Distinct Spatiotemporal Patterns of White Matter Hyperintensity Progression

White matter hyperintensity (WMH), a key imaging marker for brain health and the most prevalent brain abnormality in the general population, has prognostic implications for stroke. However, as a prognostic factor, WMH is hard to apply because it varies in both location and size. Reducing this complexity into a set of recognizable patterns would improve the utility of WMH as a cerebrovascular biomarker. In this paper, we show three distinct spatiotemporal patterns of WMH progression in ischemic stroke patients. We used a multicenter MRI database of 9,719 consecutive patients with acute ischemic stroke, plus the UK Biobank database ( n = 36,210 controls). Subtype and Stage Inference (SuStaIn) modeling was performed to simultaneously subtype and stage WMH based on longitudinal inference from the Korean multicenter cross-sectional data, after normalizing WMH volumes of our stroke patients to the age-adjusted WMH volumes of low-risk controls ( n = 13,811) from the UK Biobank database. Then, we rigorously characterized demographic profiles, vascular risk factors, and stroke outcomes across three different WMH progression trajectories: fronto-parietal (subtype 1: T1), radial (T2), and temporo-occipital (T3) progression. Subtypes remained consistent between baseline and follow-up (median [interquartile range] years of follow-up: 4.5 [2.8□7.3]) in the majority (> 77%) of patients during WMH progression. T1 showed relatively delayed WMH onset and was enriched for women and hypertension, whereas T3 showed earlier WMH onset and was enriched for men, atrial fibrillation, and coronary artery disease. Hypertension and diabetes mellitus were major risk factors that could accelerate WMH progression in all subtypes. The index stroke, i.e., acute symptomatic infarction, was more likely to be due to small vessel occlusion in T1 and cardioembolism in T3. Regarding post-stroke outcomes, early (< 3 weeks) neurological deterioration by symptomatic hemorrhagic transformation and 3-month unfavorable functional outcome were more frequent in T3, while 1-year ischemic stroke recurrence was more frequent in T1. We also found that our WMH subtyping–staging model and the widely used Fazekas WMH scale were consistent while also offering complementary strengths in profiling stroke risk factors and outcomes. Using our model, high-risk controls (without a history of neurological diseases but with vascular risk factors, n = 22,399) in the UK Biobank were assigned to their most likely WMH subtype and stage, demonstrating spatiotemporal patterns of WMH progression that closely paralleled those observed in stroke patients. Of note, we observed distinct stage distributions between high-risk controls and stroke patients, along with a higher area under the curve in the receiver operating curve analysis for distinguishing stroke patients from high-risk controls, compared to that of WMH volume. In conclusion, we identified distinct spatiotemporal trajectories of WMH progression. This WMH subtyping–staging model can capture demographic and vascular risk factors and may provide useful predictions for stroke outcomes across different WMH subtypes and stages. Further prospective investigation is required to confirm whether our model can predict future stroke occurrence more effectively than WMH volume.