Abstract: According to the characteristic of mobile C-arm X-ray imaging in image-guided minimally invasive spine surgery,an automatic lumbar vertebrae recognition method is proposed,which based on hierarchical recurrent neural network.Its purpose is to identify lumbar vertebrae automatically by learning the curvature features.First,in order to solve the problem of lumbar vertebrae texture overlapping in X-ray images,the curvature features of 3D lumbar vertebrae model,which are common to the 2D X-ray images,are taken as the input of the model.Second,in order to simulate the multi-view imaging of intraoperative C-arm,the bidirectional recurrent neural network is exploited to learn the correlation of lumbar curvature features at different imaging angles.Finally,in order to solve the problem of partial occlusion of the lumbar vertebrae in the pathological condition,a hierarchical recurrent neural network model is introduced.The natural context between human lumbar vertebrae is modeled by the layer-by-layer fusion architecture to improve the recognition rate in the pathological condition.The results of the verification on open source datasets and intraoperative mobile C-arm X-ray images show that the lumbar vertebrae recognition rate of the proposed method is superior to the other four methods in both normal and pathological conditions.Furthermore,due to the utilization of two-dimensional curvature features,the proposed method is more efficient in the training and testing phases,and more suitable for applications in intraoperative image-guided navigation.