Through software for automated detection, quantification, decision support and diagnosis, machine learning is making major inroads into medical imaging. The way things are going, the market is likely to top $2 billion by 2023, according to a new Signify Research market report.

Despite years of seemingly relentless hype, it's “becoming increasingly clear that AI will transform the diagnostic imaging industry, both in terms of enhanced productivity, increased diagnostic accuracy, more personalized treatment planning, and ultimately, improved clinical outcomes,” noted the report.

With its future key role in letting radiologists handle the ever-growing volume of diagnostic imaging data, the issue of investing in the “right” software will remain a challenge.

“Many of the AI-based solutions for medical imaging that are coming to market are positioned as workflow productivity tools, but there is often a lack of clinical validation to show how much time these tools actually save and their real impact on how radiologists work,” Signify analyst and study author Simon Harris told HCB News, noting that, “similarly, there are few large-scale clinical studies on the accuracy of quantitative tools that provide automatic measurements of image features, such as the long and short measurements of lung nodules, and the variability of the results obtained from tools from different vendors.”

Harris advised healthcare providers interested in making an AI investment “to look for vendors who have invested in clinical studies and are able to provide robust clinical evidence to back up their marketing claims.”

The study noted that the development pace “is faster than ever before,” and is leading to a surge in products from more vendors.

"The interest and enthusiasm for AI in the radiologist community has notably increased over the last 12 to 18 months, and the discussion has moved on from AI as a threat, to how AI will augment radiologists,” suggested Harris, adding, “At the same time, there are emerging clinical applications where the use of AI has been shown to both improve clinical outcomes and deliver a return on investment for healthcare providers. Examples include software to detect and diagnose stroke, and analysis tools to measure blood flow in noninvasive coronary exams.”

Still in its innovator and early adopter phase, AI for medical imaging faces several challenges, including a regulatory process that has been slow to approve products and a lack of large-scale studies to illustrate that deep learning algorithms works in real-world clinical settings.

Moreover, there are integration challenges to incorporate the information from AI systems into radiologists' workflows. This will require that developers work with imaging vendors to ensure such integration.

"Up to now, the market has mainly been driven by the many startups and specialist companies that are applying machine learning to medical imaging, but the major medical imaging vendors are now ramping up their AI activities,” Harris advised.

Some of the big tech players moving into the field “in the last year or so,” he said, included such giants as China's Tencent and Alibaba.

“The combined R&D firepower of the expanding ecosystem will knock down the remaining barriers, and radiologists will have a rapidly expanding array of AI-powered workflow and diagnostic tools at their disposal,” the study advised.

In late July, a paper in the Journal of the American College of Radiology looked at the the ways that AI overhead costs will change medical imaging economics.

“When we look at it (AI), we don’t necessarily look at if we can get more done faster. If I can get my same volume of work done instead of in a ten-hour day, in a nine-hour day, that gives me another hour I can work with patients, the hospitals, and administrators to further patient care and actually achieve better outcomes," Dr. Kurt Schoppe, the author and chair of the Reimbursement Committee of the American College of Radiology Economics Commission told HCB News.

Big data, machine AI learning has already demonstrated it can often read images more efficiently than humans. But for radiologists, this means they will get more time – and enjoyment, according to Shoppe – back in their practice. This is because radiologists will be relieved of many repetitive and mundane reading tasks.

But a challenge remains, namely, how will such efficiencies be reflected in payment policies to radiologists?

“For artificial intelligence, there may not be a physician work component, so the RUC and Medicare likely won't acknowledge the technical component of the reimbursement," Schoppe explained. “This is why vendors need to understand the nuance here, because it affects how they calculate potential returns on investment.”