End-to-end Cortical Surface Reconstruction from Clinical Magnetic Resonance Images

TL;DR

The paper tackles end-to-end cortical surface reconstruction from heterogeneous clinical MRIs, where conventional pipelines require consistent high-resolution scans. It introduces an explicit-surface reconstruction pipeline that deforms a template mesh to obtain white-matter and pial surfaces, trained entirely on synthetic, domain-randomized data to generalize across contrasts and resolutions while preserving topology. Compared with recon-all-clinical (RAC), the method reduces cortical thickness error by about 50% (0.50 mm to 0.24 mm) and better recovers aging-related thinning on high-resolution scans, with GPU runtimes around 1 s and CPU runtimes in minutes, enabling large-scale retrospective studies. The approach is publicly available and paves the way for a foundation-model-style surface pipeline that can operate across diverse clinical datasets and imaging protocols.

Abstract

Surface-based cortical analysis is valuable for a variety of neuroimaging tasks, such as spatial normalization, parcellation, and gray matter (GM) thickness estimation. However, most tools for estimating cortical surfaces work exclusively on scans with at least 1 mm isotropic resolution and are tuned to a specific magnetic resonance (MR) contrast, often T1-weighted (T1w). This precludes application using most clinical MR scans, which are very heterogeneous in terms of contrast and resolution. Here, we use synthetic domain-randomized data to train the first neural network for explicit estimation of cortical surfaces from scans of any contrast and resolution, without retraining. Our method deforms a template mesh to the white matter (WM) surface, which guarantees topological correctness. This mesh is further deformed to estimate the GM surface. We compare our method to recon-all-clinical (RAC), an implicit surface reconstruction method which is currently the only other tool capable of processing heterogeneous clinical MR scans, on ADNI and a large clinical dataset (n=1,332). We show a approximately 50 % reduction in cortical thickness error (from 0.50 to 0.24 mm) with respect to RAC and better recovery of the aging-related cortical thinning patterns detected by FreeSurfer on high-resolution T1w scans. Our method enables fast and accurate surface reconstruction of clinical scans, allowing studies (1) with sample sizes far beyond what is feasible in a research setting, and (2) of clinical populations that are difficult to enroll in research studies. The code is publicly available at https://github.com/simnibs/brainnet.

Figures (4)

• Figure 1: Overview of the synthetic data generation and training approach.
• Figure 2: Cortical surface reconstruction on $0.85 \times 0.85 \times 5 \; \text{mm}^3$ (recon-all reconstruction based on $1 \times 1 \times 1 \; \text{mm}^3$ for reference). Green arrows highlight regions of poor reconstruction by . Note that does not circumvent the hippocampus (blue arrows); this region, together with the medial wall, is masked out of the evaluation.
• Figure 3: Box plots for absolute value of mean cortical thickness error on the axial scans from GO/2 (top) and the clinical dataset (bottom).
• Figure 4: Cortical thickness variation with age on GO/2 (top) and the clinical dataset (bottom). Solid (dashed) lines show the age trend using (clinical) scans estimated using a quadratic fit. Note the different ranges on the $x$ axis.