Research And Implementation Of 3D Reconstruction System Based On Biplane X-Ray Of Spine

As one of the common diseases in modern society,spinal related diseases seriously affect patients’ normal life due to its high incidence and high mobility.In the clinical diagnosis and surgical intervention of spinal diseases,the commonly used two-dimensional image imaging method(Xray)can only provide planar information,but cannot capture effective three-dimensional spatial information of the spine,which is not conducive to the observation and judgment of the spinal morphology and structure.In order to deal with the limitations of current medical imaging technology,this study used deep learning models to predict 3D vertebral morphology based on coronal and sagittal images,and constructed a 3D reconstruction system based on dual-plane X-ray of the spine according to clinical requirements.The main work is as follows:In this study,we proposed a new method based on deep learning model to achieve automatic reconstruction of 3D vertebrae from orthogonal biplane X-ray.First,the Digitally Reconstructured Radiograph(DRR)was constructed using ITK on the CT data of the spine.Then,the 2D DRR of the spine was tailored and annotated using 3D center of mass information.Second,the centroid labeling information was used to integrate complementary information from different perspectives by selfattention mechanism and convert 2D features into 3D features.Finally,the full convolutional network model was used for supervised learning of 3D features to realize the reconstruction task from 2D image to 3D shape.Inaddition,we used Dice Loss and MSE Loss as optimization objectives,and realized 3D reconstruction of vertebrae morphology based on the full convolutional network model.In this study,1407 vertebraes(mainly thoracic and lumbar vertebrae)of 140 pieces of spinal CT data were studied,and Dice value and Hausdorff Distance were used to reflect the effect of 3D vertebrae morphological reconstruction.Through ablation experiments,the dice value on the test set and Hausdorff Distance of the comparison model TransVert reached 0.646 and 28.166 respectively,while the proposed method of the study achieved a dice value of 0.771 and a Hausdorff Distance of 14.711 on the test set,both of which were superior to the TransVert performance on the two evaluation indexes.Therefore,the proposed method of the study has better performance on the 3D reconstruction of the spine.Based on the above research and and the requirements of clinicians for the application of 3D reconstruction of spine,this study also designed a 3D reconstruction system based on the dual-plane X-ray of the spine.The system architecture and functional modules are designed in detail,and on this basis,all functional modules are designed in detail to realize the 3D reconstruction system.At present,the system is ready for operation in cooperative hospitals to further clarify the value of clinical application.