Tag Archives: annalise

Signify Premium Insight: Annalise.ai Enters into Nuanced Partnership

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. The content is only available to companies that have subscribed to this paid-for service. To view other recent Premium Insights that are part of the service please click here.

Medical imaging AI vendor Annalise.ai and Nuance Communications, a vendor which specialises in reporting and ambient clinical intelligence tools, have announced a partnership which will connect Annalise.ai’s diagnostic support solutions to more than 12,000 healthcare facilities currently on Nuance’s Precision Imaging Network globally.

With the agreement, Annalise hopes to gain exposure to a greater number of sites, allowing it to scale rapidly, while Nuance can utilise Annalise’s solution to enhance its growing Clinical Analytics Platform and complement its Natural Language Processing tools.

The Signify View

Medical imaging AI vendors are keen to extol the virtues of their partnerships. While these vendors are often quick to boast that their algorithms are being hosted by one of a growing number of AI platforms, the truth is that these platform providers are sometimes not very discerning. Some platform providers aim to simply give customers the broadest range of solutions possible. Sometimes these are bundled into clinical suites or workflow packages, but the breadth of solutions on offer is usually of paramount importance.

The approach of Nuance, bolstered by its recent acquisition by Microsoft, is subtly different. The partnerships it has fostered do help offer a range of capability to customers, but above that ambition, Nuance has been more discerning, only partnering with vendors which deliver solutions that offer providers significant clinical value. It is essentially only interested in collaborating with the vendors it deems the leaders in any product category. This marks a divergence from its original platform play, which took the form of a more conventional ‘marketplace’ approach aiming to offer a wide variety of tools to the end-user, but that platform, like many of the early marketplaces, failed to gain significant traction.

Annalise.ai, as well as Nuance’s other announced partners, Densitas and Perspectum, embody this ‘quality over quantity’ approach. In the case of Annalise, which can be regarded as a market leader given the sophistication of its comprehensive solution, the clinical value it has the potential to offer and the funding and clearances it has secured, the adoption of a comprehensive solution eschews the need for Nuance to adopt and integrate solutions from multiple providers for the same body area modality combination. Nuance’s orchestration capabilities mean that customers on its Precision Imaging Network can leverage Annalise’s strength to identify a multitude of findings, before findings are pushed to their reporting solution, ensuring they can more readily be utilised in clinical workflows.

Historic Improvements

In addition to this, however, Annalise.ai’s solutions could be used in synergy with Nuance’s strength in natural language processing (NLP). Nuance’s NLP could mine historic radiology reports to identify reports of interest. These reports could then be analysed by Annalise to identify incidental findings. While this would, in the first instance, enable providers to improve patient outcomes, it would also have broader implications, allowing the health of entire populations to be more effectively managed overall.

As well as having a presence in almost 80% of US hospitals (according to the vendor) Nuance’s network connects radiologists, providers, health-plans, self-insured employers, life sciences companies and other imaging stakeholders. The two vendors will hope that this breadth will enable such retrospective analytics to deliver value to providers beyond the clinician, and identify other areas where additional value can be delivered.

This highlights the difference between Annalise and Nuance’s collaboration, compared to other comparable partnerships. Where often vendors in partnerships essentially co-exist harmoniously, Nuance and Annalise hope to collaborate synergistically. Working together they hope to enhance the quality of reporting and efficiently enrich the quality of reports with data directly from the algorithms.

Regulation Restrictions

Wider trends in the medical imaging market also emphasise the potential offered by the partnership. Annalise has, as noted in past Insights, been progressing quickly in Australasia and Europe. However, its progress in the US has been stymied by the US-FDA’s reluctance to approve comprehensive solutions, treating the detection of an individual finding as though it were assessing a separate tool. Such an approach effectively prevents Annalise, which claims its CXR chest X-ray solution can identify 124 findings, from gaining regulatory approval in the US. Resultantly, Annalise has, been forced to break up its solution in a bid to secure approval for smaller subsets of the solution. Further, to accelerate the pace of crossing regulatory hurdles and forge an installed base in the US, the vendor has also been forced to settle for its tool’s use as a triage and notification solution, rather than one that can be used for diagnosis.

