Point Cloud Reconstruction Based on Adaptive Template Line Laser Stripe Matching for Surgical Registration

The registration between preoperative medical image and intraoperative patient spatial is a critical concern within surgical navigation systems. Conventional registration methods typically involve utilizing a probe to contact the patient's skin and acquire registration point clouds, a process characterized by low efficiency and a considerable demand for surgical expertise. This article presents a laser-based non-contact registration method that offers a solution to these challenges, which introduces a laser centerline extraction technique based on an adaptive template. By employing binocular epipolar rectification, the three-dimensional (3D) point cloud is reconstructed. Subsequently, statistical filtering is applied to denoise the reconstructed point cloud. Ultimately, a non-contact registration is performed by aligning the reconstructed point cloud with a preoperative medical image segmentation model, achieving the fusion of preoperative medical images with the patient's anatomical structures. Reconstruction and registration accuracy validation experiments were conducted on a head phantom. The outcomes revealed an average registration error of 1.54mm between the facial reconstruction point cloud and the CT model, with an average virtual-real fusion error of 0.93mm between the CT model and the head phantom. These results significantly outperformed contact-based registration methods.