A systematic review of multimodal brain age studies: Uncovering a divergence between model accuracy and utility
Summary
(reprinted from the article: “bigger picture”)
Establishing standardized biomarkers of brain aging would have immense utility for quantifying the risks of adverse life outcomes, which may lead to more precise and personalized interventions. An increasingly common approach for deriving such biomarkers is to train machine-learning models on a diverse set of neuroimaging features with the goal of accurately estimating a person’s biological age. Further, the difference between biological and chronological ages has been shown to be a useful metric that is sensitive to many phenotypes. However, no definitive guidelines exist to help researchers decide which set of neural properties would improve model accuracy and utility. Accordingly, we conducted a systematic review of multimodal brain age studies to identify which neuroimaging features provided the largest added value. We found that the multimodal models had an increased capacity to predict age, which did not translate to widespread improvement in clinical utility.
Citation
@article{jirsaraie2023systematic,
title={A systematic review of multimodal brain age studies: Uncovering a divergence between model accuracy and utility},
author={Jirsaraie, Robert J and Gorelik, Aaron J and Gatavins, Martins M and Engemann, Denis A and Bogdan, Ryan and Barch, Deanna M and Sotiras, Aristeidis},
journal={Patterns},
volume={4},
number={4},
year={2023},
publisher={Elsevier}
}