Learning transferrable and interpretable representation for brain network

Yifan Yang, Yutong Mao, Xiao Liu

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.

Keywords: Self-Supervised Learning, Masked Autoencoding, Characterizing representations, Neuroscience

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: The human brain is a complex, dynamic network, which is commonly studied using functional magnetic resonance imaging (fMRI) and modeled as network of Regions of interest (ROIs) for understanding various brain functions. Recent studies predominantly utilize Graph Neural Networks (GNNs) to learn the brain network representation based on the functional connectivity (FC) profile, typically falling into two main categories. The Fixed-FC approaches, utilize the FC profile which represents the linear temporal relation within the brain network, is limited by failing to capture the informative temporal dynamics of brain activity. On the other hand, the Dynamic-FC approaches, modeling the evolving FC profile over time, often exhibit less satisfactory performance due to challenges in handling the inherent noisy nature of fMRI data. In this study, to address these challenges, we propose Brain Masked Auto-Encoder (BrainMAE) for learning representations directly from fMRI time-series data. Our approach incorporates two essential components—an embedding-based graph attention mechanism and a self-supervised masked autoencoding framework. These components empower our model to capture the rich temporal dynamics of brain activity while maintaining resilience to the inherent noise in fMRI data. Our experiments demonstrate that BrainMAE consistently outperforms several established baseline models by a significant margin in three distinct downstream tasks. Finally, leveraging the model's inherent interpretability, our analysis of model-generated representations reveals intriguing findings that resonate with ongoing research in the field of neuroscience.

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: 8443