FlexiBrain: Resolution-Agnostic Voxel-Level Encoding for Native fMRI

2026-06-09

AI Summary

This study addresses the challenges posed by multi-source native fMRI data, which exhibit substantial variations in spatial and temporal resolution, necessitating complex preprocessing pipelines that often incur anatomical information loss and high computational costs. To overcome these limitations, the authors propose a resolution-agnostic, voxel-level encoding framework built upon a Mamba-JEPA backbone. This approach incorporates a physics-informed, dynamically resampled patch strategy defined in physical units, enabling direct modeling of native 4D fMRI signals without requiring spatial normalization or destructive preprocessing. As the first plug-and-play method capable of directly handling multi-resolution native fMRI data, the framework significantly reduces preprocessing overhead while preserving individual anatomical characteristics. It consistently outperforms state-of-the-art methods across five downstream neuroscience tasks, achieving performance gains of up to 12 percentage points without relying on external data augmentation.

Abstract

The success of large-scale deep learning models in neuroscience is fundamentally constrained by severe data heterogeneity. Native fMRI data aggregated from diverse sources exhibit substantial variation in both spatial and temporal resolutions. Consequently, most existing frameworks rely on lengthy, rigid preprocessing pipelines that enforce uniformity across datasets. This practice introduces two critical limitations: (1) potential degradation of subject-specific anatomical information; (2) significant computational overhead, often requiring hours of processing per subject. Here, we propose FlexiBrain, a resolution-agnostic voxel-level encoding framework for native fMRI based on Mamba-JEPA. FlexiBrain defines patch sizes in real-world physical units and employs a dynamic patch resizing, thereby bypassing destructive spatial standardization while enabling direct ingestion of data in native space. We instantiate the framework using an efficient Mamba-JEPA backbone to model high-dimensional 4D fMRI signals. Across five diverse downstream neuroscience tasks, FlexiBrain consistently outperforms recent state-of-the-art methods, achieving gains of up to 12 percentage points without external data augmentation. Importantly, FlexiBrain functions as a seamless plug-in module, substantially reducing preprocessing costs and accelerating the development of robust voxel-level fMRI foundation models. Code is available at https://github.com/OneMore1/FlexiBrain.

Problem

• fMRI
• data heterogeneity
• spatial resolution
• temporal resolution
• preprocessing

Innovation

• resolution-agnostic
• voxel-level encoding
• native fMRI
• Mamba-JEPA
• dynamic patch resizing