Frameless Registration Method Using a Depth Camera for Robot-Assisted Stereotactic Brain Surgery

Stereotactic surgery aims to access critical areas of the brain with high accuracy. The classical surgical process requires two separate radiological imaging datasets (MRI-CT) and their precise registration. Additionally, specific anatomical landmarks (AC, PC, TAL) are manually identified by the neurosurgeon in both datasets, and MRI-CT registration is performed using these landmarks. To address the issues of patients' double exposure to radiological imaging and the manual identification of landmarks, this paper proposes aAnew approach based on the registration of facial landmarks. The proposed approach consists of four stages. The first stage involves creating 2D facial masks (MRHead and DHead) from the MRI and depth camera data obtained from the patient. Each mask, automatically generated using Google Mediapipe software, consists of 468 points. In the second stage, the mask points are transformed from 2D to 3D. In the third stage, precise registration of the 3D mask points is achieved using singular value decomposition (SVD) and random forest (RF) methods. In the final stage, using the registration matrix, the robotic arm is guided to reach the desired target point on a 3D-printed head prototype. Using the RF method for MHead and DHead mask registration, we obtained fiducial registration error (FRE) values of 1.633 mm and 1.523 mm, and target registration error (TRE) values of 2.217 mm and 2.164 mm for each patient, respectively. These promising results will form the basis of further developments in fully autonomous brain-targeting software with robotic assistance.