Tag Archives: Machine Learning

Signify Premium Insight: Risk and Reward – The Maturation of Medical Imaging AI

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here.

Earlier this month, Viz.ai announced that it had received US-FDA clearance for its automated right ventricle/left ventricle (RV/LV) ratio algorithm, a new component of the vendor’s Pulmonary Embolism (PE) solution. The RV/LV algorithm will enable the automated assessment of potential right ventricle dilation and therefore help to identify right ventricular dysfunction, before delivering the results quickly to the entire care team using Viz’s PE solution.

The move represents the latest FDA clearance for Viz.ai, as it continues to grow its care coordination platform and expand beyond its original stroke care remit. The move also highlights a growing trend in medical imaging AI of vendors expanding product portfolios beyond a single use case, and also beyond image analysis.

The Signify View

As medical imaging AI vendors have matured and proved themselves worthy of increasingly lofty funding rounds, companies are having to expand beyond their original briefs to continue to provide value to the doctors that use them. Some of the most successful vendors have sought to offer this increased value by adding additional capabilities along the care pathway, beyond the slice of the workflow devoted to image analysis itself.

In the case of Viz.ai, this originally meant expanding into elements of stroke care such as triage and decision support, with the vendor’s care coordination platform aiming to expedite the treatment of the most urgent cases. Latterly, instead of expanding along the care pathway, vendors have been looking to leverage their expertise more broadly, with Viz, expanding into other vascular conditions.

For Viz, and other vendors, the key to adoption isn’t just about  the detection algorithms themselves. While their effectiveness is important, slight variances in specificity and sensitivity between vendors won’t make or break a provider’s decision to go ahead and make a purchase –  instead the value comes from the care coordination platform and the value that an AI developer can offer across the whole workflow. This is key as they translate their expertise into other areas. There may be niche vendors with slightly more performant algorithms in certain specific tasks, but these vendors will not be able to match the value brought about by a complete care coordination platform.

There are risks to this approach, however. Viz.ai, and other peers adopting a similar strategy such as Aidoc, and some Chinese vendors have been able to raise considerable amounts of money by advancing into new clinical areas and broadening their product portfolio. While such moves give them a head start over some more specialist vendors, they may also risk spreading themselves too thinly, stymieing their ability to fully deliver on their promises in the areas they first gained success.

Better Together?

Some vendors are forging partnerships to mitigate this exposure. Aidoc, for example, has chosen to add quantification capabilities to both its stroke care and pulmonary embolism solutions by looking externally. Aidoc’s own detect and triage capabilities are bolstered by a perfusion solution from I cometrix for stroke, and RV/LV solution from Imbio for its pulmonary embolism solution. This has allowed Aidoc to strengthen its care coordination platform, bringing quantification and stratification tools to market, while its partner gains access to many of Aidoc’s sites, giving the vendor significant potential upsell opportunities.

Unlike Aidoc, Viz developed the entirety of its stroke care platform in-house. However, for its pulmonary embolism solution, it also turned to a partner, forging links with Avicenna.ai to deliver the detect and triage capabilities for pulmonary embolism. While such a move will see the vendor relinquish some control, partnership offers a significantly expedited rollout. Rather than starting from scratch, having to develop a solution and conduct clinical validation studies over multiple years, a timespan that could result in the vendor losing ground to competitors.

Adopting such a strategy also requires Viz to further develop a back-end architecture for the native and partner algorithms to work seamlessly together, a move which could see the vendor follow in the footsteps of Aidoc and herald the commercial launch of an integrated AI platform.

The Importance of Being Useful

Regardless of the specifics surrounding vendors’ expansions into other clinical areas, be it Viz or any other AI vendor, the approach of leveraging triage and stratification tools is significant. For instance, it highlights that instead of being content with offering tools only useful for image analysis in other clinical areas, developing fully fledged care coordination platforms to serve other clinical situations is now a clear priority. Whether the actual image analysis part of that solution is developed internally, or offered via a partnership is fast becoming immaterial, as the real value of such solutions doesn’t stem from image analysis itself. Instead, in many cases, providers will benefit from leading AI vendors’ abilities to bring imaging analysis algorithms into a considered workflow, to increase their utility.

