Registration method for three-dimensional point cloud in rough and fine registrations based on principal component analysis and iterative closest point algorithm

For the registration between the preoperative computed tomography (CT) image and the intraoperative patient space in surgical navigation technology, this paper puts forward a registration technique based on the principal component analysis (PCA) and the iterative closest point (ICP) algorithm, using two feature point clouds from the medical image space and the actual patient space. Firstly, the feature point cloud of the image space was obtained through the reconstruction, segmentation and interactive operation of the CT image, while the corresponding feature point cloud in the actual patient space was collected by the optical locator in real time. Secondly, the eigenvectors of the two sets of point clouds were obtained through the PCA for rough registration, and the optimal solution of the registration matrix was found by the ICP. Finally, the effect of the proposed registration method was verified against point cloud data (the surgical navigation accuracy was evaluated through an experiment on the vertebra model), and the impacts of the number of point clouds and Gaussian noise on the registration error were investigated in details. The results show that the proposed method is simple and capable of realizing high registration accuracy, and completed registration with a less-than-2mm error in our experiment; in addition, the registration accuracy was greatly affected by the number of point clouds and the noise of the photoelectric locator. This research provides a general solution for registration in surgical navigation and lays the theoretical basis for improving intraoperative registration accuracy.