These barriers mean that Annalise would be facing a long, hard road to gain ground in the US, especially in the face of other vendors which have gained success with a single solution before expanding out to encompass increased clinical requirements. Partnering with Nuance, and gaining access to its vast installed base, immediately ameliorates that difficulty. The scale of Nuance, as well as its integration into providers’ workflows, means that for the time being, the lack of regulatory approval for detection won’t severely hinder Annalise, enabling it to be valuable as just a triage solution, albeit for a smaller number of its CXR solutions. Further, if the US-FDA does eventually rethink its approach to comprehensive solutions, it will be well placed to dramatically capitalise.

Even at present, though, both companies stand to benefit, while also granting their customers new opportunities. This is particularly true given that Nuance’s workflow integrations will help tackle another of the hurdles facing providers hoping to utilise AI for historic analysis; how to bring the analysis of historic data into current clinical workflows. Annalise needs to be able to access the data harboured by Nuance’s 12,000 care facilities, which depends on that data not only being made available, but also being formatted into a unified manner, where NLP and image analysis can be leveraged.

Patient Finding

The fruits of overcoming this challenging, in private markets at least, can be substantial. Providers connected to Nuance’s network who choose to use Annalise’s solution on their historic data could identify significant numbers of patients with incidental findings, missed findings or even misdiagnosis. In doing so, if these patients can be incorporated into hospital’s workflows, and assigned treatment pathways, they represent additional sources of revenue for providers. By utilising the collaboration between Nuance and Annalise, providers should be able identify patients that will benefit from interventions, which they themselves can charge for, while also improving outcomes for the patient.

Further, the purported access to data granted by the agreement with Nuance will also give Annalise another longer-term advantage, with the vendor being able to utilise the data as it continues to refine its algorithms and presumably expand into other clinical areas, as well as validating its solutions to increasingly convince providers and regulators alike of its merits.

Even with the apparent strengths offered by the partnership, there are several questions whose answers will be revealed over time. How invested in medical imaging is Microsoft and Nuance, for example? One of the motivations driving investment in medical imaging by cloud infrastructure providers is simply to sell more cloud services. This is likely one of the reasons for Microsoft’s acquisition of Nuance in the first place. The partnership with Annalise and other AI vendors will, if successful, aid in this regard, helping convince providers to transition to the cloud. However, Nuance’s heritage and strategy suggests this is not the sole motivation. Another question raised is why Annalise hasn’t developed its own platform? AI scale-ups offering their own platforms is fast becoming a developing trend, and Annalise are well placed to make such a move. However, the opportunity to scale with Nuance is too significant to ignore, especially in the US, and Annalise will hope to use it to “leapfrog” algorithm developers that natively developed platforms.

These are, however, relatively small matters in what is a grander ambition. The volume of platform launches throughout the year has increased dramatically, but against this backdrop, Nuance’s partnerships with Annalise, Densitas, and Perspectum have brought something different. Sophisticated AI solutions, AI orchestration expertise, a large global footprint of potential sites, backed by a global cloud technology behemoth with very deep pockets; a combination which could prove a recipe for success.

About Signify Premium Insights

This Insight is part of your subscription to Signify Premium Insights – Medical Imaging. The content is only available to companies that have subscribed to this paid-for service. To view other recent Premium Insights that are part of the service please click here

Signify Premium Insight: Annalise Hoping to get Comprehensively Ahead

This Insight is part of your subscription to Signify Premium Insights – Medical ImagingThis 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 ResearchTo view other recent Premium Insights that are part of the service please click here.

Co-written by Dr. Sanjay Parekh

Annalise AI recently announced that it has launched an AI-powered decision support solution for non-contrast CT brain studies. The Australian company boasts that its new solution, dubbed Annalise CTB, is the most clinically comprehensive brain CT solution available, and can identify 130 findings.

The tool continues along the strategic path established by Annalise with its comprehensive chest X-ray solution, with a product strategy focused on detection of all radiologically relevant findings for a given body area or scan type, therefore more closely mimicking a radiologist’s process. As such, Annalise hopes its comprehensive solutions will be more clinically useful than those offered by many of its peers, who have tended to provide ‘point’ solutions for a single or small group of radiological findings.

The Signify View

When Annalise AI launched Annalise CXR, its chest X-ray solution, it immediately attracted the attention of all who study the medical imaging AI market. It did so because of the number 124; the number of findings the solution could detect. While there is some nuance, this figure was markedly higher than that of the most comparable vendors operating in the market. At 124, rather than the 10, 20 or even 50 findings claimed by competitors, Annalise made clear it was adopting a different approach to medical imaging AI.

