Multiple sclerosis exhibits marked biological heterogeneity that conventional bedside phenotypes often fail to capture. Historically, clinicians classified disease courses into relapsing or progressive forms based purely on overt clinical attacks. However, macroscopic clinical observations frequently obscure subtle underlying microstructural degradation. Consequently, researchers leveraged machine learning to uncover distinct multiple sclerosis MRI subtypes that directly govern disease severity, cognitive decline, and long-term disability. Ultimately, decoding these complex neuroimaging patterns empowers clinicians to anticipate silent neurodegeneration and optimize therapeutic strategies.

Deconstructing Disease Heterogeneity with Machine Learning

For decades, neurologists have relied on separate radiological metrics to evaluate central nervous system demyelination. In routine practice, specialists measure T2-hyperintense white matter lesion volume, regional cortical atrophy, and deep grey matter loss independently. However, evaluating these features in isolation obscures how distinct injury patterns evolve together over time. To resolve this challenge, investigators applied an unsupervised machine learning algorithm known as Subtype and Stage Inference, or SuStaIn.

The research team analyzed high-resolution 3 Tesla brain MRI scans from 1017 patients with multiple sclerosis and 548 healthy controls. In addition, the team validated their machine learning framework within an independent cohort of 247 patients and 141 controls. The analytical algorithm assessed T2 lesion volume, mean diffusivity across white matter tracts, and regional grey matter volumes. Furthermore, longitudinal clinical follow-up spanned a median of 6.57 years in 645 patients. Therefore, this data-driven technique successfully reconstructed spatial and temporal trajectories of tissue destruction without relying on subjective clinical milestones. As a result, the algorithm established a purely biological framework for disease classification.

Characterizing the Four Multiple Sclerosis MRI Subtypes

The machine learning algorithm uncovered four biologically distinct multiple sclerosis MRI subtypes across the cohort. Specifically, the model classified patients into lesion-led (44%), cortex-led (23%), tract-led (23%), and deep grey matter or cerebellar-led (10%) subtypes. Each subtype originates in a specific anatomical compartment before spreading to other brain regions. For example, the lesion-led subtype exhibits early accumulation of focal white matter lesions before significant cortical thinning occurs. Conversely, the tract-led subtype shows early microstructural disruption within major white matter tracts, demonstrated by elevated mean diffusivity.

In contrast, grey matter-led phenotypes follow fundamentally different destructive pathways. The cortex-led subtype demonstrates prominent, early neocortical thinning prior to widespread lesion accrual. Meanwhile, the deep grey matter and cerebellar-led subtype displays early atrophy concentrated within the thalamus, basal ganglia, and cerebellum. Notably, relapsing-remitting disease dominated the lesion-led (66%) and tract-led (71%) subtypes. Conversely, progressive multiple sclerosis phenotypes heavily enriched the cortex-led (47%) and deep grey matter or cerebellar-led (56%) groups. Therefore, these imaging subtypes delineate authentic biological variations that transcend traditional clinical boundaries.

Clinical Trajectories and Relapse-Independent Disability

These imaging subtypes demonstrated remarkable divergence in long-term disability accrual and clinical disease trajectories. Crucially, the researchers analyzed progression independent of relapse activity, commonly termed PIRA. In clinical practice, progression independent of relapse activity represents insidious, smoldering neurodegeneration that standard anti-inflammatory therapies often fail to halt. Specifically, patients belonging to the deep grey matter and cerebellar-led subtype exhibited the highest baseline disability scores. Furthermore, this subgroup accumulated physical disability significantly faster during longitudinal follow-up than individuals with lesion-led disease.

Because deep grey matter and cerebellar structures coordinate complex motor networks, early injury in these zones rapidly exhausts functional brain reserves. In contrast, patients with the tract-led subtype demonstrated relatively stable ambulatory scores over time, despite substantial diffusion abnormalities. Moreover, while lesion-led patients experienced frequent acute relapses, their rate of relapse-independent progression remained comparatively low. Consequently, these findings prove that physical disability does not stem solely from cumulative white matter lesion burden. Clinicians must recognize that early deep grey matter neurodegeneration portends rapid functional decline.

