Evaluation of Depth-Depth-Matching Speed of Depth Image Generated from DICOM by GPGPU

We are developing a surgical support system to prevent surgical accidents in liver surgery. This system consists of three subsystems. First, The Liver Position and Posture Estimation System performs Depth-Depth-Matching of real depth image and virtual depth image to estimate the position and posture of the liver during surgery. The real depth image is the depth image of the real liver surface measured by the depth camera. The virtual depth image is the Z-buffer of the virtual liver generated from the DICOM captured preoperatively. Next, The Surgical Scalpel Tip Position Estimation System uses an optical 3D position tracker to measure the position of a grid pattern marker attached to the handle of the scalpel, and estimates the tip position by calculating the vector to the marker and the scalpel tip. And finally, The Liver Surgery Simulator receives the estimated position and posture of the liver and the position of the tip of the scalpel, and calculate the distance between the tip of the scalpel and the blood vessels. In this paper, we propose a high-speed Depth-Depth-Matching using GPGPU to estimate the liver position and posture using the real and virtual depth images. In the proposed method, a virtual depth image is generated from a 3D volume using GPGPU and Depth-Depth-Matching is continuously processed to speed up the position and posture estimation. We compared the performance of the proposed method with the conventional method, and confirmed the usefulness of the proposed method.