Brain MRI predicts conversion to mild cognitive impairment
Adding brain MRI to baseline traits predicted conversion from subjective decline to mild impairment with an AUC of 0.923 at five years.
medRxiv
In an observational study of 1,352 human participants with subjective cognitive decline, researchers evaluated machine learning models to forecast conversion to mild cognitive impairment. The team analyzed baseline data across four longitudinal cohorts, testing predictions at horizons of two, three, four, and five years.
Models incorporating base demographics, regional white matter hyperintensities, and gray matter volumes achieved the highest area under the curve at three years (0.900), four years (0.879), and five years (0.923). Adding cognitive test scores yielded the top performance only at the two-year horizon, reaching an AUC of 0.884. Brain imaging features provided larger performance gains over baseline features than cognitive scores at three to five years, with parietal white matter hyperintensity emerging as the most frequently selected MRI predictor.
Why it matters
Structural brain changes like white matter damage and regional atrophy often appear before overt clinical dementia. Mapping these features helps pinpoint when biological brain aging shifts from subjective complaints to measurable clinical decline.
Caveats
This work is a preprint that has not yet undergone peer review. The findings are based on observational cohorts and machine learning models that require prospective clinical validation.
The paper
Predicting Conversion from SCD to MCI: A Machine Learning Study
Douglas Mental Health University Institute
medRxiv · 8 Sep 2026 · Preprint, not peer-reviewed