Some tools, also confer other advantages. Triage tools for example, have a simpler regulatory pathway than CADe or CADx image analysis algorithms, which, are seen to harbour more potential for patient harm. This can offer vendors a more efficient route to market. While the products they will be able to sell as a result of the approval may be more limited compared to solutions cleared for diagnostic use, such clearances will at least enable vendors to begin generating revenue and launch commercially in new markets, offering them a foundation to build on.

More broadly the expansion of some of medical imaging AI’s largest vendors into wider clinical areas, seeing them apply their expertise into more diverse use cases represents the growing maturation of medical imaging AI vendors.

Remember the Objectives

The ultimate aim of medical imaging AI is not to shave seconds of the read time of a chest X-ray, for example or even identify the presence of an indicator of a clinical condition. It is, above all else intended to improve patient outcomes; a final result that is based on the totality of a patient’s care, along their entire care journey.

The portion of this journey that actually entails the analysis of medical images is small. As such, although image analysis is the use case for AI that is discussed most excitedly, there are opportunities elsewhere along the care pathway that can have a more substantial impact on patients’ eventual outcomes. The addition of risk stratification tools such as the RV/LV algorithm from Viz epitomises this.

The vendor’s USP has long been to apply its expertise beyond the image analysis portion of the workflow with its care coordination platform. Not only does this deliver the assistance to identify findings from medical images, but it also helps imaging departments, and other departments more broadly, to better manage patient care and make interventions earlier. Compared to the relatively slight impact that shaving a few seconds off a read time can have for a provider, even for high read volume applications, the use of AI in this broader way can be far more significant.

Further, this offers a more sophisticated method of identifying the leaders in the medical imaging AI market compared with simply looking at which vendor has the greatest number of FDA cleared algorithms, or which has been able to raise the most capital. Instead, it is increasingly possible to assess vendors based on how sophisticated their tools are, and how much value they can offer providers. There is no single, solitary route to adding this value, with comprehensive solutions, and some sophisticated point solutions, which alter the diagnostic pathway, also offering broader value to providers alongside some vendors’ expansion into additional clinical areas, and along care pathways (end-to-end solutions as previously termed by Signify Research).

In this regard, broader imaging IT vendors have an advantage. With large installed bases and their presence across radiology departments and beyond, these vendors, with the right tools, could alleviate many of the bottlenecks faced by providers. However, at present these vendors aren’t aggressively leveraging this advantage, leaving the likes of Viz and its peers to make the early headway.

Whether they are able to capitalise long-term remains to be seen, but for now at least, moves such as that made by Viz, and some of its peers, show the maturation of medical imaging AI away from a “one-trick” image analysis focus toward impactful care outcomes.

About Signify Premium Insights

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here

Signify Premium Insight: AI Making the Move to Maturity

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here.

Dr Sanjay Parekh, Senior Analyst

The medical imaging AI market is among the most dynamic of all the sectors in medical imaging. Its nascency, its rate of technical development and the application of the technology are combining to create a market that is changing incredibly quickly.

Despite the volatility of the market, senior analyst and author of Signify Research’s AI in Medical Imaging report, Dr Sanjay Parekh, has been able to discern several key trends in the market.

Great Growth

“The market for AI-based image analysis tools for medical imaging is set to reach $1.36bn by 2026,” Parekh states, “up from $402m in 2021.” Much of this revenue is for stand-alone AI tools, but AI-based advanced visualisation (AV) bundled AI tools are also included (accounting for 27% of the total market in 2021).

This represents a CAGR of 27% between 2020 and 2026, highlighting that the AI market is gaining momentum, many of the reasons for which are clear.

