Frequency-filtered attention for cross-species fMRI time series prediction under small-sample learning.

Authors

Chengcheng Du, Feifei Zhao, Yinqian Sun, Zeyang Yue, Ruoyu Wu, Jihang Wang, Qian Zhang, Yi Zeng

Published in

iScience. Volume 29. Issue 9. Pages 117173. Sep 18, 2026. Epub Aug 13, 2026.

Abstract

Functional magnetic resonance imaging (fMRI) time series exhibit long-term temporal dependencies and stable functional connectivity (FC) structures. However, most existing prediction models mainly focus on the temporal domain, making it difficult to jointly capture spectral characteristics and neurobiological priors. We propose a general fMRI sequence prediction model, the Frequency-Filtered Attention Transformer (FFAformer). It models low-frequency variations in the frequency domain to capture long-range dependencies and incorporates FC consistency constraints to preserve brain network structure. In addition, FFAformer introduces a trainable symmetric positive definite full-rank matrix into the attention mechanism to alleviate representation degradation under small-sample learning. The predicted fMRI time series preserve low-dimensional brain activity patterns and FC consistent with real data. Experiments on small-sample cross-species fMRI datasets (mice, macaques, and humans) demonstrate lower prediction errors, higher FC consistency, and robust cross-species generalization, supporting reliable fMRI sequence prediction.

PMID:
42633179
Bibliographic data and abstract were imported from PubMed on 23 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.