作者: Neslisah Torosdagli , Denise K. Liberton , Payal Verma , Murat Sincan , Janice S. Lee
关键词:
摘要: In this paper, we propose a novel deep learning framework for anatomy segmentation and automatic landmarking. Specifically, focus on the challenging problem of mandible from cone-beam computed tomography (CBCT) scans identification 9 anatomical landmarks geodesic space. The overall approach employs three inter-related steps. first step, neural network architecture with carefully designed regularization, hyper-parameters to perform image without need data augmentation complex post-processing refinement. second formulate landmark localization directly space sparsely-spaced landmarks. third utilize long short-term memory identify closely-spaced landmarks, which is rather difficult obtain using other standard networks. proposed fully automated method showed superior efficacy compared state-of-the-art landmarking approaches in craniofacial anomalies diseased states. We used very CBCT set 50 patients high-degree craniomaxillofacial variability that realistic clinical practice. qualitative visual inspection was conducted distinct 250 high variability. have also shown performance an independent MICCAI Head-Neck Challenge (2015).