“There was for instance a large flurry of regulatory approvals in 2020 and 2021. In the US in 2020 for example, there were more approvals than in 2018 and 2019 combined. There was also the first wave of NMPA Class III approvals in China. With these regulatory approvals vendors can commercialise their tools.

“As well as having more products on the market, there has been continued progress with regards to reimbursement. There is the continued reimbursement for HeartFlow’s FFR-CT solution in the US, the UK and Japan, as well as parts of Europe. Additionally, pockets of China are already reimbursing the use of FFR-CT tools, but national reimbursement is still pending. There has also been a flurry of Category III CPT codes [for which there is no compulsory reimbursement] provisioned for quantitative image analysis tools for ultrasound, MRI and CT, which as well as encouraging the uptake of AI, could lead to reimbursement. While NTAP payments have also been renewed and expanded, such as the recent Optellum clearance, which has defied the norm and will now receive reimbursement for its Virtual Nodule Clinic solution for lung cancer despite the CPT code remaining as Category III.

“All of these factors combined will help the markets continue to grow.”

Areas of Interest

Growth will not be equal across all clinical segments, however, with four areas, which currently represent around 87% of the market, set to continue to stand out. These are cardiology, neurology, pulmonology, and breast imaging, with each having facets that mean they are likely to continue powering growth for AI vendors.

Use of AI is the most mature in the breast imaging market; however, opportunities for growth are more limited than elsewhere. “Because of the relatively limited number of use cases; namely breast nodule detections and breast density analysis, the breast imaging market will not to be as large as the other three,” continues Parekh.

“Cardiology is likely to account for the largest proportion. This will be driven by two factors. The first is increasing uptake and continued reimbursement for FFR-CT tools. Even accounting for its failed SPAC merger, HeartFlow, one of the success stories for the medical imaging AI market, has a relatively large install base and strong commercial traction as well as still offering an appealing value proposition. There are also opportunities for FFR-CT, especially in China, as vendors like Keya Medical, Shukun Technologies, and Raysight receiving regulatory approval for their FFR-CT tools.

“In addition, clinical guidelines recommending CT imaging as a first line diagnostic procedure will drive the adoption of AI.”

Stroke care is also set to rally.

Neurology will be a growth area mainly because of stroke imaging AI solutions. The NTAP code for stroke LVO, which was first issued in 2020 to Viz.ai and then renewed in 2021, was renewed again in 2022 and it looks set to be made permanent soon, thanks to the uptake of stroke imaging AI tools and the increased use of the code in such instances.

“Not only has the payment been created, but providers are using it and its use shows that providers value the end-to-end stroke solutions which benefit the entire care pathway as well as the radiologist.”

There are also other opportunities within neurology, with brain quantification tools, for example achieving moderate success. Some vendors offering such tools are generating revenue, but, while these will continue to be valued, other drivers such as the commercialisation of drugs for neurodegenerative disease are needed before they become a major driver of growth.

“Finally, in pulmonology, the relative value of using AI market is smaller compared to FFR-CT or head CT for example. Although there are vendors working on comprehensive solutions for both chest X-ray and chest CT that do restore that value, the most successful among them are setting a benchmark for other tools looking to gain traction. Further, the continued roll-out of screening programmes for lung cancer and TB, for example, will drive further traction in this market.”

Relinquishing a Point

There are commonalities across these clinical areas, however. It is becoming clear that the utility of point solutions across modalities and clinical areas is in general, very limited. Developers who can only offer single point solutions are looking increasingly unlikely to be selected by providers.

Instead, tools that offer the most value to providers will gain success. This value, however, can manifest in various ways. Many solutions focus on efficiency, but there are also solutions that could actually slow diagnoses, but still enhance the quality of a diagnosis by offering additional metrics, for example. This value is, in some cases, also no longer derived from incremental improvements in specificity or sensitivity that new tools might offer.

