Deep orientated distance-transform network for geometric-aware centerline detection

Zheheng Jiang, Hossein Rahmani, Plamen Angelov, Ritesh Vyas, Huiyu Zhou, Sue Black, Bryan M. Williams

Abstract: Highlights•We present a novel multitask learning architecture for centerline detection.•We incorporate task uncertainty to simultaneously learn multiple losses.•A graph-cut based connectivity refinement approach is proposed.•Our proposed approach outperforms state-of-the-art methods on three datasets.