Combined magnetic resonance imaging and serum analysis reveals distinct multiple sclerosis types

Charles Willard, Lemuel Puglisi, Daniele Ravì, Mariia Dmitrieva, Rozemarijn M. Mattiesing, Frederik Barkhof, Daniel C. Alexander, Danielle E. Harlow, Daniela Piani-Meier, Arman EshaghiBrain 2025 · DOI

TL;DR

This paper introduces a novel machine learning model that combines MRI scans and serum neurofilament light chain (sNfL) levels to identify distinct subtypes of Multiple Sclerosis (MS). These "early-sNfL" and "late-sNfL" subtypes offer a more biologically grounded way to understand MS progression and predict treatment response compared to MRI alone.

What Problem Does This Paper Address?

Multiple Sclerosis (MS) is a complex neurological disease with highly variable clinical presentations and progression rates. Current diagnostic categories (relapsing-remitting, secondary progressive) often fail to capture the underlying biological differences that drive individual disease courses. This makes it challenging to predict how a patient's MS will evolve and to tailor treatments effectively. While MRI provides valuable structural information about brain damage, it doesn't fully capture the dynamic processes of neurodegeneration and inflammation. Serum neurofilament light chain (sNfL) is a promising biomarker for neuronal injury, but its utility can be limited by variability and assay standardization. The authors aim to bridge this gap by integrating these two data modalities to create a more precise and biologically informed subtyping system for MS.

Methods

The core of this research is an unsupervised machine learning approach called Subtype and Stage Inference (SuStaIn). The authors adapted SuStaIn to integrate both MRI-derived features and sNfL levels.

  • Feature Selection: They started with 35 MRI features and selected five based on their correlation with the Expanded Disability Status Scale (EDSS). These five features, along with sNfL, formed the basis of their combined model.
  • Model Training: The SuStaIn model was trained on a cohort of 189 MS patients. The algorithm learns to identify distinct "subtypes" by analyzing the order in which different biomarkers (MRI features and sNfL) reach abnormal stages. It also assigns a "stage" to each patient, indicating their position along a disease progression trajectory.
  • Harmonization: To account for potential differences in MRI scanner protocols between datasets, they used the ComBat algorithm to harmonize the data from the external test set with the training set.
  • Validation: The trained model was then applied to an independent external test set of 445 newly diagnosed MS patients to validate its performance and generalizability.
  • Comparison: They compared their combined MRI-sNfL model against an existing MRI-only SuStaIn model and a reduced 5-MRI model to demonstrate the added value of incorporating sNfL.

Datasets

  • Training Dataset: 189 participants from a previously published phase 2 clinical trial of evobrutinib. This cohort included patients with relapsing-remitting and secondary progressive MS.
  • External Test Dataset: 445 participants from a phase 3 clinical trial of interferon beta-1a. This cohort consisted of newly diagnosed MS patients and those with clinically isolated syndrome.

The study did not introduce new benchmarks but rather validated their model on existing, well-characterized clinical trial datasets.

Key Insights and Ideas

  • sNfL is a Game Changer for MS Subtyping: Integrating sNfL with MRI significantly improves the ability to correlate data-derived stages with clinical outcomes like disability (EDSS) and disease activity, outperforming MRI-only models.
  • Two Distinct Biological Trajectories Emerge: The model identified two key subtypes: "early-sNfL" (characterized by early sNfL elevation, corpus callosum injury, and lesion accrual) and "late-sNfL" (marked by early volume loss in grey matter and later sNfL elevation). This suggests fundamentally different underlying disease processes.
  • Subtypes Predict Longitudinal Outcomes: The identified subtypes are not just static classifications; they predict future disease activity, such as the risk of new lesion formation and treatment response.
  • Biological Subtypes Transcend Clinical Categories: The subtypes show distinct demographic and biological profiles, suggesting they are more biologically grounded than traditional clinical classifications. This highlights the potential for personalized medicine in MS.

