With virtual RSNA 2020 over and the year drawing to a close (finally!), I’d like to share my predictions about the world of radiology artificial intelligence (AI) for next year.
When all is said and done, 2020 was a big year for radiology AI, with even more algorithms cleared by the U.S. Food and Drug Administration (FDA) and the first-ever Medicare new technology add-on payment (NTAP) approval for AI in radiology.
In my opinion, the three major trends for 2021 that will take hold in the AI-medical imaging ecosystem are the following:
Unsurprisingly, my first prediction is that hospitals and private practices will continue to adopt radiology AI at an even faster pace than before. In a new study (Radiology: Artificial Intelligence, November 11, 2020, Vol. 2:6), Dr. Yasasvi Tadavarthi and colleagues estimated that next year the market cap for image analysis companies like Aidoc will hit a whopping $2 billion, up from $1.2 billion in 2019, due to more and more radiologists adopting AI into their workflow.
With over 30 FDA clearances issued to radiology AI this year alone and the recent announcement that a large vessel occlusion (LVO) prioritization algorithm will qualify for U.S. Centers for Medicare and Medicaid Services reimbursement under NTAP — something that all AI vendors with similar functionality can take advantage of — the return on investment from the usage of AI in clinical practice is becoming apparent.
But even with the advent of additional radiology AI use cases qualifying for NTAP, I don’t believe there will be additional reimbursement avenues, and there will be more emphasis on understanding ROI prior to deployment in clinical practices.
Even as radiology AI becomes more widespread than ever, 2021 will be a year of significant consolidation for developers. It is often said that the radiology AI market is an overhyped bubble, and COVID-19 might just put an end to that.
RSNA hosted 350 companies in 2019 in its “Machine Learning/Computer-Aided Diagnosis Systems” category. Given that the AI hype started more than half a decade ago, we can already see a divide between established players — with regulatory clearances, active users, and recurring revenues — and between new algorithm developers that will have to become more aggressive in their approach to market in order to catch up.
The numerous machine-learning companies that develop algorithms for image analysis are competing for an increasingly finite number of resources. COVID-19 has changed the playing field for everyone, especially the healthcare industry.
The combination of lower hospital budgets and reduced stability in the financial market will only widen the gap between industry players with significant adoption and those without. I believe we will see some radiology AI players begin to consolidate or liquidate, as well as potential mergers or acquisitions among the bigger players.
Since the birth of the radiology AI industry, algorithms across all imaging modalities have been lumped together under the “AI” banner. Frankly, it made the AI section at RSNA look like a big mishmash of technologies.
It has always bothered me, since what we sell is value, not an underlying technology. As products mature with different use-cases, we will begin to see a focus on concrete categories of radiology AI, e.g., acute triage and notification, screening, communication, and population health. Companies will have a broad or narrow offering in each of those categories. We may even see conferences like RSNA opt to differentiate AI providers by types of solutions.
This blog was originally published in Aunt Minnie’s Online Edition on December 15th, 2020.