Chapter 46 - Medical Image Processing Using GPU-Accelerated ITK Image Filters

Abstract

This chapter introduces GPU acceleration techniques for medical image processing using the insight segmentation and registration toolkit (ITK). ITK is one of the most widely used open-source software toolkits for segmentation and registration of medical images from CT and MRI scanners. It consists of a collection of state-of-the-art image processing and analysis algorithms that often require a lot of computational resources. Because many ITK image filters are parallel and do not require interthread communication, one can exploit the fine-grain data parallelism of the GPU. To validate these ideas several low-level ITK image processing algorithms implemented, such as convolution, statistical, and PDE-based de-noising filters, using NVIDIA's CUDA. There two main goals in this work: outlining a simple and easy approach to integrate CUDA code into ITK with only minimal modification of existing ITK code, and demonstrating efficient GPU implementations of spatial-domain image processing algorithms. It has been demonstrated that GPU computing can greatly boost the performance of existing ITK image filters. In addition, a simple approach to integrate CUDA code has been introduced into the existing ITK code by modifying the entry point of each filter. Using only a single GPU easily beats eight CPU cores for all tests that were performed, thus confirming the great potential of GPUs for highly parallel computing tasks.