Diagnostic Accuracy of MRI-Based Artificial Intelligence Models for Distinguishing Alzheimer's Disease and Mild Cognitive Impairment From Cognitively Normal Controls: A Systematic Review and Meta-Analysis

Description

Allaham reviewed how well AI models reading structural brain MRI can split Alzheimer's disease and mild cognitive impairment from cognitively normal controls. Searching five databases to February 2026 yielded seven qualifying studies. Most were deep learning systems trained on the ADNI research cohorts.

For Alzheimer's versus normal controls across five studies, pooled sensitivity reached 0.96 and pooled specificity 0.95 in the pooled analysis. That shows top discrimination in the meta-analysis. The two mild impairment head-to-heads told a murkier story. Sensitivity stayed high, between 0.91 at the low end and 0.93 at the high end. But specificity swung widely, spanning 0.54 at the bottom to 0.98 at the top between the studies. That spread makes pooled figures for early impairment hard to read, in the author's view. The review concludes MRI-based AI performs strongly for full Alzheimer's. It needs larger studies, tested outside their home data in diverse groups, for early stages.

Catalog record

Source: https://europepmc.org/articles/PMC13444374