A Spatial Registration Method Based on Rigid Coherent Point Drift of Key Features

Mixed reality and augmented reality technologies have been introduced into the field of surgical navigation for many years. Although many scholars have done a lot of research on holographic-based medical navigation techniques, the registration of preoperative MRI to the patient in a realistic clinical environment is still an existential problem. In this paper, a register method to align the preoperative medical images of patients to the intraoperative patient's body is proposed. The proposed approach simplifies the complex overall registration task of preoperative images of patients into the registration of local critical features. The proposed method extracts the patient's facial key features from the preoperative medical images and 3D information of the clinical environment reconstructed by Intel RealSense, respectively. And then the mapping transformation between the facial key features of preoperative medical images and the patient's key features in the realistic environment was estimated by the fusion of centroid movement and rigid point set registration algorithm of Coherent Point Drift. The centroid movement method performs preliminary alignment by overlapping the centers of mass of two sets of alignment data. For the two sets of alignment data containing noise and outliers, the Coherent Point Drift method is used to refine the alignment results. In the experiment of 4 key features the modified Hausdorff distance value is 3.2403 mm, the RMSE value is 5.8412 mm, and the TRE value is 5.6454 mm. The experimental results show that the proposed method is able to perform a robust and accurate alignment even though the preoperative medical images and the 3D information of the MR/AR environment come from different spatial coordinate systems and that noise and outliers are present in both data.