Machine learning-optimized Combinatorial MRI scale (COMRISv2) correlates highly with cognitive and physical...

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Composite MRI scales correlate better with clinical outcomes in multiple sclerosis (MS) than individual measures.
• Previous work developed the Combinatorial MRI scale (COMRISv1) using semi-quantitative MRI (semi-qMRI) biomarkers.
• The potential for improved prediction using quantitative MRI (qMRI) and advanced ML algorithms remains to be fully explored.

Purpose of the Study:

• To develop and validate an improved composite MRI scale (COMRISv2) for predicting clinical outcomes in MS patients.
• To assess the added value of quantitative MRI (qMRI) volumetric features and a more powerful ML algorithm compared to COMRISv1.
• To evaluate the performance of COMRISv2 in predicting both cognitive and physical disability in MS.

Main Methods:

• Prospective acquisition of brain MRI data and clinical evaluations from MS patients, divided into training (n=172) and validation (n=83) cohorts.
• Utilized the NeurEx™ App for automatic computation of disability scales from neurological examinations and the lesion-TOADS algorithm for qMRI feature extraction.
• Employed a modified random forest pipeline to select optimal biomarkers and develop COMRISv2 models, subsequently validated for predictive accuracy.

Main Results:

• COMRISv2 models demonstrated moderate correlation with cognitive disability (Spearman Rho=0.674, CCC=0.458) and strong correlations with physical disability (Spearman Rho=0.830-0.852, CCC=0.789-0.823).
• The model incorporating NeurEx™ data yielded the strongest predictive performance.
• Inclusion of qMRI features specifically enhanced the prediction of cognitive disability, suggesting semi-qMRI's accuracy in infratentorial injury assessment.

Conclusions:

• COMRISv2 models exhibit remarkable criterion validity in predicting granular clinical scales in MS patients.
• The integration of qMRI and advanced ML algorithms significantly enhances the predictive power of composite MRI scales for MS disability.
• COMRISv2 expands the scientific utility of MS research cohorts, particularly those with incomplete clinical data, by providing robust outcome prediction.