) achieved a DSC of 0.83, sensitivity of 83.09%, and FPR of 9.36% on a 60-patient dataset using a fully convolutional neural network (FCNN).
• On the ISBI 2015 dataset, models such as Aslani et al. (2019) reported a DSC of 0.7067, while Gabr et al. (2019) achieved a DSC of 0.95 for white matter and 0.96 for gray matter.
• Statistical-based methods were the most prevalent (25% of literature), followed by data-driven (21.15%) and supervised learning (15.38%) approaches.
• Despite high performance on some datasets, models showed variability in generalization—e.g., Krüger et al. (2020) reported a DSC of only 0.39 on the Zurich dataset, indicating challenges in real-world deployment.
• The study identifies a lack of standardized benchmarking and consistent evaluation across datasets, highlighting a major gap for future research in developing unified validation frameworks.

Figure 2: Different types of MRI’s