A dual space MRI radiomic network signature for risk stratification and subtyping of mild cognitive impairment

Mild cognitive impairment (MCI) is a clinically heterogeneous prodromal stage of Alzheimer's disease (AD) in which accurate risk stratification could support earlier intervention and more efficient trial design. Using baseline T1-weighted MRI from the Alzheimer's Disease Neuroimaging Initiative and the National Alzheimer's Coordinating Center cohorts, we developed a dual-space anatomical radiomic-network signature that integrates regional radiomic features and radiomics similarity network metrics extracted in standard and native spaces. The signature improved prediction of MCI-to-AD conversion relative to regional volumetric and clinical models, achieved consistent external validation performance, and separated participants into high- and low-risk groups. The same feature set identified five imaging-defined MCI subtypes with distinct anatomical patterns, conversion risks, and clinical, APOE ε4, and cerebrospinal fluid biomarker profiles; fronto-parietal and parahippocampal-temporal subtypes showed the highest risk. These findings support structural MRI radiomic-network profiling as a complementary framework for individualized prognosis and cohort stratification in AD research.