“If you offer 93% accuracy compared to 92%, is that going to make a difference,” Parekh asks. “Are you going to get a better diagnosis or is the patient going to be on a completely different treatment pathway? No. Instead value is extended beyond the analysis of the pixels in an image, to patient care and improvements to the clinical care pathway. The vendors that have started doing that are the ones that are going to succeed.

“Breast imaging tools, for example, that combine detection, quantification and classification of nodules, which are far more valuable than those which only offer nodule detection. Moreover, adding in breast density analysis will enhance the value proposition of such a solution even further. More significantly, however, are the tools that are looking at radiology more broadly and seen to offer value across the clinical care pathway (beyond the radiologist). These solutions can come from vendors which solely offer AI, or those which also offer capability to deploy and integrate AI.

“These vendors can bring in advanced visualisation capabilities, workflow capabilities and even structured reporting capabilities to address a given use case, while also offering their own native or third-party AI image analysis capabilities to create entire workflow packages. That is AI demonstrating value.”

Money to Money

Value is also forthcoming in a broader sense. Despite the turbulence in some tech markets and in some corners of the medical imaging market, investment for medical imaging AI vendors is still available, although it is becoming more discerning.

“Investors seem to be more than willing to continue to back vendors that have already shown progress,” opines Parekh, “but we are not seeing many Angel or Series A rounds.”

“Where we are seeing a lot of action is for the later-stage funding rounds, which are increasing in both size and number. This indicates that a set of market leaders are being established, such as the $100m funding club [a term coined by Signify Research including vendors that have received more than $100 million in total in venture capital funding]. Even with this greater investment in established companies we are starting to see evidence of a market shakeout.

“Last year we saw Nanox acquire Zebra Medical Vision, at the start of 2022 we have seen MaxQ-AI closing its radiology business, and Sirona acquiring Nines. RadNet, a large outpatient imaging group in the US also acquired two Dutch-based AI start-ups Aidence and Quantib to add to its portfolio after previously acquiring DeepHealth, and expand its push to deploy AI across screening for some of the most prevalent cancers. There is also some speculation about some other vendors also making pivots after not receiving funding that was expected. We have seen consolidation coming for a long time, but between the investment being focused on the largest vendors, and the difficulties for the smaller vendors, we are starting to see the shakeout take place.”

The impact of this market shakeout will be different in different regions. One area that is more difficult to make predictions for is Europe. Presently, the Western European market is starting to catch up with the US, but this growth is expected to stagnate in May 2024 when the new European Union Medical Device Regulation (EU MDR) is coming into force. There is currently a backlog of 12 to 18 months for vendors to upgrade their CE Mark to the incoming regulation, not to mention the more stringent requirements for this regulation. This raises the possibility of many vendors missing the deadline and therefore being unable to offer their products commercially in the EU.

Approval Ratings

This could have significant impacts, say Parekh.

“It is more likely that the larger vendors, the ones with the funds to pursue the MDR, will be the first to receive it. If you are a smaller vendor, then you may not want to, or be able to go for MDR approval. Ultimately, that will leave those that have MDR certification by May 2024 with an ‘early-to-market’ advantage over those that don’t. It could effectively level the playing field, and serve as a reset button, with only those that have been able to secure the new certification, regardless of past CE Mark approvals. This regulatory backlog is also therefore likely to hold back the market as a whole.

“It could also lead a lack of innovation, with smaller start-ups and research groups shifting their focus from radiology, keen to avoid the additional barriers they must pass, so there could also be a short-term innovation gap. This is another reason we could see more consolidation in the market.”

Despite these challenges the future is still bright for medical imaging AI vendors. The market increased by more than $60 million between 2020 and 2021, and growth is only set to continue. This shows a young market taking the first steps to maturity and a nascent technology making the first moves toward more mainstream adoption.

“Overall,” Parekh concludes, “it is growing, at a steady pace for now but with a big ramp up in the medium term, from 2024 onwards.”

“All signs are positive.”

