Tract-based Quantitative MRI for Resolving the Clinico-Radiological Paradox in Multiple Sclerosis

The clinico-radiological paradox in multiple sclerosis (MS) arises because conventional MRI measures, particularly total lesion volume, fail to fully capture the true burden of disability. These broad volumetric measures overlook the dual influence of where lesions occur and what is their microstructural composition. This study aimed to improve the clinical relevance of lesion analysis in MS by refining both how and where lesions are measured. While traditional approaches emphasize anatomical lesion burden, studies of structural connectivity in neurological disease have shown that functional deficits often arise from tract-specific interruptions within distributed brain networks. We hypothesized that filtering lesions through clusters of functionally meaningful white matter tracts, rather than broad anatomical compartments, would enable more accurate identification of clinically relevant damage. To test our hypothesis, we studied 132 participants, including 89 patients with MS (49 relapsing–remitting, 17 primary progressive, 23 secondary progressive; 62 women and 27 men; mean age 41 years, range 18–68) and 43 healthy controls (28 women and 15 men; mean age 36 years, range 23–58). All underwent standardized 3 Tesla MRI including fluid-attenuated inversion recovery (FLAIR), T1 mapping, magnetization transfer ratio (MTR), and diffusion tensor imaging (DTI) with fractional anisotropy (FA) and mean diffusivity (MD).

We evaluated associations between imaging measures and disability, and assessed predictive performance with ridge-penalized regression across Expanded Disability Status Scale (EDSS), and motor scores Timed 25-Foot Walk (T25FW), Nine-Hole Peg Test (9HPT), as well as progression risk defined by the MSPro scale. Filtering lesions through functionally critical tracts revealed associations obscured by conventional approaches. Tract-based lesion T1 values correlated more strongly with disability than classical regional metrics (e.g., EDSS r = 0.83 in occipitoparietal tracts vs r = 0.47 in periventricular regions). Multimodal tract-specific metrics including T1, MTR, FA, and MD significantly predicted disability. In multivariate models, tract-based frameworks consistently outperformed classical region-based ones, achieving higher discrimination and better model fit for binarized EDSS and MSPro (AUC = 0.86–0.96 for tract-based vs 0.57–0.86 for classical anatomical regions). For continuous outcomes (T25FW, 9HPT), tract-based models explained substantially more variance (R² = 0.19–0.43 for tract-based vs 0.01–0.34 for classical regions). Including demographic covariates improved model fit in both approaches, yet tract-based quantitative metrics remained superior to classical region-based measures.

By integrating lesion location and microstructural composition, tract-based quantitative MRI enhances disability prediction and provides interpretable biomarkers to support individualized monitoring and therapeutic decision-making.