Particle.news
Download on the App Store

WPI Machine-Learning Model Flags Alzheimer’s on MRI With Near-93% Accuracy

Researchers say the MRI-based approach could support earlier diagnosis pending validation in broader, more diverse cohorts.

Overview

  • Published in Neuroscience, the study analyzed 815 Alzheimer’s Disease Neuroimaging Initiative MRI scans from participants aged 69–84 across 95 brain regions.
  • The model achieved 92.87% accuracy in distinguishing Alzheimer’s disease from normal cognition and mild cognitive impairment.
  • Volume loss in the hippocampus, amygdala, and entorhinal cortex emerged as the strongest predictors across groups.
  • Both sexes in the youngest cohort studied (ages 69–76) showed right hippocampal volume loss, pointing to a potential early marker.
  • Sex-specific patterns were reported, with females showing loss in the left middle temporal cortex and males in the right entorhinal cortex, and the team plans follow-up tests using deep learning and assessing factors such as diabetes.