Many vendors, particularly in the earlier days of medical imaging AI, were preoccupied with improving the sensitivity and specificity of the detection of a single radiological finding. While there are some scenarios in which such competent ‘point’ solutions are valuable, for the most part they offered only incremental gains compared to an unassisted radiologist, while frequently disrupting that radiologist’s workflow. Furthermore, the radiologist would often still need to thoroughly read a scan, to look for every other finding that the algorithm was not searching for. The benefits, in short, often failed to outweigh the drawbacks.

Various approaches to this challenge were adopted by vendors, many sought to expand their product’s utility along a clinical workflow, turning their algorithms into just one component of an expanded solution. Annalise, on the other hand, sought to expand its breadth of capabilities across multiple findings. Significantly, aiming to detect all possible findings for a single modality/body area combination. In doing so, the tool would more closely resemble the approach of radiologists. This would offer greater clinical value, also helping identify incidental findings and expediting the read for a radiologist.


As with Annalise’s CXR solution, this philosophy permeates the vendor’s head CT solution, which identifies numerous findings, including those related to the brain, such as brain bleeds and midline shifts, as well as to the head more broadly, with, for example, eye orbits and paranasal sinuses both assessed, as well as findings on the scalp and neck. This breadth expands upon the focus of many of Annalise’s competitors, which often only address findings for the brain. Annalise’s solution, which addresses a broader range, could therefore prove attractive to providers, particularly when some time-consuming reads, such as C-spine assessment, are considered.

Beyond tackling some of these exams that are less well catered for, the adoption of comprehensive solutions can also herald an approach more focused on population health. In mimicking a radiologist, comprehensive solutions can help avoid missing findings, particularly those that are not part of the primary diagnosis, enabling patients to be put on a treatment pathway for these incidentals as well as primary findings. In doing so, missed diagnoses or misdiagnoses can be reduced enabling patients to be treated sooner and outcomes to be improved.

While such population health advantages can be valuable, their impact will be most significant in single payer markets, where the payer and provider, represent the same entity. In such a system, regardless of where or when the patient continues treatment, any downstream savings made and any reduction in care costs over the longer term will ultimately benefit the same payer. Such an advantage cannot be conferred in predominantly private markets, where there is no guarantee that identifying additional findings in a scan will bring benefits to the same provider. Instead, allocating resource on an AI solution may only benefit a different  provider, where the patient eventually seeks treatment.

Another similar challenge in private markets will be in convincing providers to utilise comprehensive solutions. Although they may have some clinical advantages, for providers it is often advantageous to conduct, and therefore bill for, multiple specialist scans, rather than a single, comprehensive scan. As such, there may be limited motivation among providers to adopt such tools.

Regulatory Burden

Adoption of Annalise’s Head CT solution, along with other comprehensive tools, also faces another challenge in the world’s largest private healthcare market, the US: regulation. While Annalise’s solutions have received regulatory clearance in Europe and Australia as a single comprehensive tool, in the US, the FDA has held-out, insisting that each of a solution’s findings are treated as if a narrow, point solution. Each finding must, in effect, be treated as a single product.

The rationale behind such an approach is that each individual finding promised by a comprehensive solution should be subjected to the same regulatory rigour as a point solution, thereby ensuring that a comprehensive solution can demonstrably perform as effectively as an approved point solution in any single task. In Europe, conversely, comprehensive solutions have been regulatorily palatable provided they meet a minimum viable threshold.

There are merits to each of these approaches. The US FDA’s demanding criteria is, in the short term at least, arguably good for clinical practice. It ensures patient safety, minimising the opportunity for misdiagnosis, and prioritises patient outcomes. But, it is a regulatory framework that will stifle innovation, and in the longer-term prevent US patients benefiting from some tools. While these high barriers are appropriate in some cases, such as common findings where there are considerable training data, they severely hamper vendors’ abilities to address more obscure findings targeted in Annalise’s CTB, in the paranasal sinuses or in the pineal gland, for example.

There are some routes that purveyors of comprehensive algorithms for more obscure findings can take in the US. They could, for example, seek approval via the triage (CADt) rather than detection (CADe) regulatory pathway; a more straightforward route to market. Another option for vendors offering comprehensive solutions is to break up their offerings in the US, only offering the specific algorithms which have been approved. Both these approaches may help a vendor get a toe in the market, but neither are ideal, both potentially robbing solutions of their strengths.

