AI scan model sorts healthy older brains by Alzheimer’s risk

Clinical takeaway: Watch for prospective studies that test the wiring score against amyloid and tau markers, since that comparison determines whether it ever reaches the clinic. 

Clinicians see older patients every day whose memory scores look normal, and nothing at the visit separates those who will stay sharp from those heading toward mild cognitive impairment or Alzheimer’s disease. Beneath those scores, brain aging and early amyloid and tau pathology vary widely. 

SuperAgers, adults 75 and older whose memory rivals that of middle-aged adults, have mostly been studied against typical older adults, a group that mixes steady agers with people in early disease. This analysis tests a different comparison. Researchers scored resting-state brain connectivity by contrasting SuperAgers with patients who have Alzheimer’s, then asked whether that score could separate cognitively normal older adults who played no part in building it. The work extends evidence from the two ends of cognitive aging into the middle. 

The wiring score came from resting-state fMRI, a scan of brain activity taken while patients lie at rest, which measures how distant regions work in step. About one in four cognitively normal adults scored AD-like in each cohort. That group declined faster than the SuperAger-like group on clinical dementia ratings and executive function tests. The groups looked alike at baseline apart from age, with the AD-like group two to three years older at the median. In the replication cohort, the gap in decline held after separating baseline age from time in the study. AD-like adults also progressed to mild cognitive impairment or Alzheimer’s disease more often in both cohorts, though the replication gap was borderline. 

Baseline amyloid and tau did not differ between the groups. Over follow-up, amyloid built up faster in the AD-like group, most in the frontal and temporal cortex, while tau did not diverge. Among amyloid-positive adults in the replication cohort, the AD-like group still worsened faster on the clinical rating after adjusting for baseline amyloid, tau, and clinical score. 

The wiring score alone sorted adults who progressed from those who did not better than chance in replication. Adding it to amyloid and tau markers improved that sorting, yet it explained only 3 to 4 percentage points more variance in clinical decline. 

Using resting-state fMRI, researchers trained the model on 180 SuperAgers and 110 patients with Alzheimer’s disease from OASIS-3 and the Harvard Aging Brain Study, then tested it on 143 people in the Alzheimer’s Disease Neuroimaging Initiative. They applied it to 818 cognitively normal adults in the first two cohorts and 497 in the third, drawing on archived research scans, with the SuperAger memory criteria differing by cohort. 

The next test is whether the score holds with the standard SuperAger criteria, which this analysis adapted. The cohorts were largely white, so it also needs testing in more diverse groups. Longer follow-up in larger cohorts would show whether the wiring reflects resilience to pathology or resistance to its buildup, a distinction this design cannot make. 

The authors, who call the wiring score the Alzheimer’s Resilience-Vulnerability Connectome, or ARC, concluded: “These findings support the potential utility of ARC for early risk stratification among cognitively unimpaired older adults.” 

Source: Zhao K, et al. (2026 Nov 4) Sci Adv. A functional connectome identifies vulnerability to cognitive decline and Alzheimer’s disease progression