AI Insight
This study demonstrates that kernel ridge regression using functional connectivity matrices outperforms current brain foundation models in predicting individual phenotypes, and identifies that this advantage stems from miscalibrated eigenvalues in the connectivity data. The researchers developed a spectral filtering method to recalibrate these eigenvalues and used this approach to pretrain a compact encoder model on approximately 4,000 hours of fMRI data from 162 datasets, achieving performance comparable to larger published models while using significantly fewer parameters. The resulting model shows particular advantages for short scanning sessions, small cohorts, and brain fingerprinting tasks.
Why it matters
This work challenges the current trajectory of brain foundation models by showing that simpler methods with proper calibration can match their performance more efficiently. The findings could enable more practical clinical applications of brain imaging analysis, particularly in settings with limited data or shorter scanning protocols, and the released pretrained model provides an accessible tool for researchers working with fMRI data.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Self-supervised pretraining reshaped prediction in language and vision, and brain foundation models (BFMs) inherited its promise. Representations learned from large unlabelled corpora should capture individual functional dynamics and generalise across cohorts. However, kernel ridge regression (KRR) fitted on functional connectivity (FC) matrices still predicts individual phenotypes more accurately than any BFM we tested. In this paper, we show that KRR is weighted by the eigenvalues of the FC which are miscalibrated for phenotype prediction. We apply an efficient spectral filter to recalibrate the eigenvalues of each subject’s FC matrix, enabling the model to exploit more inter-individual variance. Across the 5 datasets, 11 parcellations and 6 prediction targets we tested, we match or exceed the KRR baseline. Based on this finding, we then pretrain a small encoder model on about 4,000 hours of fMRI from 162 open datasets, whereby we align the pairwise similarities between the embeddings of recording snippets with those between the recalibrated connectomes. Our model performs on par with the best of the 6 published BFMs we tested while having an order of magnitude fewer parameters. Our encoder performs better than FC on short scans and in smaller cohorts, especially in fingerprinting. We release the pretrained model weights, the code and the pretraining data, preprocessed and parcellated.