AI Brain Age Score Predicts Dementia Risk
A study from the Alzheimer Center in Amsterdam, published in Neurology, shows AI-estimated brain age predicts dementia better than standard scan review in people with mild memory decline.

A team of researchers at the Alzheimer Center at Amsterdam University Medical Center (Amsterdam UMC), led by Stefan de Vries, has shown that an artificial intelligence algorithm that estimates brain age from an MRI scan predicts the transition to dementia better than standard visual reading of the scan, in patients who report memory complaints but show no measurable deficit on cognitive tests. The study appeared in the journal Neurology, DOI 10.1212/wnl.0000000000218418. According to the announcement and the report on Medicalxpress, the study was published on August 19, though another source lists the paper as available online as early as August 10.
Memory clinics today typically assess brain atrophy through visual scoring of MRI scans, using scales such as medial temporal lobe atrophy or global atrophy ratings. Those scales are coarse and miss subtle changes in the early stages of memory decline. The measure the Dutch researchers tested, known as brain-PAD, has already been linked in earlier studies of general populations and memory clinic patients to dementia risk, independent of other known risk factors. The new research set out to check whether that measure adds anything beyond what doctors already see on a standard scan review.
What does the brain-PAD measure show?
Brain-PAD stands for brain-predicted age difference, the gap between the brain’s age as estimated from a scan and the patient’s actual age. Researchers used two pretrained algorithms, brainageR and cNeuro, which calculate how old a brain looks based on a structural MRI scan. When the brain appears older than the patient’s chronological age, brain-PAD is positive, and according to the study’s findings, each additional year of that gap carried a 6 percent higher risk of dementia, a hazard ratio of 1.06. When the researchers added standard visual scan ratings to the model, the extra risk per year of difference remained at 4 percent, meaning the measure carries information independent of what a doctor sees with the naked eye.
The study included 799 patients from memory clinics: 412 with subjective cognitive decline and 387 with mild cognitive impairment. The data came from the Amsterdam Dementia Cohort and the SCIENCe project, covering the period from 2002 to 2024. The average age of participants was 63.3, and women made up 39.4 percent of the sample. Patients were followed for an average of 3.2 years, during which 235 of them, or 29.4 percent, progressed to dementia. Alongside brain-PAD, researchers had standard visual scales for each patient, including medial temporal lobe atrophy scores, global atrophy, the Koedam scale, the Fazekas scale, and amyloid status.
The difference between the two patient groups was pronounced. In people with subjective cognitive decline, meaning those who report memory complaints without a measurable deficit on tests, each additional year of brain age raised the relative risk of dementia by 9 percent, a hazard ratio of 1.09. From that data, the researchers drew a concrete threshold: patients whose brain on the scan appeared at least 2.6 years younger than their actual age had a 91 to 99 percent chance of not developing dementia over the following two to ten years. The picture was different for patients with mild cognitive impairment, where brain-PAD added little beyond what visual scan ratings already showed, since atrophy in that group is typically already visible to the eye.
What the study does not prove
The authors note that the data are retrospective and come from a single center, the Alzheimer Center in Amsterdam, so the finding still needs to be confirmed in independent cohorts before the tool moves into clinical use. An association between an older-looking brain and later dementia is not proof of causation. Brain-PAD is a risk indicator at the level of a patient group, not a diagnostic test for an individual. The result also does not mean the algorithm can replace a doctor’s examination on its own, researchers see it as an addition to visual scales, not a replacement for them. For patients with mild cognitive impairment, the gain from measuring brain age is, based on this data, small, so for now the clearest practical benefit is limited to the earliest stage, when subjective cognitive decline is the only sign.



