Unified Brain Surface and Volume Registration

TL;DR

NeurAlign presents a unified, learning-based framework for joint cortical and subcortical brain registration by coupling volumetric and spherical registrations through an intermediate $S^2$ domain. The method uses a 3D volumetric network and a 2D spherical network, connected by a cortical-consistency loss that enforces alignment coherence between the volume and the cortex, enabling a single forward pass without requiring cortical meshes or segmentations at inference. Empirically, NeurAlign achieves substantial improvements in cortical Dice scores (up to about 7 Dice points) while preserving subcortical accuracy and delivering dramatically faster inferences than classic methods like CVS, across multiple datasets and disease populations. The approach advances practical whole-brain registration for large-scale neuroimaging studies, with potential extensions to multimodal data and broader cortical topologies.

Abstract

Accurate registration of brain MRI scans is fundamental for cross-subject analysis in neuroscientific studies. This involves aligning both the cortical surface of the brain and the interior volume. Traditional methods treat volumetric and surface-based registration separately, which often leads to inconsistencies that limit downstream analyses. We propose a deep learning framework, NeurAlign, that registers $3$D brain MRI images by jointly aligning both cortical and subcortical regions through a unified volume-and-surface-based representation. Our approach leverages an intermediate spherical coordinate space to bridge anatomical surface topology with volumetric anatomy, enabling consistent and anatomically accurate alignment. By integrating spherical registration into the learning, our method ensures geometric coherence between volume and surface domains. In a series of experiments on both in-domain and out-of-domain datasets, our method consistently outperforms both classical and machine learning-based registration methods -- improving the Dice score by up to 7 points while maintaining regular deformation fields. Additionally, it is orders of magnitude faster than the standard method for this task, and is simpler to use because it requires no additional inputs beyond an MRI scan. With its superior accuracy, fast inference, and ease of use, NeurAlign sets a new standard for joint cortical and subcortical registration.

Figures (4)

• Figure 1: Unified cortical and volumetric registration framework. We employ two CNNs, a volumetric one to perform 3D image alignment, and a 2D spherical one to align the cortical surfaces in the spherical domain. In training, the models require the MRI images $I$, the cortical surface meshes, and the inflated spheres. The resultant deformation fields of both models are used in a loss term ($\mathcal{L}_{cons}$) that encourages cortical registration to be consistent in the spherical and volumetric pathways. The models are trained jointly to obtain accurate and consistent cortical and subcortical alignments. At inference time, the 3D model does not require meshes nor spheres to perform registration.
• Figure 2: Violin plot of distributions of Dice score across all subjects in each dataset. Our method (blue) consistently achieves the highest mean Dice for cortical structures, with mass centered near the mean. We preserve subcortical performance while delivering much stronger cortical alignment.
• Figure 3: Example registrations for three image pairs. Columns show the moving and fixed images, followed by warped images produced by uGradICON-seg, SynthMorph-wm, FireANTs, CVS, and our method. Corresponding deformation fields are visualized on the right. Our method (NeurAlign) yields accurate alignments with smooth and regular warp fields.
• Figure A.1: Example registrations for one image while varying $\kappa$. Columns show the moving and fixed images, followed by warped images and corresponding deformation fields are visualized on the right. Deformation fields become more irregular with increasing $\kappa$, however all demonstrated values produce accurate and smooth alignments.