In a move that hopes to counteract years of promoting AI as a viable alternative to a radiologist interpretation, AI vendors seemingly have changed direction and have become like Mr. Rogers, asking radiologists, "Won't you be my neighbor?" Radiologists have answered the question in return with a definitive "We'll see."
The AI Showcase at RSNA 2019 has 123 vendors exhibiting this year, nearly double the number of vendors last year and triple that of the year before. Interest has never been higher.
Yet two questions remain: how can AI be used and how can it show a return on investment? I walked the technical exhibits of McCormick Place and spoke with exhibitors in the AI Showcase to find out.
PACS consultant Michael J. Cannavo.
The makeup of the AI market remains pretty much the same, with unbundled single-solution AI providers making up better than 90% of the market space, consolidators around 8%, and a bundled single-source provider the domain of just one vendor so far.
There have been a few new entries into the AI marketplace (or consolidator) space, which previously was the domain of less than a handful of vendors, and this area is poised to grow as the market grows.
A "marketplace" vendor offers a range of different solutions from different AI vendors. The customer pays a single interface fee to connect to the consolidator and then picks and chooses the algorithms from the multiple vendors offered by the marketplace vendor.
There also are a few companies at RSNA that allow the end user to develop (and ostensibly use) their own algorithms, but the legality of this approach was not explored or discussed. I did not see any algorithms from academic or teaching institutions, but they may have been located in the poster exhibit area of McCormick Place, as we have both heard and read about them.
The overall AI field has seen total investments of $6.6 billion through 2018, with context-aware processing (medical imaging) accounting for almost 20% of that ($1.2 billion). Nearly half of that ($580 million) has been raised since 2018 alone.
Medical imaging makes up about 10% of the total AI market share today and is growing to more than double its share by 2024, with an average investment in AI companies at around $9 million.
So where is this money coming from? For the most part venture capitalists. Venture capitalists a few years back were handing out money like candy to trick-or-treaters, but now they seem to be asking many more questions of AI developers before continuing to invest in them, and they seem to be getting a bit more selective in the investments they are making.
The growth of companies in the AI space outside the U.S. is impressive, with Korea seemingly dominating the market space, followed by China and others. Israeli companies also have had a strong presence. About one in four companies are first-time exhibitors at RSNA, with half as many exhibiting for the second year in a row.
Confusion about clearances
There remains confusion relative to how many companies have U.S. Food and Drug Administration (FDA) 510(k) certification for their algorithms. It was initially thought that one in three vendors had 510(k) certification, but the number is actually closer to one in four (if that), with all but a few obtaining their 510(k) clearances in the past two years.
At RSNA 2019, many companies I talked to stated that their 510(k) was pending, although several of these companies had pending certifications last year as well.
CE Mark certification, the European equivalent to a 510(k) that allows products sold within and outside the European Economic Area (EEA), has been much more prevalent at RSNA 2019 than initially expected. About half of the products displayed had CE certification, with a few having certifications from the respective countries of development (such as Korea, Japan, and Thailand). That left roughly half of all companies without any certification whatsoever. In the U.S., a 510(k) is required before a product can be used clinically and included as part of the final report.
Why are so many companies getting into AI? Ben Panter, CEO of Blackford Analysis, addressed this in a chat with me.
"Significant progress in the underlying technology stack has lowered the bar for development of AI in all sectors. Combine this with strong VC interest and the long-term potential to reduce healthcare spend and you generate considerable excitement through startups and hospital-based initiatives," Panter said.
All the executives I chatted with at RSNA 2019 cited the ease of entry into the AI market as the reason why so many vendors are jumping into the fray. Gene Saragnese, CEO of MaxQ AI, summed it up nicely.
"There is clearly a need -- the healthcare sector today is under a level of pressure that is unprecedented, with less money than ever before and a shortage of healthcare providers attempting to serve a growing number of patients," he told me.
Yet many, including Saragnese, feel that market growth is sustainable.
"To truly relieve the pressure on healthcare, any new AI-enabled solutions must augment radiologists and care providers and allow them to make critical decisions quickly, correctly, and that saves lives. ... Those companies that pursue a generalized, scalable, and secure approach will be standing in two years from now, and those who do not might not be," he said.
John Axerio-Cilies, CEO of Arterys, explained, "There will be companies that will have trouble fundraising, as they won't be able to reach the revenue targets they promised. Lots of them have focused on the AI aspect and don't properly estimate what it takes to build clinical applications, regulating them, selling them, and integrating them," he said. "Market consolidation is expected to happen for most in the next 18 to 24 months, with some vendors getting acquired and others who lack adequate funding going by the wayside."
