Mind the Gap: Does Brain Age Improve Alzheimer's Disease Prediction?

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Brain age models, trained on large datasets, serve as general brain health markers.
• Large language models demonstrate success when finetuned for specific applications.
• Direct models often outperform surrogate models in predicting specific outcomes.

Purpose of the Study:

• To compare the predictive performance of large-scale brain age models versus direct models for specific health outcomes.
• To investigate the efficacy of brain age models in the context of Alzheimer's Disease (AD) dementia prediction.

Main Methods:

• Utilized anatomical T1 scans from 1,848 participants across three continents.
• Compared predictive accuracy of non-finetuned brain age models against direct models.
• Evaluated finetuned brain age models against direct models, considering pretraining data scale differences.

Main Results:

• Direct models demonstrated superior performance compared to non-finetuned brain age models in predicting AD dementia.
• Finetuned brain age models achieved performance comparable to direct models.
• Despite extensive pretraining data (N=53,542), finetuned brain age models did not outperform direct models trained on smaller datasets (N=50).

Conclusions:

• Brain age remains a valuable indicator of general brain health.
• For predicting specific neurological outcomes like AD dementia, direct models show significant value.
• Small-scale, targeted approaches for extracting specific brain health markers remain crucial, even with the advent of large-scale brain age models.