Reconstruction of Cortical Surfaces with Spherical Topology from Infant Brain MRI via Recurrent Deformation Learning

Xiaoyang Chen, Junjie Zhao, Siyuan Liu, Sahar Ahmad, Pew-Thian Yap

Primary Area: applications to neuroscience & cognitive science

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Cortical surface reconstruction, infant brain, diffeomorphic

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Cortical surface reconstruction (CSR) from MRI is key to investigating brain structure and function. While recent deep learning approaches have significantly improved the speed of CSR, a substantial amount of runtime is still needed to map the cortex to a topologically-correct spherical manifold to facilitate downstream geometric analyses. Moreover, this mapping is possible only if the topology of the surface mesh is homotopic to a sphere. Here, we present a method for simultaneous CSR and spherical mapping efficiently within seconds. Our approach seamlessly connects two sub-networks for white and pial surface generation. Residual diffeomorphic deformations are learned iteratively to gradually warp a spherical template mesh to the white and pial surfaces while preserving mesh topology and uniformity. The one-to-one vertex correspondence between the template sphere and the cortical surfaces allows easy and direct mapping of geometric features like convexity and curvature to the sphere for visualization and downstream processing. We demonstrate the efficacy of our approach on infant brain MRI, which poses significant challenges to CSR due to tissue contrast changes associated with rapid brain development during the first postnatal year. Performance evaluation based on a dataset of infants from 0 to 12 months demonstrates that our method substantially enhances mesh regularity and reduces geometric errors, outperforming state-of-the-art deep learning approaches, all while maintaining high computational efficiency.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6817