Predicting future cognitive impairment in preclinical Alzheimer's disease using multimodal imaging: a multisite
Braden Yang1, Tom Earnest1, Murat Bilgel2
1Mallinckrodt Institute of Radiology, Washington University School of Medicine in St. Louis, St. Louis, MO, USA 63110.
Medrxiv : the Preprint Server for Health Sciences
|November 24, 2025
Summary
Machine learning models can predict Alzheimer's disease (AD) progression in preclinical individuals. This stratification enhances clinical trial power for new AD therapies targeting early disease stages.
Area of Science:
- Neuroscience
- Medical Imaging
- Artificial Intelligence
Background:
- Predicting Alzheimer's disease (AD) dementia progression is crucial for clinical trials.
- Machine learning (ML) shows promise, but models for preclinical AD are lacking.
- Preclinical AD involves early amyloidosis without cognitive impairment.
Purpose of the Study:
- Develop and evaluate ML classifiers for predicting preclinical AD progression.
- Assess model generalizability across sites and radiotracers.
- Demonstrate ML's utility in enriching clinical trial cohorts.
Main Methods:
- Trained ML classifiers on imaging features from amyloid PET and MRI.
- Utilized data from seven sites and two radiotracers ([18F]-florbetapir, [11C]-Pittsburgh-compound-B).
- Employed leave-one-site-out and leave-one-tracer-out cross-validation for generalizability.
Main Results:
- Achieved out-of-sample AUC of 0.66+ across sites and 0.72+ across tracers.
- Demonstrated increased statistical power in a retroactive A4 trial analysis.
- ML stratification improved detection of amyloid accumulation differences.
Conclusions:
- ML models can effectively predict cognitive decline in preclinical AD.
- Patient stratification using ML enhances clinical trial efficiency.
- This approach supports the development of targeted AD therapies.
