Global Citizenship
Letter from our CEOSocial Impact of the GPUFY11 Citizenship Report

Improving the Quality of Healthcare, Many Steps at a Time

Medical imaging technologies running on GPUs produce better, safer results in less time.

For people facing the possibility of serious illnesses like cancer, accurate, timely test results are vital. Exposure to radiation, a necessity to perform some tests and treatments, may itself pose risks over time. At the very least, delays in imaging processing and diagnosis can hinder both a hospital's efficiency and a patient's peace of mind.

To address these challenges, graphics processing units (GPUs), most commonly known as the engines that render rich, dynamic graphics for computer games are today being used in a variety of medical imaging technologies. And they're well-suited for the job.

The 3D images produced by CT, ultrasound, MRI and PET scans are computationally intensive. GPUs use parallel processing to break down such complex computing problems into many smaller tasks that run simultaneously. This core capability is driving dramatic improvements across the spectrum of medical imaging technologies and, ultimately, helping make the healthcare system better for both providers and patients.

Watch the video here:
Third Pillar of Science - Leading researchers and companies are using NVIDIA GPUs to transform imaging and other areas of science.
Making cancer detection and treatment more precise and safer
One of the more advanced diagnostic tests for detecting cancer is fluorescent tomography (DFT), which uses the absorption and scattering of light in tissues to spot dangerous growths. One challenge with DFT is that light disperses as it moves, so it can be difficult to see malignant cells that are deeper within the body.

To overcome this issue, researchers at the Applied Physics Institute at the Russian Academy of Science developed algorithms to reconstruct the 3D position of the fluorescent markers used in DFT. The result: they could pinpoint cancer cells with greater accuracy.

The researchers used the Monte Carlo method (which uses repeated random sampling) to achieve their simulations, which typically require the calculation of roughly 1 billion random paths. While this process can be painfully slow using traditional CPU-only processing, it's perfectly suited to parallel processing. When the scientists switched to a GPU-based system, the average runtime for the tests went from about 2.5 hours to 1.5 minutes — a hundred-fold speed-up. The researchers were also able to add more paths to the calculations, which ultimately increases accuracy.

In a separate study, researchers at the University of California, San Diego were looking to minimize the damage to healthy tissue when treating cancer patients with radiation therapy. To effectively accomplish this, they needed live images to guide the treatment.

They deployed an advanced form of CT imaging, called Cone Beam CT (CBCT), for image-guided radiation therapy. This type of therapy uses multiple, repeated scans to precisely target how radiation is delivered to the site of a tumor.

To reduce the level of radiation often necessary in CBCT scans, Jiang's team developed a new iterative algorithm based on compressed sensing technology, to reconstruct images from undersampled and noisier data. This resulted in much less radiation exposure for patients. Like any other iterative reconstruction algorithm, this new algorithm is also computationally intensive and thus time consuming. To reduce the amount of time needed to create a clinically acceptable image, Jiang's team also applied GPU-based technologies, reducing the reconstruction time to less than two minutes – 100 times faster than similar approaches using non-GPU technologies.

"GPU technology has enabled us to dramatically reduce the amount of radiation exposure a patient must receive, while speeding up the time that it takes to be scanned," said Steve Jiang, UCSD associate professor of radiation oncology. "One of the biggest beneficiaries of these advances is young children, who can now receive state-of-the-art cancer treatment without being exposed to high doses of imaging radiation."

Dr. Jiang and his team are working to extend this technique to general diagnostic imaging, thereby reducing exposure to radiation for a greater population of patients.

“GPU technology has enabled us to dramatically reduce the amount of radiation exposure a patient must receive, while speeding up the time that it takes to be scanned.”
Steve Jiang
Associate Professor of Radiation Oncology University of California, San Diego
Detecting breast cancer in minutes with GPU-powered Ultrasound
According to the American Cancer Society, about 1 in 8 women in the United States will develop invasive breast cancer over the course of her lifetime. Early detection can play a critical role in treating the disease.
TechniScan Whole Breast Ultrasound system

TechniScan, a Salt Lake City, Utah-based medical device company, has developed a GPU-powered mammogram imaging system called Warm Bath Ultrasound. The system, which uses rotating scanners to generate a three-dimensional image of a patient's breast, captures eight to nine million voxels of data. (A voxel is the 3D equivalent of a pixel.)

A traditional CPU-only system can process that amount of data in about 4.5 hours. It takes the GPU-based system just 20 minutes. In addition to providing a clear, detailed image, that turnaround time delivers priceless peace of mind. Ultrasound's use of sound waves instead of radiation also makes this approach safer for the patient.

Advancing the state-of-the-art across medical imaging technologies
Massachusetts General Hospital is experimenting with a new class of CT imaging called iterative scanning, which uses a physics-based model. Instead of a bi-directional reflection of X-rays, iterative scanning propagates the X-ray through the body to get an integrated view. This requires a lot of computational horsepower, but with a server running eight NVIDIA C2050 "Fermi" GPUs, Mass General has the tools for the job.

"When you look at the kernel time [the amount of processor time requested by an application], with all eight GPUs, the back projector is 4,000 times faster than a quad-core CPU. So the potential speedup is enormous," said Dr. Homer Pien, director of the research group for the radiology unit at Mass General.

To advance ultrasound, Siemens is expanding its boundaries into 3D. Coupled with NVIDIA 3D Vision, which is powered by GPUs and includes custom glasses, Siemen's syngo.fourSight Workplace software delivers an immersive, high-definition 3D view of a fetus. Physicians can more readily detect abnormalities and parents can have a lifelike view of their child.

Researchers with the University of California, Los Angeles are seeing improvements when GPUs are used in PET scans. In a PET scan, a radioactive isotope tracer is injected into the body to detect cancer cells, which are more likely to consume the isotope. GPUs allow for higher quality image processing in real time, so as the isotopes pass through the body, their travels can be monitored more accurately. And of course the GPUs are fast. PET scan renderings were cut from 4 hours with CPU-only processing to under five minutes with the GPU's parallel processing.

A project at Simon Fraser University in Canada found that processing of Dynamic PET scans, where the scanner watched live tissue in the body to see how it behaved, could be cut from many days to just 13 seconds using GPUs and CUDA-based software algorithms. And at the University of Illinois at Urbana-Champaign, researchers are improving MRI scans by developing a reconstruction algorithm that enhances image quality using GPU technology.

For computer gamers, GPUs enable the creation and destruction of fantasy worlds in the blink of an eye. For healthcare professionals and their patients, medical imaging technologies running GPUs are delivering safer, faster, higher quality care.

Watch the video here:
Siemens and NVIDIA Quadro: The Wonders of 3D Imaging