Research On2D-3D Non-rigid Registration Method In3D Femur Model Reconstruction

This thesis deals with the problem that the overlapping DR images can notreflect the3D space information in the orthopaedic surgery diagnosis. A2D-3Dnon-rigid registration and pose estimation method is proposed to realize the3D posevisualization of the patient femur before the surgery, which uses the DR images ofthe patient femur from anterior and lateral views and the CT volume data of ageneric femur. With the advantages of low cost, fast imaging speed and littleradiation to the patients, this method can provide more effective3D imaginginformation for the femur fracture diagnosis and preoperative plans.Firstly, the basic theory and methods of medical image registration areintroduced and the main research steps of femur2D-3D registration and3D poseestimation are stated. Secondly, the basic princ iple of Zhang Zhengyou calibrationmethod is expounded. The traditional calibration board is redesigned and thecalibration parameters of DR are acquired accordingly. Thirdly,3D generic femurmodel is reconstructed and its DRR images are acquired using the CT and DRRtechnique. Then the edge detection step of the DR/DRR images are completed toobtain the contour coordinates of the patient or generic femur respectively.Based on the iterative closest point(ICP) theory, this thesis proposes a2D-3Drigid registration method using ICP algorithm, which can complete the3D poseestimation of the patient femur. Then a2D-3D non-rigid registration method usingaffine ICP algorithm is proposed. The method considers the similarities anddifferences between the patient femur and generic femur by increasing theproportion deformation, which makes the3D model after registration closer to thereal patient femur. Finally, our method is method is validated and evaluated througha large number of2D-3D registration experiments.The registration experiments shows that the proposed method can effectivelyrecover3D pose of the patient femur with a fast, accurate and robust registrationresult, which will satisfy the need of clinical application.