SuD-CoTAN: Sulcal Depth-guided Anatomically Consistent Fetal Cortical Surface Reconstruction

Irina Grigorescu, Jiaxin Xiao, Yourong Guo, Vanessa Kyriakopoulou, Alena Uus, Vyacheslav Karolis, Kaili Liang, Mohamed A. Suliman, Qiang Ma, Daniel Rueckert, Bernhard Kainz, A David Edwards, Jo Hajnal, Mary Rutherford, Maria Deprez, Emma C. Robinson

Keywords: fetal MRI, cortical surface reconstruction, deep learning

TL;DR: The paper introduces SuD-CoTAN, a sulcal-depth-guided framework that fits anatomically and topologically consistent cortical meshes directly to fetal T2-weighted MRI and performs alignment to age-matched templates in one single step.

Abstract: Accurate and anatomically consistent fetal cortical surface reconstruction is essential for studying early brain development, yet existing methods often lack reliable vertex-wise correspondence and fail to harmonise their outputs across heterogeneous MRI datasets. We introduce Sulcal Depth-guided CoTAN (SuD-CoTAN), a learning-based framework that fits anatomically and topologically consistent cortical meshes directly to T2-weighted MRI and performs alignment to age-matched templates in one single step. All models are trained exclusively on normative samples from the developing Human Connectome Project (dHCP) and evaluated within-sample and on a different acquisition protocol. Results show that SuD-CoTAN generalises to new datasets in ways that harmonise global morphometric properties by better capturing the surface geometry of individual cases; its template fitting is precise, delivering vertex-wise anatomical correspondences that result in sharp weekly averages of sulcal depth and curvature maps in template space. This supports direct vertex-wise Gaussian Process regression of neurodevelopmental trends without a need for any additional registration. Collectively, this whole pipeline runs in ∼3 seconds. This suggests that SuD-CoTAN offers promise as a screening tool for cortical malformations during fetal development.

Primary Subject Area: Image Registration

Secondary Subject Area: Geometric Deep Learning

Registration Requirement: Yes

Reproducibility: https://github.com/irinagrigorescu/SuDCoTAN

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 315