Reproduce “MRExtrap: Longitudinal Aging of Brain MRIs using Linear Modeling in Latent Space” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jaivardhan Kapoor, Jakob H. Macke, Christian F. Baumgartner
MRExtrap forecasts MRIs by compressing a 3D scan into a latent code, advancing it by an estimated age-related rate, and decoding it. A single scan can use a population-average rate. On 178 ADNI subjects, its hippocampus-volume MAE was 0.020 ± 0.001 % TBV, versus 0.060 ± 0.003 % TBV for DANINet, whose scores came from prior work; ventricle error was higher.
Key insights
- Autoencoders trained on brain MRIs tend to create latent spaces where age-related changes, and thus aging trajectories, appear approximately linear.
- A direct linear relationship between age and latent representations (z(a) = beta * a + z0) is supported by empirical observation and neuroimaging literature on scalar brain volume changes.
- The derived subject-specific progression rate (beta) itself is a clinically meaningful signal, correlating with disease status and age, consistent with known atrophy patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.