About Signify Premium Insights

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here

Signify Premium Insight: VC’s $3.6bn in a Quickly Maturing Market

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here.

Co-written by Dr. Sanjay Parekh

In our recently released analysis of venture capital funding for medical imaging AI vendors (download free report here – note this report was published prior to Viz.ai’s very recent $100m series D funding round), Signify Research found that venture capitalists’ appetite for medical imaging AI is yet to be satisfied. Total VC investment in medical imaging AI has reached almost $3.6bn since 2015. What’s more, despite a dip in 2019, overall levels of funding have continued to rise making 2021 a record year in terms of investment raised at $815m.

This record-breaking year also saw the emergence of the ‘$100m club’ of vendors which have raised more than $100m in capital funding. Many of the vendors also highlight another trend detailed in the report; the eastward transition of funding, with companies from Asia and especially China accounting for an increasing and significant proportion of investment.

In this more nuanced analysis for Premium Insights clients, we dig deeper into these findings.

The Signify View

These headline figures only tell half the story, with large numbers of vendors choosing not to reveal the levels of funding they have secured. Circle Cardiovascular Imaging, for example, has reported raising almost $20m in funds, however, their latest undisclosed funding is estimated to be in the range of $75m-$100m, pushing the Canadian vendor’s total funding above $100m. This is a trend that is mirrored across the world. China’s Infervision raised $70m in series B funding, and $140m in series D funding, but did not disclose the results of its series C round. If this undisclosed round is taken into account, it is likely that the vendor has raised upwards of $300m in total, far higher than the $225m disclosed. Other vendors also share this dynamic, with Shukun Technology and Deepwise both harbouring far higher sums than publicly disclosed.

These high figures show the strength of these vendors and the emergence of prominent market leaders. Already members of the ‘$100m club’, these vendors are sitting on cash piles while also having NMPA Class III approvals, allowing commercial operation within China. This makes them formidable competition in the market, especially given that their lack of disclosure makes them difficult to accurately assess and limiting most international vendors from targeting the Chinese market.

Had these rounds been disclosed, we would be discussing a group of unicorns (those estimated to be worth more than $1 billion). However, other vendors have been more transparent. Viz.ai is an example of one vendor that can lay claims to this status, having confirmed its series D funding round of $100m itself earlier this week. This funding round took the total the company has raised to over $250 million, valuing it at $1.2 billion.

Chinese Action

More broadly, these funding rounds for Chinese vendors evidence the trend identified in Signify Research’s 2021 report, which showed that China is increasingly the hotbed of medical imaging AI VC funding. This is expected to continue, with 2022 likely to surpass other years as more vendors mature and secure larger, later-stage funding rounds. As detailed in a recent Premium Insight which addressed trends in regulatory clearances, more solutions are receiving Class III NMPA approval in China, a factor that is likely to encourage investment, with approved solutions often a more attractive investment target than those with no approvals. This growth in funding will also be compounded by the possibility that the Chinese market offers. As well as its insular nature, which makes penetration difficult for non-Chinese vendors, limiting competition, the market is also the world’s second-largest.

AI uptake has, so far, been very regionalised, with vendors targeting specific patient cohorts in certain areas. In China, where there are 15 provinces with populations of more than 40 million, sub-regionalisation is likely, with vendors targeting specific diseases based on provincial priorities, providing VC investors with plenty of viable targets, even if they are, individually, unlikely to become global powerhouses.

However, investment will peak, potentially as soon as 2022. In the US, VC investment peaked in 2018, at over $400m, after which it slipped backwards, plateauing around $150m. This is a result of market leaders emerging, securing later stage funding, increasingly growing revenues, and then receiving undisclosed private equity investments or attempting to list publicly. The Chinese market is approaching a similar scenario, with market leaders being established and some reaching the end of their VC funding journeys.

