Winning a fresh $1.27 million to continue advancing its Quantitative Imaging Biomarkers Alliance (QIBA), the Radiological Society of North America plans to develop a quantitative-imaging data warehouse that, eventually, will be free and open to the public.

The funding has been flowing from the National Institute of Biomedical Imaging and Bioengineering, as the early-October allocation was the fourth from that NIH body in as many years.

QIBA, which pulls together researchers, clinicians and industry people, has been working since 2007 to reduce variability in medical images across devices, patients and time. Using quantitative imaging biomarkers—quantifiable features from medical images—the research applies sophisticated algorithms to test the technical performance not only of scanners but also of computer hardware, software and workstations.

The QI data warehouse is one of several initiatives to which QIBA hopes to apply the new funding. Also planned are the continued development and testing of phantoms and digital reference objects, along with research to characterize the sources of bias and achievable precision associated with quantitative imaging.

Right now most of the warehouse-ready scans are of test objects, or phantoms, but the intent is to introduce clinical scans into the warehouse as well, according to QIBA chair Daniel Sullivan, MD.  

“By superimposing synthetic lesions on a clinical scan, which the algorithm is designed to run against, we have a lesion in which we know what truth is,” he told RadiologyBusiness.com. “We can evaluate the bias and precision with which the algorithm determines that lesion.”

QIBA’s long-range hope, added Sullivan, is to make data sets available to anyone who wants to run an algorithm against a set of data—or do anything else they want to do with the data. “It would essentially be a public resource,” he added, “but it would also be potentially a validation facility so that users, consumers or buyers of an algorithm could have some independent assessment of how a given algorithm performed.”

QIBA’s work also involves “accelerating the development and adoption of standards” for hardware and software needed “to achieve accurate and reproducible quantitative results from imaging methods,” according to RSNA’s QIBA website.

Sullivan was kind enough to offer a serious response to an unconventional question: Can he picture a day when quantitative imaging allows for reads by artificial-intelligence robots rather than real live radiologists?

“I can’t say never, but I think that day is far off,” he replied. “There is such an infinite amount of variability in human bodies and in scans of humans. There are patterns that the eye-brain system, the trained observer, can pick up that would be extraordinarily difficult right now to create an algorithm to duplicate. Could artificial intelligence ever get there? I don’t know—but, if so, it’s very far in the future.”