SuStaIn-inferred sequences of blood-brain barrier permeability and cortical thinning in cognitive impairment: a data-driven DCE-MRI study

Blood-brain barrier (BBB) disruption and cortical thinning are both implicated in neurodegenerative cognitive impairment, yet their temporal relationship remains unresolved. Cross-sectional studies have yielded conflicting results, likely reflecting disease heterogeneity. We aimed to identify subtypes with distinct model-inferred sequences of K trans elevation and cortical thinning and to compare their clinical characteristics. This retrospective multi-center study included participants with normal cognition, mild cognitive impairment (MCI), and dementia recruited from two centers. BBB permeability (K trans ) was derived from dynamic contrast-enhanced MRI, and cortical thickness was measured from structural MRI across six bilateral cortical regions. The Subtype and Stage Inference (SuStaIn) algorithm was used to infer subtypes and the relative ordering of imaging abnormalities from cross-sectional data. Subtype differences in cognition, cerebrospinal fluid biomarkers, APOE genotype, and small vessel disease markers were assessed. Among 216 participants (mean age 70.1 years; 64.4% female), SuStaIn identified three groups: Subtype 0 (normal-appearing; n = 99), Subtype 1 (atrophy-first; n = 58), and Subtype 2 (BBB-first; n = 59). Subtype 1 showed the highest dementia prevalence (46.6%), poorest cognitive performance, greatest cortical thinning, and higher APOE ε4 carrier frequency than Subtype 2 (20/50 [40.0%] vs. 10/50 [20.0%], p = 0.029). Subtype 2 exhibited the highest MCI prevalence (50.8%) and the highest K trans . CSF total tau was elevated in both modeled subtypes. Subtype 2 showed lower white matter hyperintensity burden than both other groups. SuStaIn identified atrophy-first and BBB-first patterns with differing clinical characteristics, suggesting heterogeneity in the relationship between BBB disruption and cortical thinning. Longitudinal studies are needed to establish whether these model-inferred sequences reflect changes within individuals. CRIS, KCT0008510, registered 2023-06-09, http://cris.nih.go.kr/cris/en/use_guide/cris_introduce.jsp ; KCT0008699, registered 2023-08-11; KCT0003181, registered 2018-09-14; all retrospectively registered.