Lapping the Shore

These dynamics in funding, show that so far, there have been two waves of funding for medical imaging AI. The first peaked in 2018 (total funding of $772m; average deal size of $16.4m), but the second peak in 2021 was far greater (total funding of $875m) with an average deal size more than double ($33.6m) that of 2018.

The quantity of deals was higher in 2018 (47 deals), as a higher number of vendors looked to secure their initial rounds of funding, fostering the creation of the market for radiology AI. Since then, some tools have matured, and in a bid to increasingly add value to clinicians, become more sophisticated. The vendors who have developed these tools have grown and are facing significant costs in the commercialisation of their products. Resultantly, they have needed to raise more capital to fund this expansion and commercialising, spurring the recent second wave.

One indicator of this current wave is the growth in the number of vendors entering the ‘$100m club’. As vendors have matured and sought out later-stage rounds, the number of deals has declined (26 deals in 2021). However, these deals are frequently for very sizable amounts, taking increasing numbers of vendors past the $100m mark (ten vendors to date), a barrier that, until recently had only been broken by a very select few vendors, which met a very demanding criteria.

Vendors that have entered the ‘$100m Club’ of VC funding raised

These dynamics are, of course, not immune to wider influences, with the Covid-19 pandemic also playing its part. Part of the second peak rebound in 2021 was due to the market emerging from the pandemic, and investors were once again able to confidently invest in medical imaging AI start-ups. After husbanding cash throughout the pandemic VC investors could once again seek opportunities and take on more risk. Healthcare technology would also have been an attractive sector in which to invest, given the Covid-induced excitement around digital healthcare. Investors, with available cash and looking for companies in which to invest, bought into the story of AI helping radiologists become more efficient and deal with the backlogs they were left with. The pandemic had created a problem, which radiology AI could aid in solving.

The Looming Threat

These trends could contribute to the growing spectre of consolidation that is hanging over the medical imaging market, with a distinct gap between the minority of vendors that have secured sizable funding rounds and the long tail of vendors which have not. Since 2015, the top 10 vendors have, after all, raised an overwhelming 55% of all funding, while the top 25 vendors account for almost 80% of funding. These larger, better-funded vendors look set to increasingly take control of their specific market segments, making life difficult for smaller vendors to gain any traction and increasing the risk of casualties in this market.

Life could yet be difficult for larger vendors too, though, with higher VC investment rounds representing greater pressure on these vendors to start delivering the sizable revenues that they have promised. Reimbursement for medical imaging AI also remains limited and sporadic; the question of who will pay for AI, in many instances, is still unanswered. This may prove more of a challenge than expected, leaving many investors disappointed.

As these problems are worked through, vendors will go on to list publicly, as some including Keya Medical and HeartFlow have already unsuccessfully attempted; some may also be bought out by private equity, giving early VC investors the sizable returns they were hoping for. As this happens, there could be another wave of VC investment in medical imaging AI, given the technology’s transformative potential. However, the next cohort of start-ups backed by VC funding is unlikely to be in any of the established clinical segments (e.g., breast imaging, neuroradiology, cardiac imaging, chest imaging) as the opportunity for a start-up to break into these well-catered segments is negligible. This future wave of funding, as VC firms look to find the next stalwarts of medical imaging AI, could catalyse interest in tools targeting different clinical specialities or different regions. Tools for liver imaging, for example, could blossom, as drugs to manage fatty liver disease become increasingly available and backing from VC firms looking to capitalise is forthcoming. Additionally, the use of MR imaging as a first-line diagnostic imaging procedure for prostate cancer may create an opportunity for vendors with prostate imaging AI tools, a niche segment targeted by advanced visualisation incumbents today. The small wave of regulatory approvals for prostate AI solutions in the past 12-18 months may be indicative of this trend.