Key Results

  • Improved Correlation with EDSS: The MRI-sNfL model showed a significantly stronger correlation with EDSS in the training set (Spearman's ρ = 0.420 vs. 0.231 for MRI-only) and maintained a significant correlation in the external test set (ρ = 0.163 vs. 0.067 for MRI-only).
  • Predicting Radiological Activity: The early-sNfL subtype had a 144% increased risk of new lesion formation compared to the late-sNfL subtype in the training set (HR = 2.44).
  • Treatment Response Differences: In treated patients, the early-sNfL group showed a significantly faster reduction in active lesion counts (P = 0.01) and a faster rate of brain volume loss (P = 0.038 in the external test set) compared to the late-sNfL group.
  • sNfL Dynamics: The early-sNfL subtype showed significant reductions in sNfL levels over time, particularly with treatment, while the late-sNfL subtype showed less pronounced changes.

Limitations and Future Work

  • Clinical Trial Cohorts: The study primarily used data from clinical trials, which have specific eligibility criteria. This means the findings may not fully represent the broader MS population, including those with comorbidities or primary progressive MS.
  • Limited EDSS Range in Test Set: The external test set consisted of newly diagnosed patients, leading to a limited range of EDSS scores, which could affect the correlation analysis.
  • Model Assumptions: The SuStaIn algorithm assumes a monotonic progression of biomarker changes, which might not capture all complex disease trajectories.
  • Future Work: The authors suggest the need for longer prospective studies to validate these subtypes with broader outcome measures (cognition, quality of life). They also emphasize the potential for this multimodal model to evolve into a clinical decision aid system, especially as sNfL becomes more widely available. Further research could also explore the integration of other fluid biomarkers or advanced imaging modalities for a more comprehensive understanding of MS progression.

Summary updated

Abstract

Multiple sclerosis (MS) is a highly heterogeneous disease in its clinical manifestation and progression. Predicting individual disease courses is key for aligning treatments with underlying pathobiology. We developed an unsupervised machine learning model integrating MRI-derived measures with serum neurofilament light chain (sNfL) levels to identify biologically informed MS subtypes and stages. Using a training cohort of patients with relapsing-remitting and secondary progressive MS (n = 189), with validation on a newly diagnosed population (n = 445), we discovered two distinct subtypes defined by the timing of sNfL elevation and MRI abnormalities (early- and late-sNfL types). In comparison to MRI-only models, incorporating sNfL with MRI improved correlations of data-derived stages with the Expanded Disability Status Scale in the training (Spearman's ρ = 0.420 versus MRI-only ρ = 0.231, P = 0.001) and external test sets (ρ = 0.163 for MRI-sNfL, versus ρ = 0.067 for MRI-only). The early-sNfL subtype showed elevated sNfL, corpus callosum injury and early lesion accrual, reflecting more active inflammation and neurodegeneration, whereas the late-sNfL group showed early volume loss in the cortical and deep grey matter volumes, with later sNfL elevation. Cross-sectional subtyping predicted longitudinal radiological activity: the early-sNfL group showed a 144% increased risk of new lesion formation (hazard ratio = 2.44, 95% confidence interval 1.38-4.30, P < 0.005) compared with the late-sNfL group. Baseline subtyping, over time, predicted treatment effect on new lesion formation on the external test set (faster lesion accrual in early-sNfL compared with late-sNfL, P = 0.01), in addition to treatment effects on brain atrophy (early sNfL average percentage brain volume change: -0.41, late-sNfL = -0.31, P = 0.04). Integration of sNfL provides an improved framework in comparison to MRI-only subtyping of MS to stage disease progression and inform prognosis. Our model predicted treatment responsiveness in early, more active disease states. This approach offers a powerful alternative to conventional clinical phenotypes and supports future efforts to refine prognostication and guide personalized therapy in MS.
BibTeX
@inproceedings{willard2025combined,
  author = {Charles Willard and Lemuel Puglisi and Daniele Ravì and Mariia Dmitrieva and Rozemarijn M. Mattiesing and Frederik Barkhof and Daniel C. Alexander and Danielle E. Harlow and Daniela Piani-Meier and Arman Eshaghi},
  title = {{Combined magnetic resonance imaging and serum analysis reveals distinct multiple sclerosis types}},
  booktitle = {Brain},
  year = {2025},
  doi = {10.1093/brain/awaf331},
}
Figure