Discovery of MRI biomarker for early Alzheimer's disease integrating regional atrophy and ventricular expansion

Abstract

Timely detection of neurodegeneration in Alzheimer’s disease, particularly during the early and late mild cognitive impairment stages (EMCI and LMCI, respectively), is crucial for effective clinical intervention and disease management. Despite widespread use of absolute and normalised brain or ventricular volumes as imaging biomarkers, the relative expansion of ventricles per unit of brain tissue remains underexplored. We introduce the Ventricle-to-Tissue Ratio (VTR), and intuitive biomarker capturing this ratio, and define its global (VTRG) and estimated (VTRE) forms. Through machine learning analyses, we demonstrate that VTRE robustly approximates VTRG, accurately tracks age and group related atrophic changes, and exhibits superior discriminative performance in distinguishing EMCI from LMCI when compared to alternative composite features. Moreover, explainable AI techniques confirm VTRE’s dominant and interpretable role in model predictions. Together, these findings establish VTRE as a sensitive, scalable, and clinically interpretable biomarker for early neurodegeneration, offering significant translational value for Alzheimer's disease diagnostics and research.