QSM separates multiple sclerosis lesions into four distinct microstructural phenotypes
A new multiple sclerosis imaging analysis suggests that white matter lesions are not a single pathological entity on MRI, even when they appear similar on conventional scans. Using quantitative susceptibility source separation, investigators divided lesions into four reproducible phenotypes and then tied one of those phenotypes to greater physical disability. The work matters because standard lesion counts often flatten biology into a single burden metric, while susceptibility imaging can separate paramagnetic and diamagnetic tissue contributions that conventional quantitative susceptibility mapping tends to blur together through phase cancellation.
The study examined 183 people with multiple sclerosis and 3,327 quality-controlled white matter lesions acquired at 3T with a five-echo gradient-recalled echo sequence. A subsampled two-echo dataset was also tested to see whether a shorter acquisition could preserve the same microstructural signal. Investigators used a hybrid strategy: unsupervised clustering on whole-lesion paramagnetic and diamagnetic features, then rule-based assignment based on compartmental edge-core gradients. The resulting clustering was highly stable, with an adjusted rand index of 0.924, which indicates strong internal reproducibility across the lesion set.
Four lesion phenotypes emerged from susceptibility source separation
The four phenotypes were not evenly distributed. 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 was the smallest subgroup at 8.6%, but it carried the most clinically interesting signal: high χ para with a positive edge-core paramagnetic gradient.
A single lesion count could never have captured that spread.
The biology behind the classification is plausible and clinically important. Paramagnetic susceptibility can reflect iron-rich material, while diamagnetic effects are linked to myelin and other tissue constituents. When those signals are mixed together in standard QSM, opposing effects can cancel each other out and hide heterogeneity inside the lesion core, rim, or edge. Source separation tries to pull those components apart, making it easier to ask whether some lesions reflect more inflammatory activity, while others reflect more severe tissue destruction or repair failure.
Disability tracked with diamagnetic loss inside selected lesion compartments
The most important clinical association appeared not across all lesions, but inside specific compartments of specific subtypes. Higher Expanded Disability Status Scale scores were selectively associated with more negative signed χ dia within Type 4 lesions. The relationship held in the whole lesion, core, and edge, with beta coefficients of –0.37, –0.37, and –0.34, respectively, and q values of 0.020, 0.006, and 0.030. In Type 2 lesions, the edge also showed a significant association with disability, with a beta of –0.33 and q of 0.044.
The direction of effect matters. More negative diamagnetic signal implies less diamagnetic tissue contribution, which may point toward loss of myelin or other intralesional microstructural remodeling. The authors interpret this as a marker of complex tissue change rather than a simple binary lesion score. Reactive astrogliosis is one possible contributor, and non-myelin tissue remodeling is another. Either mechanism would fit a lesion environment that has moved beyond focal demyelination into a more chronic, compositionally altered state.
The Type 4 core association held up under additional models adjusting separately for total lesion volume, total lesion count, and disease duration. The coefficients remained negative and statistically significant in each setting, with q values of 0.004, 0.008, and 0.033. In practical terms, the disability signal was not just a reflection of having more lesions or a longer disease course. It was embedded in lesion microstructure itself.
Shorter echo sampling diluted the compartment signal
The abbreviated two-echo acquisition did not preserve the same strength of association. When investigators repeated the analysis with GRE2, the Type 4 compartmental links weakened to non-significance across all tested compartments, including the core where the beta was –0.20 with q = 0.215. This result suggests that reducing echo sampling may come at a cost when the goal is to resolve subtle susceptibility differences at the lesion edge and center.
That finding has a direct imaging implication. If a scanner protocol is too compressed, clinically relevant compartment information may disappear even when the overall lesion load still looks acceptable. The effect is unlikely to be random noise alone; susceptibility estimates depend on how much phase information is available across echoes, and fewer echoes can reduce the precision needed to separate paramagnetic from diamagnetic sources.
What the lesion biology implies for MS imaging practice
The study moves lesion assessment closer to tissue phenotyping rather than simple lesion counting. In multiple sclerosis, disability often reflects the cumulative impact of heterogeneous lesions distributed across time and anatomy, but two lesions with the same apparent size can differ sharply in their underlying composition. A lesion with preserved diamagnetic signal may represent a different pathological stage than one showing marked diamagnetic depletion and paramagnetic enrichment at the core or rim.
The clinical promise lies in stratification. If susceptibility-defined phenotypes can be reproduced in other cohorts, MRI could eventually distinguish lesions with greater potential to track disability from lesions that are biologically quieter. For now, the dataset is cross-sectional, so it cannot show progression or prove that one phenotype evolves into another. The cohort is also drawn from a single research environment, and the imaging approach still needs validation across scanners, sequences, and external populations.
Even so, the message is sharp: total lesion burden is only part of the story. Within the white matter of people with multiple sclerosis, susceptibility-defined microstructure carries information that conventional lesion metrics leave behind, and the most disability-linked signal in this preprint sits inside the diamagnetic core of a small but distinct lesion subtype.