A New 3D Multi-modality Medical Bone Image Registration Algorithm

Abstract

Three-dimensional (3D) multi-modality medical bone image registration is an important technology in surgical application, especially in large computer-aided orthopedic surgery. To improve registration accuracy, we propose a new 3D multi-modality medical bone image registration algorithm based on local features through analyzing the bone structure. In this method, the image Hessian matrix is introduced for local features extraction, and the local behavior of the 3D bone image is described by the eigenvalues of Hessian matrix. This method can automatically extract and select the most representative feature points (blob-like structure) in different scales. Then we adopt the idea of triangle matching to get stereo matching point pairs. Improve random sample consensus (RANSAC) algorithm is adopted to remove wrong matching point pairs. We use the right matching point pairs to establish rigid transformation model and solve this non-linear model by Levenberg-Marquardt algorithm to get geometric transformation parameters. Simulated experiments and real experiments demonstrate that the proposed method can achieve a high image registration accuracy.