From Scuds to tumors: Qualia applies military AI to mammography

System may be alternative to more expensive workstations

As the processing power of desktop computers expands exponentially, radiologists are gaining access to a growing arsenal of powerful yet inexpensive image-analysis software tools. These processing tools, some developed for sophisticated defense and aerospace purposes, will assist doctors in performing even low-cost screening examinations. If adopted by hospitals, the tools could reduce healthcare costs by both detecting cancers earlier-allowing for timely therapeutic intervention-and reducing the number of biopsies performed on benign lesions.

Algorithmic expertise used in programming F-16 fighter jets to search and destroy Scud missile sites and other ground targets has been brought to bear by Qualia Computing (QCI) of Beavercreek, OH, on the targeting of potentially malignant regions of the breast, said president and CEO Steve Rogers.

Rogers developed military targeting technology during his career with the U.S. Air Force, he told SCAN. Following his retirement three years ago, he decided to apply this expertise in new areas, medical image processing being the first, and brought with him a team of experienced software engineers to form QCI.

QCI signed on Canada's Briana Bio-Tech in April to provide development funding and royalties in exchange for worldwide distribution rights to QCI's Second Look digital image analysis system when the product reaches the market. Second Look uses sophisticated neural network software to detect both microcalcifications and suspicious masses in mammographic images (SCAN 5/14/97).

Consistent with today's cost-conscious healthcare environment, the artificial-intelligence (AI) algorithms in Second Look will produce inexpensive laser-printed images on paper, marked up and annotated with advice for the radiologist in reading the original films, Rogers said. That differs from the more expensive strategy of many competing computer-aided diagnosis (CAD) system developers.

"Most of our CAD competitors are going for high-end workstations. That is not the direction we are going because that is not what we hear radiologists saying. They want to stay on film for now," Rogers said. "Although we are going low-end in terms of hardware expense, our Pentium-based solution is very quick and precise. That is our niche: very quick, very portable, very inexpensive."

Real-time processing is key to the programming of fighter jets, he said. As data enters the AI processors embedded in a plane, locational answers must come out the other end fast enough to keep up with the data flow. Speed is key in medical image processing as well. The mammographic analysis system must be fast enough to be transparent to the physician and not put a drag on throughput. In this application, QCI defines real-time as a minute or less.

"After the film gets developed, it is put into our digitizer. Then, before the radiologist ever looks at the film, we will have an answer," Rogers said.

Neural networks are a subclass of artificial intelligence technology, Rogers explained. Second Look's neural network software is designed to mimic the workings of neurons in the brain, and as with the human brain, the software learns from experience. In mammographic imaging, past case examples, such as biopsy results, are fed through the program until the system learns through adaptive signal processing to distinguish malignant from benign lesions.

While Rogers feels that QCI's image analysis technology is capable of providing tumor diagnosis, QCI does not plan to make that claim for the initial version of Second Look. In large part, this is because firms can expect a significantly longer certification process at the Food and Drug Administration when they submit claims to be able to diagnose pathology.

"It seems pretty clear that the only way we are going to get through the FDA in any sort of reasonable amount of time is (by claiming) detection only," he said. "After we get that product approved, we will then go back to the FDA and present our clinical work on diagnosis."

As a detection-only system, Second Look will point out regions of the breast where possible malignancies may exist, allowing the radiologist to refine the initial diagnosis if necessary, Rogers said. This will help catch more breast tumors earlier. The second clinical advantage-reducing unnecessary biopsies on questionable lesions-will come about when the technology is applied as a direct diagnostic aid.

When it reaches the market, Second Look may be provided to physicians either as a stand-alone system of digitizer, PC, and printer, or as an online service, he said. In the latter case, the hospital would digitize on-site and send the image to a centralized location run by QCI. The company would then respond with an image report via the Internet. Prospective international customers have expressed an interest in this type of remote service.

Which product strategy or combination of strategies QCI and Briana will use for Second Look is still an open question, Rogers said. The companies plan to show a prototype of Second Look at this year's Radiological Society of North America conference in November.