January 27, 2010, 4:40 PM — University and hospital researchers have taken a cue from the gaming industry by using 3D video graphics chips to develop a parallel software platform that can speed the processing times, in this case for medical digital imaging, by 10 to 1,000 times.
Northeastern University in Boston and Massachusetts General Hospital (MGH) were jointly awarded a $1.3 million grant from the National Science Foundation in December to develop the technology and use it to enhance several biomedical imaging applications, including software designed for breast and brain imaging.
The project has started and the joint team expects to begin delivering software to vendors by year's end.
Early on in the effort, MGH has already shown that the parallel technology can cut the time it takes to read a 3D-radiological breast image from hours to seconds, meaning a patient could be cleared to have a biopsy during a visit to the doctor for a breast exam.
"That way they can determine whether there's a malignant tumor in the breast," said Professor Dave Kaeli, who leads Northeastern's parallel software development team. "The implications for a patient's mental health as well as cost of care for patients are huge,"
The new software libraries will enable radiologists to more quickly assess tissue images, which will allow clinicians to perform image-guided biopsies earlier than can be done now. The faster process will lead to quicker cancer diagnoses detection of cardiovascular plaque. The software will also allow for the rapid delivery of data needed by surgeons; and for the delivery of more accurate radiation treatment.
Kaeli's team is developing the software on systems running Graphics Processing Units (GPUs), very high-performance processors with as many as 200 cores on a single chip.
The development team is using proprietary development software, called CUDA, from GPU maker Nvidia , and the open source OpenCL, Open Computing Language, that was originally developed by Apple Inc. Cuda and OpenCL act as a framework for writing software programs that can execute across heterogeneous GPU platforms.
"We can use one or a few cards. What we do is take a serial application and rewrite it to exploit the data-level parallelism available on graphics cards," Kaeli said.
Homer Pien, PhD, director of the laboratory for medical imaging and computation in MGH's Department of Radiology, said the parallel processing software will enable the use of a radiological imaging technique called iterative reconstruction. The technique can complete x-ray tracing in the same way computer graphics in the gaming industry uses light ray tracing as a technique for generating an image by tracing the path of light.