SimCortex: Collision-free Simultaneous Cortical Surfaces Reconstruction

TL;DR

SimCortex is introduced, a deep learning framework that simultaneously reconstructs all brain surfaces (left/right white-matter and pial) from T1-weighted MRI volumes while preserving topological properties while maintaining state-of-the-art geometric accuracy.

Abstract

Accurate cortical surface reconstruction from magnetic resonance imaging (MRI) data is crucial for reliable neuroanatomical analyses. Current methods have to contend with complex cortical geometries, strict topological requirements, and often produce surfaces with overlaps, self-intersections, and topological defects. To overcome these shortcomings, we introduce SimCortex, a deep learning framework that simultaneously reconstructs all brain surfaces (left/right white-matter and pial) from T1-weighted(T1w) MRI volumes while preserving topological properties. Our method first segments the T1w image into a nine-class tissue label map. From these segmentations, we generate subject-specific, collision-free initial surface meshes. These surfaces serve as precise initializations for subsequent multiscale diffeomorphic deformations. Employing stationary velocity fields (SVFs) integrated via scaling-and-squaring, our approach ensures smooth, topology-preserving transformations with significantly reduced surface collisions and self-intersections. Evaluations on standard datasets demonstrate that SimCortex dramatically reduces surface overlaps and self-intersections, surpassing current methods while maintaining state-of-the-art geometric accuracy.

Figures (6)

• Figure 1: Overview of the SimCortex pipeline.
• Figure 2: Workflow for segmentation-driven cortical surface initialization. From MNI152-aligned T1w MRI, a 3D U-Net segments tissues, masks are converted to smoothed SDFs, topologically corrected, and extracted as collision-free meshes for deformation.
• Figure 3: Multiscale deformation network. A 3D U-Net predicts SVFs from MNI152-aligned T1w MRI, deforming initial surfaces into precise, collision-free cortical surfaces.
• Figure 4: Violin plots illustrating the distributions of Hausdorff Distance (HD) and Average Symmetric Surface Distance (ASSD) for pial and white matter surfaces, comparing SimCortex, CFPP, and V2C models. The plots clearly demonstrate the variance and central tendency for each metric across methods, highlighting SimCortex's competitive geometric accuracy alongside reduced metric dispersion, particularly on pial surfaces.
• Figure 5: Cortical surface reconstructions: SimCortex (yellow), CFPP (blue), V2C (red), and ground truth (green). (a) Coronal view with MRI overlay. (b) RH pial surfaces detail. (c) LH white matter surfaces detail.