Diverging Details

Despite these drawbacks, these are the concessions that Annalise is likely to have to make if it seeks to gain ground in the US. There is no reason to expect that either of the vendor’s solutions are to be any less well received than its chest X-ray solution has been in Europe and Australia. However, if it is to establish a footprint in the US, the vendor will have to take on the more time-consuming piecemeal approach to approval, beginning with the most clinically common solutions, before working its way through its broader array of findings.

The ramifications of such a requirement could, over time be significant. In Europe and Australia, comprehensive solutions could flourish, becoming providers’ preferred methods of AI adoption. In the US however, the FDA’s approach could mean that platforms which offer one or multiple suites for certain clinical use cases could become the norm. Enabling a range of vendors, each focused on their own regulatory challenges, to effectively be offered together through a platform to provide hospitals with a useful breadth of capability.

Although in some sense these platforms appear antithetical to Annalise’s comprehensive approach, they could themselves be an opportunity. Like Aidoc before it, Annalise may choose to offer a platform itself, including a version of CXR and CTB, which has been cut down to secure regulatory approval, alongside some other solutions from partner vendors. Over time though, as Annalise receives approval for more findings, and releases other products with different focuses based on modality, body area or clinical focus, it could incorporate them into its own platform displacing third parties. As such, it could adopt a gradual approach to the US market while capitalising on the different regulatory frameworks elsewhere.

As ever, there are varying routes to success, and in this nascent market, there is no certainty about which one is preferable. Other leading vendors have made their mark in their own unique ways, with, for example, Heartflow and Cleerly changing the diagnostic pathway for diagnosing patients with coronary artery disease, Viz.ai, RapidAI, and others altering care pathways for stroke care, and Perspectum helping reduce the need for liver biopsies. Similarly, Annalise, with its head CT solution, emphasises its intent to be the top comprehensive solution.

Other vendors are following a similar path, some with considerable advantages in other settings – one only need look at Lunit’s breadth of partners to see the esteem in which that firm is held – but Annalise has ensured that providers must at least consider its comprehensive potential.

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 ResearchTo view other recent Premium Insights that are part of the service please click here

Signify Premium Insight: Annalise.ai, Fujifilm and the Perils of Living on the Edge

This Insight is part of your subscription to Signify Premium Insights – Medical ImagingThis 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 ResearchTo view other recent Premium Insights that are part of the service please click here.

Co-written by Dr Sanjay Parekh & Bhvita Jani

Earlier this month Australian AI vendor Annalise.ai announced that it had partnered with Fujifilm Australia to offer its CXR Edge solution on the Japanese vendor’s portable and stationary X-ray machines.

The move will enable Fujifilm to distribute versions of Annalise’s comprehensive decision support solutions that are made to be embedded on medical imaging hardware in Australia, as well as in other select markets such as New Zealand, the United Kingdom and India.  Two variants of Annalise’s CXR Edge software will be available to healthcare providers. One, CXR Edge Comprehensive, can detect 95 clinical findings and is designed for use in inpatient, outpatient, and emergency settings. The second, Annalise CXR Edge Critical Care detects 35 findings, and is designed to be used in trauma, emergency, and intensive care settings.

The Signify View

Time is the enemy when it comes to many medical conditions. As such there is a great premium placed on devices and technologies which can save clinicians time and allow patients to be treated more quickly. While efficiency is an important consideration for all X-ray exams, it is often most important for those conducted using mobile systems, which are frequently used in emergency rooms and in other trauma settings. In these situations, doctors need to be alerted to time-critical conditions (e.g., pneumothorax) quickly, so embedding an AI solution to identify these conditions directly on the modality makes sense. In doing so, possible bottlenecks stemming from processing delays or network reliability issues are sidestepped, allowing doctors to see the results of the AI seconds after acquisition. What’s more, embedding AI tools on the modality itself also avoids the requirement for any additional IT infrastructure (e.g., dedicated servers), a factor that could be particularly beneficial in some emerging markets such as India in which Fujifilm has a strong presence.

Fujifilm is not the first medical imaging vendor to take this approach, GE Healthcare, for example originally released its Critical Care Suite in 2019 which also embedded medical imaging AI on the device. The key difference however are the tools’ remits. GE’s Clinical Care Suite is focused only a smaller number of findings (endotracheal tube placement, pneumothorax triage, quality assurance tools), while the Fujifilm and Annalise combination addresses up to 95 findings. This could prove far more attractive to providers. Mobile x-ray systems are often not dedicated to one specific department, instead being used for multi-disciplinary purposes amongst different clinical applications. Offering a wider range of solutions could therefore be a competitive advantage. This is especially true given the challenge of embedding multiple solutions from multiple vendors on the same device. It is likely unfeasible to have numerous separate solutions embedded on the modality, so a multipurpose, comprehensive system is arguably a more realistic way for providers to benefit from such a wide range of capability.