Cognitive Impairment and Genetic Severity Associations

Cognitive impairment heavily burdens patients with multiple sclerosis, yet conventional imaging provides minimal predictive insight. In this study, standardized neuropsychological testing in 501 patients revealed striking cognitive differences across the four imaging subtypes. Notably, patients assigned to the cortex-led subtype exhibited the earliest age of disease onset alongside the most severe cognitive deficits. Because cortical regions govern information processing speed, executive function, and verbal memory, early cortical volume loss precipitates pronounced neuropsychological impairment.

Furthermore, genetic profiling in 650 individuals established critical biological validation for these computational subtypes. The researchers demonstrated that patients carrying genetic variants linked to severe disease clustered predominantly within the cortex-led and deep grey matter-led groups. Specifically, these patients carried higher polygenic risk for progressive neurodegeneration rather than purely inflammatory lesion formation. In addition, individuals within grey matter-predominant subtypes reached severe disability milestones significantly faster. Therefore, combining genetic susceptibility profiles with neuroimaging subtypes confirms that these clusters reflect genuine biological entities. Ultimately, this comprehensive profiling strategy enhances our understanding of neurodegenerative mechanisms in chronic autoimmune disease.

Translational Value for Clinical Practice and Future Directions

For Indian neurologists, managing multiple sclerosis requires balancing complex diagnostic considerations against variable healthcare resources. Clinicians across India frequently encounter diagnostic challenges when distinguishing multiple sclerosis from neuromyelitis optica spectrum disorders or acute disseminated encephalomyelitis. Moreover, recognizing aggressive biological trajectories early remains essential to prevent permanent motor disability in young working-age individuals. Incorporating automated quantitative neuroimaging into routine tertiary practice offers actionable prognostic guidance. As 3 Tesla MRI systems become widespread across Indian diagnostic centers, clinicians can integrate volumetric and diffusion analysis into standard protocols.

Additionally, identifying biological subtypes assists physicians in selecting tailored treatment strategies. For instance, patients presenting with the lesion-led subtype typically achieve excellent disease control with high-efficacy anti-inflammatory disease-modifying therapies, such as monoclonal antibodies. Conversely, individuals exhibiting cortex-led or deep grey matter-led subtypes demand aggressive early intervention alongside neuroprotective strategies and targeted cognitive rehabilitation. Meanwhile, ongoing computational advances will soon enable automated subtype assignment directly within diagnostic radiology workstations. Consequently, adopting advanced neuroimaging paradigms will substantially elevate precision multiple sclerosis care across Indian healthcare institutions.

Frequently Asked Questions

How do multiple sclerosis MRI subtypes differ from traditional clinical phenotypes?

Traditional phenotypes categorize multiple sclerosis purely by subjective clinical relapses and observable symptom progression. In contrast, MRI subtypes identify objective biological damage patterns using machine learning. These computational subtypes distinguish between patients whose disease begins primarily in white matter lesions, major tracts, the cerebral cortex, or deep grey matter, revealing underlying pathophysiological differences.

Which imaging subtype carries the highest risk of progression independent of relapse activity?

The deep grey matter and cerebellar-led subtype carries the greatest risk of progression independent of relapse activity. Patients classified within this subgroup experience accelerated accumulation of sustained physical disability. Because basal ganglia and cerebellar nuclei govern essential motor and coordination networks, early structural degradation within these regions rapidly overwhelms neurological functional reserve.

Can standard 3 Tesla MRI protocols identify these biological subtypes in routine clinical workflows?

Standard clinical 3 Tesla MRI protocols can capture the necessary volumetric and diffusion imaging sequences. However, classifying individual subtypes requires automated post-processing software pipelines to calculate tissue volumes and mean diffusivity accurately. As artificial intelligence tools integrate into clinical neuroimaging software, automated subtype classification will become practical for routine neurological practice.

Disclaimer: This content is for informational and educational purposes only. It is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified health provider with any questions you may have regarding a medical condition. Refer to the latest local and national guidelines for clinical practice.

References

1. Preziosa P et al. Imaging subtypes reveal distinct biological substrates and disability profiles in multiple sclerosis. Brain. 2026 Nov 04. doi: 10.1093/brain/awag320. PMID: 42773544.
2. Eshaghi A, Young AL, Wijeratne PA, et al. Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data. Nat Commun. 2021;12(1):2078. doi: 10.1038/s41467-021-22265-2.
3. Giovannoni G, Popescu V, Wuerfel J, et al. Smouldering multiple sclerosis: the 'real MS'. Ther Adv Neurol Disord. 2022;15:17562864211066751. doi: 10.1177/17562864211066751.