An Automatic Spatial Registration Method for Image-Guided Neurosurgery System

Objective: This study aimed to investigate the feasibility of an automatic marker-free patient-to-image spatial registration method based on the 4-points congruent sets (4PCS) and iterative closest point (ICP) algorithm for the image-guided neurosurgery system (IGNS). Methods: A portable scanner was used to obtain the point cloud of the patient's entire head. The 4PCS algorithm, which is resilient to noise and outliers, automatically registered the point cloud in the patient space to the surface reconstructed from the patient's pre-operative images in the image space without any assumptions about initial alignment. A variant of the ICP algorithm was then used to finish the fine registration. Two phantoms and 3 patients' experiments were performed to demonstrate the effectiveness of the proposed method. Results: In the phantom experiments, the mean target registration error of 15 targets on the surface of the rigid and the elastic phantoms were 1.02 +/- 0.18 mm and 1.27 +/- 0.36 mm, respectively. In the clinical experiments, the mean target registration error of 7 targets on the first, second and third patient's head were 1.88 +/- 0.19 mm, 1.84 +/- 0.19 mm, and 1.89 +/- 0.18 mm, respectively, which was sufficient to meet clinical requirements. The registration accuracy and registration time using the proposed method are better than that using the method based on manually coarse registration and automatic fine registration. Conclusions: It is feasible to use the automatic spatial registration method based on the 4PCS and ICP algorithm for the IGNS. Moreover, it can replace the spatial registration method based on manually selected anatomical landmarks combined with the automatic fine registration in the currently used IGNS.