Region-Aware Reconstruction Strategy for Pre-training fMRI Foundation Model

2025-11-01

AI Summary

Conventional self-supervised pretraining of resting-state fMRI suffers from semantically arbitrary masking strategies—typically random voxel or patch masking—that lack neuroanatomical grounding and hinder interpretability. Method: We propose an anatomically informed, region-aware reconstruction strategy: leveraging the AAL3 brain atlas to guide ROI-level masking—replacing random masking—and performing structure-guided self-supervised reconstruction over full 4D fMRI volumes. Contribution/Results: This approach substantially enhances model representation fidelity and semantic consistency for functionally relevant regions—particularly the limbic system and cerebellum. On ADHD-200, downstream classification accuracy improves by 4.23% over baseline; attribution analysis confirms that the anatomical guidance effectively focuses on pathology-sensitive regions. To our knowledge, this is the first work to deeply integrate neuroanatomical priors into the masking design of fMRI foundation model pretraining—simultaneously advancing task performance and neural interpretability—thereby establishing a novel paradigm for biomarker modeling in neuropsychiatric disorders.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Other Foundations of Computer Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

The emergence of foundation models in neuroimaging is driven by the increasing availability of large-scale and heterogeneous brain imaging datasets. Recent advances in self-supervised learning, particularly reconstruction-based objectives, have demonstrated strong potential for pretraining models that generalize effectively across diverse downstream functional MRI (fMRI) tasks. In this study, we explore region-aware reconstruction strategies for a foundation model in resting-state fMRI, moving beyond approaches that rely on random region masking. Specifically, we introduce an ROI-guided masking strategy using the Automated Anatomical Labelling Atlas (AAL3), applied directly to full 4D fMRI volumes to selectively mask semantically coherent brain regions during self-supervised pretraining. Using the ADHD-200 dataset comprising 973 subjects with resting-state fMRI scans, we show that our method achieves a 4.23% improvement in classification accuracy for distinguishing healthy controls from individuals diagnosed with ADHD, compared to conventional random masking. Region-level attribution analysis reveals that brain volumes within the limbic region and cerebellum contribute most significantly to reconstruction fidelity and model representation. Our results demonstrate that masking anatomical regions during model pretraining not only enhances interpretability but also yields more robust and discriminative representations. In future work, we plan to extend this approach by evaluating it on additional neuroimaging datasets, and developing new loss functions explicitly derived from region-aware reconstruction objectives. These directions aim to further improve the robustness and interpretability of foundation models for functional neuroimaging.

Problem

• Develops region-aware fMRI reconstruction using anatomical atlas masking
• Improves ADHD classification accuracy through selective brain region masking
• Enhances model interpretability by analyzing limbic and cerebellar contributions

Innovation

• ROI-guided masking strategy using AAL3 atlas
• Selectively masking semantically coherent brain regions
• Anatomical region masking enhances interpretability and robustness