Accelerating biomedical image processing using massively parallel processors



 DESCRIPTION (provided by applicant): During the last decade the quantity of bioimaging data has grown tremendously. Current estimates indicate that the average hospital in the USA houses some 665 TB of data of which approximately 80% is composed of unstructured image data from CT, MRI, and X­ray machines. This huge quantity of data is expected to grow at a rate of 20­40% annually, meaning hospitals could generate a total of one exabyte of new biomedical imaging data this year. The last decade has also seen the development of several new computing platforms. In particular, multi­core and massively parallel processors are ubiquitous. Of these new platforms, the sheer computational power in modern Graphical Processing Units (GPUs) have created a computing era where it is feasible for a developer to purchase a personal supercomputer with 10+ teraflops of processing ability for less than $20,000. One of the most popular components of modern biomedical imaging software, the Insight ToolKit (ITK), could benefit greatly from GPU computing. There have been two attempts to implement ITK's functionality on the GPU and although there were impressive results (accelerations between 5 ­ 800x); both projects were ultimately abandoned. As it stands, our GPU accelerated ArrayFire library already contains about 26% of ITK's core functionality, more than any competing software. Within the context of this proposal we seek to expand ArrayFire's support of ITK's functionality and create tools that will help developers use ArrayFire to leverage the massively parallel computing capabilities of GPUs from their ITK applications.