A deep learning pipeline for lifespan cortical surface reconstruction, spherical mapping, and anatomical correspondence

Motivation: Accurate lifespan cortical surface reconstruction (CSR) is important for understanding brain anatomy and longitudinal studies. Goal(s): We present a deep learning pipeline that simultaneously reconstructs cortical surfaces, performs spherical mapping, and establishes anatomical correspondences across subjects. Approach: Our model learns multi-scale velocity fields to iteratively deform a spherical template to match the inner or outer cortical surfaces, with a constraint that registers subjects to a reference surface for one-to-one vertex correspondence. Results: Evaluation on a dataset spanning ages from 1 week to 90 years demonstrates that our model accurately predicts surfaces with inter-subject correspondence. Impact: Our end-to-end pipeline offers a fast and accurate method for generating cortical surfaces, making it an efficient tool for surface-based analysis of cortical morphology.