Fully Automated Hippocampus Segmentation using T2-informed Deep Convolutional Neural Networks

Maximilian Sackl, Christian Tinauer, Martin Urschler, Christian Enzinger, Rudolf Stollberger, Stefan Ropele

Abstract: Highlights•High-resolution T2-weighted MR data provides a better basis for manual labeling.•T2-based ground truth labels improve hippocampus segmentation on T1-weighted scans.•Proposed hippocampus segmentation is based on stacked deep hourglass CNNs.•Our deep SHG-CNN outperformes FreeSurfer, ASHS, HippoDeep, and HippMapp3r.