BrainSF: A Foundation Model for Whole Brain Functional Signals Forecasting

Li Yang, Hanqi Jiang, Xintao Hu, Tianming Liu, Tuo Zhang

Keywords: Foundation Model, Brain Functional Signal, Time Series Forecasting

Abstract: Foundational models hold significant potential for advancing brain function research, particularly with recent technological advancements enabling the capture of spatiotemporal dynamics of brain signals. However, existing methods are primarily limited to characterizing observed brain signals and cannot infer continuous future signals—an essential component for understanding the brain's causal structure and its role in various cognitive states. Current research leaves a substantial gap in forecasting whole-brain signal sequences. To address this, we propose a self-supervised model that embeds momentary whole-brain fMRI signals into vector representations and predicts continuous future signals. Our model is trained on a large-scale fMRI dataset, encompassing both resting-state and naturalistic stimuli conditions. Experimental results demonstrate that the model performs effectively in zero-shot forecasting of future whole-brain signals on unseen data and excels in downstream tasks such as task-based functional state decoding. To the best of our knowledge, this is the first approach to forecast and model whole-brain signals at such a large scale. The experimental results validate the feasibility of our method, offering new directions for theoretical research on brain signal time series and potential applications in diagnosing and treating brain disorders.

Primary Area: applications to neuroscience & cognitive science

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 34