For AI to become clinically feasible in women’s imaging, it must excel in the areas of performance, time, workflow and cost, according to an opinion piece published online Feb. 19 in the American Journal of Roentgenology.
The adoption of any new technology must achieve certain metrics to become viable and quantify its impact, wrote authors Ray C. Mayo, MD, and Jessica W. T. Leung, MD, each with the department of diagnostic radiology at The University of Texas MD Anderson Cancer Center in Houston.
“As patients, physicians, hospitals, and insurance companies look for value, AI must earn a role in medical imaging,” Mayo and Leung wrote.
Below are four areas of women's imaging AI must impact in order to become clinically feasible:
Improving the performance of mammography is the most important condition AI must achieve to be useful in women’s imaging, according to the authors. Without this there is no interpretative purpose for the technology. In turn, the high rate of false-positive results in women’s imaging must also decrease.
“If AI earns a place in breast imaging, its best opportunity is to decrease false-positive flags compared with the number associated with currently available (computer aided detection) CAD programs,” the authors wrote. “There are limited research data, but so far AI programs analyzing mammograms have exhibited performance equal to that of radiologists in assigning breast density.”
To successfully implement AI into diagnostic radiology and clinical practice, it also must not increase interpretation time for radiologists or delay the delivery of images to radiologists. This is especially vital for radiology exams performed in the emergency department (ED) where quick and timely intervention are crucial.
The integration of AI into clinical workflow must be seamless, according to the authors, and should start before image acquisition by combing through medical records to identify patients in need of appropriate imaging exams. The ordering process and scheduling may then be initiated automatically on behalf of the referring physician.
Additionally, an ideal computer format for AI would be one in which it can be toggled on an off by the radiologists in an image overlay manner. The algorithms must also be compatible and fully integrated with the hardware and software systems of various manufacturers.
An example of how AI could improve imaging workflow and patient care, the authors noted, is the concept of “rereading” where the algorithm reviews recently archived imaging exams not subject to initial AI CAD evaluation.
“If even one cancer per thousand cases were to be detected, this process would positively affect the cost-benefit ratio,” the authors noted.
Lastly, the cost of providing and integrating AI into women’s imaging must not be so high that it “tilts the value equation against its use,” according to the authors.
“Currently, CAD is included as part of a bundled charge for both screening and diagnostic mammograms rather than an individual separate charge. Introducing a new code for billing is unlikely,” they wrote.
The speed and availability of high-performance computers should lower any barriers to developing and implementing AI. Over time, costs should decrease as the emergence of numerous AI models in the market quantify the cost-benefit ratio of AI, the authors concluded.