Using Deep Learning to Measure MS-Specific Brain Damage Beyond Aging

A new study published in Neurology aimed to differentiate age-related brain changes from multiple sclerosis (MS)-specific neurodegeneration using MRI. The study proposed a disease-specific model to complement the brain-age gap (BAG).

Age is often treated as a confounder in neuroimaging analyses; however, brain aging and MS are intertwined. A deep learning 3D DenseNet architecture was trained to predict disease duration (DD) from brain MRI scans in people with MS (PwMS), alongside the DeepBrainNet model for age predictions. MRI scans from 4392 patients with MS were analyzed cross-sectionally. Additionally, 252 patients with early MS were studied longitudinally. The “DD gap” was calculated as the difference between predicted and actual DD, serving as a potential MS-specific biomarker.

Researchers found that the model predicted DD better than chance, with a mean error of 5.63 years and R2 of 0.34. The DD gap showed low correlation with BAG, suggesting that the 2 metrics are largely independent. It was also more sensitive to MS-specific lesions than BAG and was a significant predictor of disability as measured by the Expanded Disability Status Scale (EDSS), both cross-sectionally and longitudinally. Longitudinal increases in DD gap correlated with greater EDSS change and worsening disability, explaining more variance than BAG alone.

The researchers concluded that the DD gap reflects MS-specific brain damage beyond typical aging and may serve as a unique biomarker for MS severity and progression. Combining BAG with DD gap offers a more complete picture of brain health in PwMS by capturing both aging and disease-specific changes. However, limitations of the study include biases in the DD model, and additional studies are required to account for factors like treatment effects, comorbidities, and imaging inconsistencies.

“In conclusion, we demonstrated that the DD gap is a clinically meaningful measure of multiple sclerosis-specific brain damage, adding to models of healthy brain aging,” researchers stated. “By condensing the complex information contained in routinely acquired brain MRI scans into a simple and intuitive biomarker of disease severity and progression, it may represent a powerful tool for the stratification of PwMS in both clinical and research settings.”

Reference

Pontillo G, Prados F, Colman J, et al. Disentangling neurodegeneration from aging in multiple sclerosis using deep learning. Neurology. 2024;103(10):e209976. doi: 10.1212/WNL.0000000000209976