Multi-GPGPU Based Medical Image Processing in Hip Replacement

Abstract

Image processing techniques are widely used in medical software applications. For instance, feature extraction methods such as Canny edge detector and Hough transform can be applied to radiographic images for the purpose of computing certain parameters. The large number of pixels in a radiographic image leads to slow computing times for such algorithms. GPGPU Implementations of these algorithms that reduce the computing times already exist. The novelty of the proposed approach is in taking advantage of the power of multi-GPGPU accelerated computers and other mechanisms of the CUDA architecture. The particularities of orthopedic radiographic images are also taken into account to further reduce the computing times. The new implementations were tested on a computer with two graphic cards and were compared to CPU and other GPGPU implementations. The comparison between the CPU and existing GPU implementations show that multi-GPGPU applications can add a significant performance gain to image processing applications.