In the more immediate future however, funding in the US and Europe should continue its plateau, while funding in Asia is expected to continue to grow, with the country accounting for an ever-greater share of imaging AI funding. The size of these deals is also set to grow, albeit there will be fewer of them, as VC investment continues assist in the forging of medical imaging vendors. From this, vendors will be able to establish clear market leadership positions, and, assuming vendors live up to their own ambition drive significant revenues. At this point, the VC’s job will be complete, and the medical imaging AI market, established.

About Signify Premium Insights

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here 

Signify Premium Insight: The Key Trends of RSNA 2021: Medical Imaging AI

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here.

Co-written by Alan Stoddart.

As highlighted in our other RSNA previews (Imaging IT, Medical Imaging Modalities), this year’s meeting looks set to be primarily focused on solutions that improve efficiency and productivity. The ongoing coronavirus pandemic has stretched providers, forcing them to redirect resources to certain departments during the pandemic and postponing elective surgeries. Now, these providers will have to deal with this backlog of surgeries and the imaging exams that they will entail. This, as well as the ever-present need for providers to effectively manage budgets and maximise the productivity of their stretched radiologists is set to shape the show’s focus for many. AI vendors will be keen to illustrate how they can help.

The Signify View

While most medical imaging markets are well established and relatively stable, the medical imaging AI market is still young and changing rapidly. As such, vendors’ offerings are developing, use cases are evolving and, significantly, customer requirements are maturing. Just a few short years ago, RSNA would have provided an introduction to AI for many radiologists. The meeting, and the plethora of AI start-ups it hosted, would have presented an exciting new technology and all the potential it entailed to a curious, if not yet entirely convinced, audience. That has changed. No longer content with being shown what AI might be able to do in the future, providers will this year attend RSNA with specific problems that they are looking to AI vendors to solve.

That being the case, these vendors will be exhibiting solutions that aren’t merely commercialisations of existing technologies but are instead tools much more focused on specific challenges and bottlenecks in clinical workflows. This has, so far, most commonly been achieved by developers expanding beyond narrow AI algorithms that are focused on one single, specific task, and offering broader solutions that offer providers more value. Some of these are comprehensive AI solutions, a topic discussed in a previous Premium Insight. These solutions overcome the inefficiencies of point solutions which despite being able to accurately identify a single finding, often don’t make a material difference to workflow efficiency. Other vendors will be promoting tools that extend beyond the core image analysis capability into other areas such as workflow and case management. These tools could be designed for very different scenarios, from screening programmes to critical conditions such as stroke or pulmonary embolism, but all focus on utilising AI to offer a clinical benefit to patients, rather than just giving a radiologist additional data or shaving a few seconds from read times.

Making Introductions

One of the key uses for RSNA’s annual meeting is networking. This will be among the priorities for many AI developers, who will look to form partnerships with other vendors to be able to offer providers more compelling packages. Developers want to be able to offer tools that complement their own, and offer a fuller toolkit that brings providers a more rounded solution. These additional tools would be too onerous for a small vendor to develop in-house, so joining with another vendor who already has that capability makes sense.

There are also other reasons that vendors will be looking to forge partnerships at RSNA. In addition to improving a developer’s portfolio, partnerships could facilitate better distribution channels to help vendors grow in different markets. Alternatively, if a smaller, unknown developer partners with one that is larger and better established, then the smaller vendor stands to gain enhanced credibility and visibility.

Although the trend towards partnerships has been growing over recent years, the coronavirus pandemic and the disruption to providers and vendors will have accelerated the trend. Vendors will, in some instances, have been forced to reassess their priorities and acknowledge that a more specialised approach is better for them, targeting a particular modality, or body area, for example, rather than trying to offer a gamut of solutions. This re-evaluation will hasten the move towards partnerships, as vendors, who are looking to grow, will be reluctant to spread themselves too thinly trying to compete with an established leader in another area, when they could instead look to partner.

