TAD-Net: tooth axis detection network based on rotation transformation encoding

Yeying Fan, Qian Ma, Guangshun Wei, Zhiming Cui, Yuanfeng Zhou, Wenping Wang

Abstract: Highlights•We propose a novel learning-based method for automatic and accurate tooth axis detection, which is the first work in this field to greatly accelerate the workflow of digital orthodontics.•We convert the tooth axis prediction problem to a point-wise dense rotation transformation prediction task on 3D point cloud. Importantly, a feature confidence-aware attention mechanism is introduced to learn dynamic weights for point-wise features, which in return enhance the learning at reliable points.•The proposed method can simultaneously predict multiple tooth axes on different kinds of teeth. It has achieved the state-of-the-art performance on the benchmark validated by various experiments.