Vision Multi-Hop Integration Graph Convolutional Network for X-Ray Landmark Detection

X-ray Imaging landmark detection aims to automatically identify the locations of predefined points and it can be applied in medical-aided diagnosis. Despite graph convolution demonstrating encouraging results, the scale of the adjacency matrix is too single to fuse information between landmarks. In this work, we present a new multi-hop graph learning approach for accurately detecting medical landmarks, called MHGN. The proposed method builds a graph via the pre-estimated coordinates by a heatmap generator and fuses multi-hop adjacency matrix to mine the relationship between different landmarks. The proposed multi-hop graph convolutional network is trained in cascade with convolutional neural network and graph convolutional network together. Experiments are implemented on three X-ray medical datasets. The results compared with the state-of-the-art methods indicate the better performance in accuracy.