Atlas-free Brain Network Transformer

TL;DR

The paper tackles limitations of fixed brain atlases in brain-network analysis by introducing atlas-free Brain Network Transformer (atlas-free BNT). It derives individualized parcellations from rs-fMRI, builds ROI-to-voxel connectivity in a standardized voxel space, and applies a block-wise transformer to produce comparable subject embeddings for group-level tasks. Across sex classification and brain-connectome age prediction on ABCD and EHBS datasets, atlas-free BNT outperforms state-of-the-art atlas-based methods, showing improved precision, robustness, and generalizability. This approach holds promise for personalized neuroimaging biomarkers and clinical diagnostic tools in precision medicine.

Abstract

Current atlas-based approaches to brain network analysis rely heavily on standardized anatomical or connectivity-driven brain atlases. However, these fixed atlases often introduce significant limitations, such as spatial misalignment across individuals, functional heterogeneity within predefined regions, and atlas-selection biases, collectively undermining the reliability and interpretability of the derived brain networks. To address these challenges, we propose a novel atlas-free brain network transformer (atlas-free BNT) that leverages individualized brain parcellations derived directly from subject-specific resting-state fMRI data. Our approach computes ROI-to-voxel connectivity features in a standardized voxel-based feature space, which are subsequently processed using the BNT architecture to produce comparable subject-level embeddings. Experimental evaluations on sex classification and brain-connectome age prediction tasks demonstrate that our atlas-free BNT consistently outperforms state-of-the-art atlas-based methods, including elastic net, BrainGNN, Graphormer and the original BNT. Our atlas-free approach significantly improves the precision, robustness, and generalizability of brain network analyses. This advancement holds great potential to enhance neuroimaging biomarkers and clinical diagnostic tools for personalized precision medicine.

Figures (9)

• Figure 1: The Shen-368 brain atlas applied to standardized T1-weighted images in MNI space from two different subjects. Taking ROI $\#$52 as an example, both anatomical misalignment and functional heterogeneity arise due to inter-individual variability. (a) A zoomed-in view highlights the anatomical misalignment of the ROI between subjects. (b) Within this ROI, intra-correlations between the voxel-wise and mean BOLD time series are low, indicating substantial internal functional heterogeneity.
• Figure 2: Individualized brain parcellation: the ROIs parcellated on a subject's brain using the agglomerative and spectral clustering methods.
• Figure 3: Each channel ${\bm{F}}_j$ of the brain map ${\bm{F}}=\{{\bm{F}}_1,\cdots,{\bm{F}}_D\}$ encodes the functional connectivity between the ROIs and a specific voxel ${\bm{v}}_j$.
• Figure 4: The proposed atlas-free brain network transformer.
• Figure 5: The atlas-based methods use four atlases to define ROIs in the brain: the AAL, Craddock-400, Shen-368 and HCP-360 atlases.