Abstract Details

Although there is poor correlation between changes in brain volume and physical disability, brain magnetic resonance imaging (MRI) volumetric outcomes are currently used to evaluate treatment efficacy in phase II trials targeting neuro-degenerative aspects of multiple sclerosis (MS). Recent studies suggest spinal cord (SC) volume may be a better predictor of MS disability.

To develop, optimize, and validate computational models predicting clinical disability in MS using brain and SC MRI volumetric data.

MRI imaging and neuroexam were performed on 649 subjects that were split into a training (n=457: 80 healthy volunteers; 100 Non-MS; 277 MS) and a validation cohort (n=192: 60 Non-MS; 132 MS). Participants were enrolled in the natural history protocol (Clinicaltrials.gov identifier NCT00794352) and gave informed consent. Lesion-TOADS algorithm implemented on QMENTA platform derived brain volumetric data. Spinal Cord Toolbox calculated upper cervical SC volume by investigators blinded to clinical outcomes. Neurological examinations were transcribed into NeurEx App that automatically calculates MS disability scales: Expanded Disability Status Scale (EDSS), Combinatorial weight-adjusted disability score (CombiWISE), NeurEx, and Symbol Digit Modalities Test (SDMT). Predictive models were optimized using Generalized Boosting Machine (GBM) algorithm.

GBM models using MRI volumetric data to predict clinical disability, optimized in the training cohort, were all strongly validated in an independent validation cohort: EDSS (Rho=0.58, R²=0.32, p<2.2e-16), CombiWISE (Rho=0.6, R²=0.35, p< 2.2e-16); NeurEx (Rho=0.61, R²=0.32, p<2.2e-16); SDMT (Rho=0.47, R²=0.22, p=9.9e-11). Age was the strongest predictor in all models except for NeurEx. SC and cerebellum white matter volumes were the most influential factors in predicting physical disability (CombiWISE, EDSS, and NeurEx), while thalamus and T2 lesion volumes were important for predicting cognitive disability (SDMT).

Optimized models that utilize brain and SC volumetric measures can be used to predict clinical disability in subjects with available imaging data and increase the  utility of existing datasets lacking clinical evaluation.