3D Medical Point Transformer: Introducing Convolution to Attention Networks for Medical Point Cloud Analysis

General point clouds have been increasingly investigated for different tasks,and recently Transformer-based networks are proposed for point cloud analysis.However, there are barely related works for medical point clouds, which areimportant for disease detection and treatment. In this work, we propose anattention-based model specifically for medical point clouds, namely 3D medicalpoint Transformer (3DMedPT), to examine the complex biological structures. Byaugmenting contextual information and summarizing local responses at query, ourattention module can capture both local context and global content featureinteractions. However, the insufficient training samples of medical data maylead to poor feature learning, so we apply position embeddings to learnaccurate local geometry and Multi-Graph Reasoning (MGR) to examine globalknowledge propagation over channel graphs to enrich feature representations.Experiments conducted on IntrA dataset proves the superiority of 3DMedPT, wherewe achieve the best classification and segmentation results. Furthermore, thepromising generalization ability of our method is validated on general 3D pointcloud benchmarks: ModelNet40 and ShapeNetPart. Code is released.