The Developing Human Connectome Project: A fast deep learning-based pipeline for neonatal cortical surface reconstruction

Qiang Ma, Kaili Liang, Liu Li, Saga Masui, Yourong Guo, Chiara Nosarti, Emma C. Robinson, Bernhard Kainz, Daniel Rueckert

Abstract: Highlights•A deep learning-based neuroimage processing pipeline is presented for neonatal cortical surface reconstruction in the Developing Human Connectome Project (dHCP).•A multiscale deformation network is introduced to learn diffeomorphic cortical surface reconstruction end-to-end from neonatal brain MRI.•The entire workflow of our pipeline only requires 24 s on a modern GPU.•Our pipeline achieves superior cortical surface quality while being nearly 1000 times faster than the previous dHCP structural pipeline.