high performance computing on graphics processing units: hgpu.org

This paper presents an acceleration method for tracking a bronchoscope based on image registration. This method tracks a bronchoscope by image registration between real bronchoscopic images and virtual ones derived from CT images. However, since the computation cost of image registration, especially generating virtual bronchoscopic (VB) images, is quite expensive, it is dificult to track the bronchoscope in real-time. To solve this problem, we try to accelerate the process of image registration by utilizing GPU (Graphics Processing Unit) with CUDA language. Specifically, we accelerate two parts: (1) VB image generation by volume rendering, and (2) image similarity calculation between a real and a virtual bronchoscopic images. Additionally, to obtain the maximum performance of GPU as much as possible, we minimize (i) the amount of memory transfer between CPU and GPU, and (ii) the number of GPU function calls from CPU. We applied the proposed method to ten pairs of real bronchoscopic video and CT images. The experimental results showed that the proposed method could track the bronchoscope 16 times faster than the method using only the latest CPU.