Eyal Gura, CEO and chairman of Zebra Medical Vision, whose company has one of the longest tenures in the AI market, summed it up nicely. He cited "access to data for multiple applications and multiple verticals, funding, strong team with published results, proven deployment ability with customer success, and proven regulatory clearances execution" as the keys to a company making it in the market.
TeraRecon, developer of EnvoyAI technology, takes a slightly different approach as a marketplace provider. Jeff Sorenson, president and CEO of TeraRecon explained, "Success (or lack thereof) is directly tied to successful clinical implementation and adoption -- which relies on the development of valuable clinical use cases that AI innovation is purpose-built to solve."
According to Elad Walach, CEO of Aidoc, "AI companies should focus on showing their value as mainstream enterprise software solutions and start acting like this -- with true customer-centric methodologies, like user engagements or customer satisfaction, and business-level discussions." He goes on to say, "Algorithm accuracy is important, but there's a lot more to consider when thinking about first productizing and then commercially selling an AI solution."
So what's important? With so many AI vendors, the logical question is what differentiates one AI company from another. It certainly isn't the company name, as no fewer than eight companies had the word "deep" as part of their corporate identity, while other companies had product or corporate names that had no relation to AI whatsoever.
Walach explained what he felt was important, "I would put an emphasis on (1) workflow integration from the radiologist perspective, (2) comprehensiveness of the suite, (3) ease and maturity of integration model, and (4) a flexible AI partnership approach."
Saragense said, "To be successful, I believe that these vendors must focus on AI-enabled solutions that solve deep clinical problems. ... To truly relieve the pressure on healthcare, any new AI-enabled solutions must augment radiologists and care providers and allow them to make critical decisions quickly, correctly, and that saves lives."
Panetr believes a company needs to understand the commercial realities of the medical imaging market in the territory they are trying to address. For example, tools for radiology productivity matter hugely to an individual radiologist and the group they read in, but they are less important to a traditional hospital that simply contracts with the radiology group.
Who is AI really for?
That brings up the topic of who is really the target market for AI. As you walk through the halls of RSNA 2019, most of the messaging is radiologist-centric. Considering this is a radiology show, that messaging is to be expected. Yet for AI to be completely accepted, the messaging has to resonate as much for the hospital, which no doubt will be making a major financial investment in AI. No vendor that I saw either directly or indirectly addressed the needs of the hospital.
Interestingly, the largest push back against AI has been from radiologists who fear the impact it will have on not just the performance of radiologic interpretations but on job security as well. Blackford's Panter believes that AI solutions need to fit within the radiologist's workflow and provide real value for all stakeholders -- the radiologist, the referrer, and the radiology business manager. Axerio-Cilies of Arterys recommends that AI vendors deliver full clinical integration, while Walach of Aidoc believes vendors must show workflow integration and clear value.
So what will RSNA 2020 and beyond bring in terms of AI vendors? A market shake-up is definitely in the works and the market simply isn't big enough to accommodate the number of vendors offering products today. Nearly all the major modality vendors also seem to have one form or another of an AI solution either as a standalone offering or integrated with their modality offerings, yet only a few were represented in the AI Showcase at RSNA 2019.
While most OEMs have elected to display their AI solutions on the main show floor alongside their other products, nearly all used terms like "AI-enabled," "AI-powered," or similar adjectives that reference AI. Companies like Microsoft, Google, and Amazon have also chosen to use relationships with AI partners to promote their products and take a partner-led strategy rather than to act as a reseller.
TeraRecon's Sorenson was concise in his evaluation of the marketplace.
"A shakeout of sorts has already begun to hit the market," he said. "The hype of a few years ago is now being replaced by a demand for clinical domain expertise."
Zebra's Gura agreed with that assessment.
"We already saw in the last few months the initial wave of shakeout and this will continue in the next 24 months," Gura said. "A rollup is also in the horizon."
Michael J. Cannavo is known industry-wide as the PACSman. After several decades as an independent PACS consultant, he worked as both a strategic accounts manager and solutions architect with two major PACS vendors. He has now made it back safely from the dark side and is sharing his observations.
His healthcare consulting services for end users include PACS optimization services, system upgrade and proposal reviews, contract reviews, and other areas. The PACSman is also working with imaging and IT vendors developing market-focused messaging as well as sales training programs. He can be reached at firstname.lastname@example.org or by phone at 407-359-0191.
The comments and observations expressed are those of the author and do not necessarily reflect the opinions of AuntMinnie.com.
Copyright © 2019 AuntMinnie.com