Biomedical imaging ecosystem and the role of the GPU

Abstract

The biomedical imaging chain is continuously being challenged to reconstruct, analyze, and visualize increasing amounts of data in shorter amounts of time. Parallel computing on multi-core devices and clustered computers has allowed for continued innovation of compute and processing technologies but not without facing serious constraints of cost, space, and power consumption. Over the last three years the graphics processing unit (GPU) and its increased programmability has played an integral role in defining a new dimension to parallel computing with its single chip, many-core architecture as well as evolving the graphics pipeline to enhance visualization techniques. Image reconstruction, segmentation and registration algorithms architected to take advantage of the GPU parallel architecture not only realize massive processing speedups but also set the stage for scalability. High resolution rendering of 3D and 4D datasets are navigated in interactive, real-time approaches. Real time ray tracing and 3D stereoscopic solutions bring increased realism to images. Understanding the optimized mix of GPU and CPU, both in the sense of hardware and software, is necessary for imaging applications to innovate, realize cost/performance efficiency and continue to enhance visualization. Several approaches for GPU programmability are available and will be explored. Innovations in the compute, graphics and visualization space will be discussed to show the relevance of the GPU throughout the imaging chain.