Vox2Cortex
Generating brain surfaces within seconds

Vox2Cortex: Fast Explicit Reconstruction of Cortical Surfaces from 3D MRI Scans with Geometric Deep Neural Networks

Vox2Cortex deforms a brain template to the inner and outer boundary surfaces of the cortex based on features extracted from the input MRI.

The obtained surfaces are geometrically accurate and topologically correct.

The novel curvature-weighted Chamfer Loss reweights the ratio between geometric accuracy and surface regularity locally based on ground truth curvature

See the paper and our CVPR 2022 poster and presentation slides here:

Watch our presentation at CVPR 2022 here:

Authors

Fabian Bongratz

fabi.bongratz@tum.de

PhD student

Anne-Marie Rickmann

a.rickmann@tum.de

PhD student

Sebastian Pölsterl

PostDoc

Christian Wachinger

Professor

AI-Med

Lab for AI in Medical Imaging