Multivariate pattern classification of gray matter pathology in multiple sclerosis

Area of Science:

• Neuroimaging
• Neurology
• Machine Learning

Background:

• Univariate analyses reveal gray matter (GM) alterations in multiple sclerosis (MS) patients.
• Multivariate methods, such as support vector machines (SVM), can identify complex neuroanatomical patterns of GM differences.

Purpose of the Study:

• To identify neuroanatomical GM patterns relevant for individual classification of MS patients using multivariate linear SVM analysis.
• To assess the efficacy of SVM in differentiating MS patient groups based on disease stage, lesion load, and disease progression.

Main Methods:

• Utilized multivariate linear SVM analysis with leave-one-out cross-validation.
• Applied SVM to GM segmentations from T1-weighted 3D MR imaging scans of MS patients.
• Compared early vs. late MS, low vs. high white matter lesion load, and benign MS (BMS) vs. non-benign MS (NBMS) groups.

Main Results:

• Identified GM patterns in cortical and deep GM structures (thalamus, caudate) crucial for MS patient classification.
• Achieved high classification accuracy: 85% for early vs. late MS, 83% for lesion load, and 77% for BMS vs. NBMS.
• Demonstrated that neuroanatomical GM patterns contain sufficient information for single-case MS classification.

Conclusions:

• Multivariate SVM analysis effectively classifies MS patients at the individual level based on neuroanatomical GM patterns.
• This approach shows promise as a clinical application for MS patient stratification and management.
• Neuroanatomical patterns derived from MRI scans provide valuable insights into MS heterogeneity.