AI Connectome Predicts Alzheimer’s Risk Years Before Symptoms

Overview

• The study, published Nov. 4, 2026, introduces the Alzheimer’s Resilience-Vulnerability Connectome (ARC), a machine-learning score derived from resting-state fMRI that ranks brain patterns from SuperAger-like resilience to Alzheimer’s-like vulnerability.
• Researchers trained ARC on extreme groups — 180 SuperAgers (people 75+ with unusually preserved memory) and 110 patients with Alzheimer’s — and validated the model across independent cohorts.
• ARC distinguished SuperAgers from people with Alzheimer’s with about 85% accuracy when tested on holdout data.
• When applied to more than 1,300 cognitively unimpaired adults aged 55–70, baseline ARC scores predicted faster cognitive decline, greater amyloid-β accumulation, and higher conversion rates to mild cognitive impairment or Alzheimer’s over follow-up periods up to 12 years, and the score added prognostic value beyond standard biomarkers and genetics.
• Investigators stress that ARC is a research biomarker not a diagnosis, and they are planning trials to test whether targeted interventions such as cognitive training or brain stimulation can shift ARC patterns toward resilience; the work used multi-site data (OASIS, Harvard Aging Brain Study, ADNI) and received NIH and Alzheimer’s-focused funding.