Susceptibility mapping splits multiple sclerosis lesions into four reproducible phenotypes
A quantitative MRI approach has separated multiple sclerosis lesions into four internally stable phenotypes and linked a subset of those patterns to disability. The work uses quantitative susceptibility source separation, or χ-separation, to pull apart the competing magnetic signatures of paramagnetic and diamagnetic tissue sources inside white matter lesions, where conventional susceptibility mapping can blur iron-related and myelin-related effects into a single composite signal.
That separation matters because MS lesions are not biologically uniform. Some contain preserved diamagnetic signal, some carry strong paramagnetic content, and others show gradients across the edge and core that likely reflect different stages of injury and repair. In practical terms, a lesion counted once on routine imaging may hide several microstructural states with different clinical meaning.
How the imaging framework classified lesion biology
The cross-sectional study included 183 people with MS and analyzed 3,327 quality-controlled white matter lesions acquired at 3T. Investigators used a five-echo gradient-recalled echo sequence, then repeated the analysis with a subsampled two-echo dataset to test whether a shorter acquisition could retain biologic sensitivity. The imaging pipeline combined unsupervised clustering of whole-lesion paramagnetic (χ para) and diamagnetic (χ dia) features with rule-based compartmental assignment at the lesion edge and core.
That hybrid approach yielded four reproducible lesion phenotypes with an adjusted Rand index of 0.924, a level of agreement that points to a highly stable classification structure. Type 1 accounted for 22.7% of lesions and showed preserved χ dia. Type 2 made up 32.9% and showed high χ para without a positive edge-core gradient. Type 3 represented 35.8% and showed attenuated χ dia. Type 4, the smallest group at 8.6%, combined high χ para with a positive edge-core paramagnetic gradient.
A single lesion class did not dominate the dataset. Instead, the lesion burden fragmented into a mixed microstructural population, and the smallest subtype carried the clearest clinical signal.
Disability tracked with diamagnetic loss inside the most altered lesions
Higher physical disability, measured by the Expanded Disability Status Scale, was selectively associated with more negative signed χ dia in Type 4 lesions. The association appeared across the whole lesion, the core, and the edge, with q values of 0.020, 0.006, and 0.030, respectively. In Type 2 lesions, the edge also carried a significant association, with more negative signed χ dia linked to higher EDSS and a q value of 0.044.
The pattern is biologically plausible. Diamagnetic signal in this setting is generally interpreted as a surrogate for myelin-rich tissue, so a more negative χ dia may reflect deeper demyelination, altered tissue architecture, or related remodeling in lesions that already show strong paramagnetic features. The authors interpret the signal as potentially reflecting non-myelin tissue remodeling or reactive astrogliosis, and the data fit that framework better than a simple lesion-count explanation.
The Type 4 core findings held up after adjustment for total lesion volume, total lesion count, and disease duration. Across those sensitivity models, the core association remained significant, with β values ranging from –0.32 to –0.37 and q values between 0.004 and 0.033. The relationship therefore did not depend on the obvious burden variables that usually drive disability analyses in MS.
A shorter echo protocol weakened the compartment signal
The abbreviated two-echo acquisition changed the picture. When the same lesion phenotype analysis was repeated with GRE2 rather than GRE5, the compartment-specific associations in Type 4 fell to non-significance, with all q values above 0.05 and the core association shrinking to β = –0.20 with q = 0.215.
This is not a trivial technical footnote. It suggests that the extra echo information in the five-echo protocol captures subtle susceptibility differences that two echoes may miss, especially in the most heterogeneous lesions. For centers considering whether shorter MRI protocols can preserve biologic granularity, the answer from this dataset is cautious: a reduced acquisition may be adequate for broad lesion detection, but it may erode the compartment-level signal needed to connect lesion microstructure with disability.
Why paramagnetic and diamagnetic signals need to be separated
Conventional quantitative susceptibility mapping can be difficult to interpret in MS because paramagnetic and diamagnetic sources push the signal in opposite directions. Iron deposition, inflammatory remnants, and other paramagnetic contributors can coexist with myelin loss and protein-related diamagnetic effects inside the same lesion. When those signals cancel each other out, the resulting image may look averaged and biologically flat even when the lesion is internally diverse.
χ-separation tries to break that cancellation apart. By estimating paramagnetic and diamagnetic components separately, the method gives a more direct readout of lesion composition rather than a blended susceptibility score. In a disease where disability often lags behind obvious lesion counts, that kind of separation could help explain why two patients with similar MRI burden do not walk, think, or function the same way.
What this adds to MS imaging and where the evidence still stops
The study supports a view of MS lesions as structured microenvironments rather than interchangeable scars. It also suggests that lesion phenotype, not just lesion number, can carry disability information. The strongest clinical association emerged in the small but distinctive Type 4 group, where high paramagnetic content and a positive edge-core gradient coincided with worse EDSS scores and lower diamagnetic measures in the lesion core.
Because the work is cross-sectional, it cannot show whether these phenotypes predict future disability, relapse biology, or treatment response. The cohort also had smaller clinical subgroups for some outcome analyses, including SDMT and disease duration, so those relationships need more cautious interpretation than the EDSS findings. Even so, the data establish a reproducible imaging taxonomy and a clear signal that lesion microstructure contains clinical information beyond lesion burden alone.
The next question is not whether MS lesions are heterogeneous; this dataset shows they are. The remaining task is to determine whether susceptibility-defined lesion phenotypes can be used prospectively to forecast progression, monitor treatment effects, or identify patients whose disability is being driven by a particularly destructive lesion biology.