Massachusetts General Hospital speeds up colonoscopy screening processing time with Intel

The 3D Imaging Research Department at Massachusetts General Hospital (MGH), in collaboration with Intel, Microsoft and Vectorform, used Intel Parallel Studio to optimize key image-processing libraries, reducing the processing time for computationally intensive colon screening from 60 to 3 minutes.

The Company 

Massachusetts General Hospital (MGH) provides sophisticated diagnostic and therapeutic care in virtually every specialty and subspecialty of medicine and surgery. MGH has consistently been named one of the best hospitals in the United States by U.S. News and World Report and is the oldest and largest teaching affiliate of Harvard Medical School.

MGH Imaging provides a full range of diagnostic testing services using state-of-the-art imaging equipment, and has distinguished itself with expertise in the following subspecialties: abdominal and interventional, breast, cardiac, musculoskeletal, neurological, nuclear, pediatric, thoracic and 3D imaging.

With 3D imaging, images obtained from computed tomography and magnetic resonance scanners are used to create realistic 3D images that physicians can use to visualize organs and diseases and to make diagnoses, treatments and surgical plans. This level of accuracy not only increases clinical productivity, but also improves patient care.

The PRODUCT 

Virtual colonoscopy (VC), also known as computed tomography (CT) colonography, is a CT scan of the colon. 

The CHALLENGE 

Manage highly parallel processing and virtual colon cleansing in real time, while reducing wait and processing times and improving reliability and quality of diagnostic images.

The RESULTS

The MGH 3D imaging research experienced a 20-fold speedup (from 60 minutes to 3 minutes) and developed a cutting-edge cross-platform solution using a single code base for multiple processor targets.

The IMPACT

Reduces wait times for patients and medical experts by enabling more performance and reliability, in addition to faster image processing and display speeds.

“Using technology we developed in close collaboration with Intel, Microsoft and Vectorform, we were able to dramatically reduce processing time and make virtual colonoscopy a practical, effective and potentially life-saving colon cancer screening technique.”

Explains Hiro Yoshida, director of 3D imaging research, Massachusetts General Hospital and associate professor of radiology, Harvard Medical School

MGH Takes Advantage of the Use of Parallelism

Virtual colonoscopy (VC) is a computed tomography (CT) scan of the colon, and it has been shown to be sufficiently accurate to detect precancerous polyps. Research on 3D imaging of the MGH has recognized that a successful VC procedure should be performed in less than five minutes and at a lower cost.

To achieve this goal, the MGH needed suitable processor and software technology that could handle highly parallel processing and virtual colon cleansing in real time. The 3D imaging research also required the software development tools necessary to build a parallel virtual colonoscopy application that leverages the latest processor technology. Speed and performance were major requirements to sufficiently reduce wait and processing times, while improving reliability and quality of diagnostic images.

ACCELERATE PROCESSING TIME

The study results showed that treatment time could be accelerated to make VC an increasingly effective option for a growing number of patient candidates for the procedure. The 3D imaging research resulted in performance gains equivalent to a 20-fold speed-up, reducing treatment time from 60 to 3 minutes. Productivity was also improved, enabling the hospital to meet a critical deadline for demonstrations at the Supercomputing and Radiological Society of North America conferences. Utilization of cross-platform capabilities was improved by using a single code base for multiple processors.

The improvements and reduction in image processing time have saved patients time and money.

Intel Parallel Studio and a Structured Workflow approach enabled on-time delivery of code that met performance, robustness and maintainability requirements. This also dramatically reduced the cost of a colonoscopy, making the procedure more comfortable and widely available. In recognition, MGH’s 3D imaging research received the “Excellence in Design” award from the Scientific Assembly at the 2011 Radiological Society of North America conference.

HOW INTEL’S SOFTWARE DEVELOPMENT PRODUCTS ENABLED THE SOLUTION

The Intel Parallel Studio suite of tools was instrumental in improving the speed of the MGH virtual colonoscopy application by leveraging the performance advantages of cluster and multicore processor technologies. The suite was used to accelerate the virtual colonoscopy processing application by up to 20 times. To achieve this goal, several components of the Intel Parallel Studio tool suite were used.

The Intel Parallel Advisor component was used first. This first step identified the most time-consuming areas of the application. Advisor helped clarify where parallelism would benefit the most, including those that have the most to gain from a significant reduction in execution time, whether through serial improvements, parallelizing code, or using existing optimized libraries. Some data structures and algorithms have been modified, and the convolution and compression features of Intel Performance Primitives (IPP) have been introduced.

Following this step, various strategies for introducing parallelism were proposed and the performance benefit of each was simulated.  For a given proposal, Intel’s parallel advisor also simulated potential conflicts in the parallel regions (e.g., deadlocks, race conditions). This was used to evaluate the correctness of a proposal. All errors identified by the tool were corrected. Implementation performance was characterized using Intel Parallel Amplifier, while accuracy was tested using Intel Parallel Inspector. Inspector was then used to test the now-parallelized unit, and on a few occasions it uncovered obscure issues regarding exactly how the parallelism had been added. He quickly found the causes of crashes that were very hard to see in the source code, and very hard to debug with a conventional debugger.

The working program was then characterized using Intel Vtune Profiler. Parallelism could be seen, and the next site to be addressed could be chosen. This allowed the work to be focused on the high-performance areas.

Intel Parallel Studio incorporated the parallel model that best fit the virtual colonoscopy application for display, processing and performance. The result was parallel code that met the requirements of the objective. At the same time, the structured workflow approach helped the developers achieve their performance goals in a fast and well-organized manner.

Source and illustrations: software.intel.com
Translation: ritme.com