Research On Key Technologies Of Brain Virtual Surgery Navigation Based On Computer Vision

Brain diseases such as brain tumors are currently one of the main diseases leading to human death,seriously endangering human life and health.Craniocerebral surgery is an important means of treating these diseases.In traditional surgical treatment,the doctor judges the location of the lesion based on personal clinical experience,estimates the location of the cranial entry and the surgical path,but the traditional method relies on the doctor’s long-term experience accumulation,which has large errors and is difficult to meet the requirements of modern fine and minimally invasive surgery.In recent years,with the development of research fields such as medical imaging technology and image processing technology,surgical navigation systems are playing an increasingly important role in the field of minimally invasive surgery.However,the virtual surgical navigation technology has problems such as low efficiency of face point cloud registration and inaccurate positioning of surgical cranial drills.Therefore,this thesis focuses on the key issues in surgical navigation.The main research work and contributions are as follows:Aiming at the problem that the face 3D point cloud data under visible light is difficult to acquire accurately due to the influence of the surrounding environment and the acquisition error is large,this thesis proposes to use a deep learning-based face detection model to quickly locate the face position,thereby reducing the point cloud acquisition.Scope,according to the 68 key point models based on the face,use the RGB-D camera to extract the 3D point cloud data of the face area,and preprocess the point cloud data through statistical filtering and downsampling to make it meet the registration requirements,and Using 3D reconstruction technology,based on CT data,the face is reconstructed in 3D,and the 3D point cloud data of the face is extracted to facilitate the3 D point cloud registration of the face under multi-modality.Aiming at the problem of inaccurate registration of face 3D point cloud when the distance is large,this thesis proposes to use point cloud two-step registration method to register face 3D point cloud under multi-modality,that is,use Coarse registration based on sampling consistency based on Fast Point Feature Histograms(FPFH)and fine registration based on Anderson accelerated(AA)Iterative Closest Point(ICP)method solve the problem that registration efficiency is sensitive to distance requirements through coarse registration,and then through fine registration,thus improving the accuracy of registration speed and accuracy.Aiming at the difficult positioning and inaccurate tracking of cranial drills in cranial surgery,this thesis proposes a checkerboard-based cranial drill positioning and tracking technology,which solves the problem of cranial drill positioning and tracking in cranial surgery navigation.Therefore,this thesis designs a brain surgery navigation system,which establishes the intraoperative RGB-D camera coordinate system and the preoperative CT imaging coordinate system through the intraoperative 3D face point cloud registration of visible light and preoperative CT face transformation matrix between them.At the same time,the three-dimensional space pose of the cranial drill is calculated and tracked by the RGB-D camera,and projected into the CT imaging coordinate system to evaluate the degree of coincidence between the real path of the cranial drill and the virtual navigation path.Finally,the real surgical path conforms to the virtual navigation path by adjusting the pose of the cranial drill.