MRI Tract Patterns Sort ALS Patients into Three Subtypes Linked to Disease Severity

A new analysis of diffusion MRI scans from people with amyotrophic lateral sclerosis (ALS) suggests that the disease’s white matter damage does not follow a single uniform pattern. Instead, a statistical model grouped 73 patients into three distinct subtypes based on which tracts showed abnormal diffusion, and assigned each patient a disease stage that tracked with clinical measures of severity. The study, published in the Journal of Neurology, used a machine learning approach called Subtype and Stage Inference, or SuStaIn, to move beyond the averaged group comparisons that dominate conventional diffusion tensor imaging research.

ALS is a progressive neurodegenerative disease that affects motor neurons, and patients vary widely in how quickly symptoms advance and which parts of the nervous system appear most affected. Conventional diffusion tensor imaging analyses typically compare a group of patients against healthy controls and report average differences across the brain. This approach provides limited information on whether tract-level abnormalities follow distinct patient-level patterns. The researchers behind the new work set out to address that gap by treating fractional anisotropy and mean diffusivity values from individual white matter tracts as biomarkers that could be modeled simultaneously across patients.

The study was a retrospective secondary analysis that included 73 patients with ALS and 30 healthy controls. The researchers selected 11 predefined white matter tracts and used two standard diffusion measures from each: fractional anisotropy, which reflects the directional organization of water diffusion and tends to drop when white matter integrity is disrupted, and mean diffusivity, which rises as tissue barriers break down. Healthy controls provided the normative reference for Z-score standardization, meaning each patient’s tract values were expressed as deviations from the healthy range. The SuStaIn algorithm then searched for patterns in how these deviations co-occurred across patients, testing whether the data were better explained by one homogeneous group or by multiple subtypes progressing through ordered stages of tract involvement.

The three-subtype model emerged as the best fit. It classified patients into a cingulate callosal motor mixed subtype with 34 patients, a hippocampal cingulum posterior projection subtype with 32 patients, and a posterior callosal projection subtype that was mean diffusivity predominant with 7 patients. Each subtype name reflects the tracts that showed the earliest or most prominent abnormalities in that group, giving the patterns an anatomical interpretation rather than a purely statistical one.

Exploratory comparisons between the subtypes revealed differences in clinical measures. Patients in subtype 0, the cingulate callosal motor mixed group, had lower peak width of skeletonized mean diffusivity, or PSMD, than patients in subtypes 1 and 2, with both comparisons reaching p = 0.026. PSMD is a diffusion-derived marker of white matter microstructural damage that has gained attention in dementia research and is increasingly applied to other neurodegenerative diseases. Lower PSMD in subtype 0 suggests less widespread white matter damage in that group compared with the other two.

Patients in subtype 2, the posterior callosal projection MD predominant group, showed lower scores on the Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised, or ALSFRS-R, than patients in subtype 1, with p = 0.026. The ALSFRS-R is a widely used clinical questionnaire that measures functional ability across speech, swallowing, fine motor tasks, and mobility, with lower scores indicating greater impairment. The same subtype 2 group also had higher SuStaIn stage estimates than subtype 0, with p = 0.025. Because subtype 2 contained only 7 patients, these comparisons are exploratory and should be interpreted with caution.

Beyond the subtype differences, the SuStaIn stage estimates correlated with established clinical staging tools. Higher SuStaIn stage was associated with higher King’s stage, a clinical system that grades ALS severity based on the anatomical spread of symptoms beyond the initial region, with a Spearman correlation of rho = 0.318 and p = 0.006. Higher SuStaIn stage also correlated with lower ALSFRS-R scores, with rho = −0.289 and p = 0.013. These correlations indicate that the imaging-derived staging captured meaningful variation in disease burden, though the effect sizes are modest.

The researchers then tested whether these associations held after statistical adjustment. Higher SuStaIn stage remained associated with lower ALSFRS-R scores, with a standardized beta of −0.333 and p = 0.013, and with more advanced King’s stage, with an odds ratio of 1.181 and p < 0.001. The adjusted results suggest that the imaging-based stage estimates carry information about clinical severity that is not simply explained by other measured variables.

The authors conclude that tract-level diffusion SuStaIn identified anatomically interpretable patterns of white matter involvement in ALS and provided patient-level stage estimates associated with clinical disease burden. If validated in larger and independent cohorts, this approach could eventually help stratify patients in clinical trials by biological subtype rather than by symptom onset alone, potentially reducing the heterogeneity that has complicated ALS drug development. For now, the study demonstrates that a data-driven decomposition of diffusion MRI can reveal structure in what previously appeared to be a single averaged disease process.

Subject of Research: Diffusion MRI subtyping and staging of white matter tract abnormalities in amyotrophic lateral sclerosis

Article Title: Subtype and stage inference of tract-level diffusion abnormalities in amyotrophic lateral sclerosis

Article References: Yin, Y., Tu, Y., Wang, B., Chen, Y., Zhao, Y., Tang, J., Zhang, M., & Zhang, J. (2026). Subtype and stage inference of tract-level diffusion abnormalities in amyotrophic lateral sclerosis. Journal of Neurology, 273(11), Article 698. https://doi.org/10.1007/s00415-026-14138-9

Image Credits: AI Generated

DOI: 10.1007/s00415-026-14138-9

Keywords: ALS, diffusion tensor imaging, SuStaIn, white matter tracts, fractional anisotropy, mean diffusivity, PSMD, ALSFRS-R, King's stage, disease subtypes, neurodegeneration, machine learning