This direction is likely to also result in a greater focus on AI platforms at RSNA, with some new platform launches likely. This could be driven by larger AI vendors that need to bring a number of separate algorithms together to improve their accessibility. However, these platforms are also likely to be promoted by AI vendors who have worked with other partner vendors to create a more encompassing toolset. While their partnerships will have bestowed them with expanded capability, these vendors still require a robust way of seamlessly integrating these disparate tools into providers’ workflows. Such platforms also provide an increasingly important competitive differentiator. For smaller vendors forging partnerships, they are an attractive strategy, for the established imaging IT vendors however these platforms have become an expectation. As such, RSNA could witness announcements from any major vendors which do not yet offer a platform or marketplace for incorporating AI, and do not wish to fall behind the competition. For example, Philips is expected to showcase its AI platform and marketplace at this year’s meeting. We may also see some of the specialist AI vendors reposition themselves as platform providers, taking on the established imaging IT vendors.

AI’s Changing Role

Alongside the evolution of medical imaging AI vendors, and their increasingly frequent transition away from a broad developer to a more specialised partner, AI applications themselves are also evolving. This will also be seen at RSNA. Previously, much of the excitement around AI was centred on image analysis and patient diagnosis. While there are still many vendors which aspire to accomplish this ‘radiologist work’, there are other tasks to which, particularly in the near term, AI may be better used. There are solutions that have proved successful in imitating a radiologist’s performance in some studies and under very specific and limited conditions, but, for the most part they still don’t compare well to a radiologist.

What’s more there are tasks for which AI could be more suitable. Instead of developers promoting tools that attempt to replicate a radiologist’s core diagnostic role, tools which can enhance a radiologist by automating workflow tasks, such as feature quantification, are likely to be of more use. Afterall, hospitals already struggling from a shortage of radiologists are having to deal with an enormous backlog of patients who were unable to receive scans and treatment last year because of the COVID-19 pandemic. In this situation, tools that can enable radiologists to work more efficiently will be warmly received. If many of the simple, yet time consuming, workflow tasks can be accomplished by AI tools, a radiologist will have more time to devote to the high value tasks which make a clinical difference.

AI’s Changing Face

More generally, RSNA will also reflect the developing identity of AI in medical imaging. This year’s show is set to be different to those of previous years because of the disruption caused by the pandemic. Despite the show taking place in Chicago, ongoing travel restrictions in many countries and at many companies mean that the show will still have a large virtual element. In addition to this there are likely to be several vendors that will eschew the cost of a booth, and instead walk the show, holding more targeted meetings. These changes mean that the presence of AI developers at this year’s meeting will not be directly comparable to that of previous years, but should serve as an interesting barometer nonetheless.

The move towards consolidation in the AI market, along with the difficulties faced by small vendors and start-ups in particular, means that there is likely to be a smaller number of AI vendors in attendance. Since the last show in Chicago, some of the smaller vendors with limited point solutions will have struggled to secure sales, engagement in pilot studies and funding, and will have lost out to other larger competitors. Another indicator of change on display at RSNA will be in the positioning of some AI vendors. Where previously an AI developer may have been among its peers in the AI Showcase section of the show, this year, the more successful, better established AI vendors could move elsewhere. Subtly refocusing attention away from their AI technology, and onto the medical problems that they solve. The value of the solution, not the buzzworthy technology that powers it, will be the focus.

In such ways, this year’s RSNA conference will show a technology that has passed its honeymoon phase. Radiologists and providers are increasingly savvy about AI, and the use of technology alone is no longer worthy of note. RSNA will epitomise this. The vendors enjoying the most success at the show will be those that have a clear strategy, whether alone or with partners, for solving clinical problems using their technology. Providers will want solutions that can materially impact patient outcomes, demonstrably improve hospital efficiency and provide an evident return on investment. The AI developers that do well at RSNA will be those which can meet these clear needs.

About Signify Premium Insights

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. This content is only available to individuals with an active account for this paid-for service and is the copyright of Signify Research. Content cannot be shared or distributed to non-subscribers or other third parties without express written consent from Signify Research. To view other recent Premium Insights that are part of the service please click here