Patient Waiting

However, despite these advantages, AI solutions embedded on the scanner still have some drawbacks comparted to their PACS-deployed siblings. One of the main disadvantages is that they are unlikely to receive upgrades as frequently. Whereas a PACS-based system will usually receive both feature updates and refinements remotely, frequently bringing both new capabilities and improvements to AI solutions, edge AI solutions are more likely to be left running older versions. This is particularly true if the modality doesn’t have cloud connectivity. This could not only leave radiologists relying on outdated software, but also makes performance issues more likely. This is a particular issue given the nascency of AI adoption and the impressive rate of development.

While Fujifilm is initially offering Annalise’s CXR solution in Australia, it also has plans to offer it in other parts of the world, including India. Here, and in other emerging markets, this lack of updates is unlikely to be a significant issue. On the contrary, these are likely to be some of the best opportunities for embedded AI systems. Many emerging markets have, after all, a distinct shortage of radiologists, and increasing volumes of images. In some situations, having even an outdated version of a comprehensive solution embedded onto an X-ray system could flag critical cases and expedite treatment, rather than the images languishing until a radiologist is available to read them, possibly resulting in missed diagnoses. This is particularly true for the critical care version of the software, which, although detecting 35 findings compared to 95 for the comprehensive solution, is focused on detecting time-sensitive conditions that require immediate treatment.

Earlier is Better

Further, the use of embedded AI is also likely to bring financial benefits to providers. Early detection isn’t only good for patients, it will also facilitate the earlier treatment of patients, which in many cases will save a patient having to undergo more serious, and more expensive treatments.

This will be especially true in single-payor health systems where there is a greater incentive to reduce costs, but private markets such as the US could also benefit. Missed radiological findings can be very costly. Not only because of the added care costs, but also because of the costs of litigation. The fact that the embedded AI tools may help to note findings missed or seen at a more progressive stage by a radiologist, therefore have the potential to save providers considerable amounts of money lost in lawsuits.

Despite this potential, how willing providers will be to pay for such capability remains to be seen. This could depend on how Fujifilm chooses to offer embedded AI. The vendor could add the capability at no extra cost as a way to sweeten a deal and encourage a hardware purchase, alternatively, Fujifilm may choose to only offer it on its highest-end systems in a bid to differentiate the systems in its lineup and encourage customers to spring for the more expensive, higher-tiered systems. Additionally, Fujifilm may also consider upselling these edge solutions across its install base, generating a new source of revenue and enticing potential customers to its AI-enabled premium fleet when their contracts are up for renewal.

Taking the Subscription

The Japanese vendor may also explore some more innovative options such as ongoing subscriptions. Such deals could result in more of a partnership between Fujifilm, Annalise.ai and the customer, ensuring that providers have the support they need to maximise the value they are deriving from the systems. This could also benefit the vendors by turning transactional hardware sales into sticky subscription revenues. It would also overcome the issues of upgrades and would ensure that customers continually benefit from the latest version of the software.

Such a move would be particularly valuable at present. Fujifilm, as with other vendors which offer mobile X-ray systems, capitalised on the demand for the modality during the Covid-19 pandemic. As detailed in Signify Research’s General Radiography and Fluoroscopy Equipment report, global revenues derived from mobile digital radiography systems increased to almost $750m in 2020, up from $413m in 2019 as a result of pandemic-induced demand. Fujifilm, like other vendors will look to continue to capitalise on this spike, and will hope to continue to derive revenues from this expanded install base. Offering AI solutions under a subscription could be one tool that helps the vendor achieve it.

However, regardless of which options are utilised, embedding AI on modalities will not be a huge revenue generator in and of itself for either Fujifilm or Annalise. Fujifilm will boast of the capability to snag some extra deals and upsell to some customers in Australia, but the bigger opportunity lies when it is able to sell its products in India and other emerging markets. Fujifilm has some form in these regions, with its products tending to be more affordable than those of the likes of GE Healthcare, Philips and Siemens Healthineers, while it has also shown it is happy to use innovative methods to try to create business in these areas. In this context, offering embedded AI which seamlessly supplements a limited number of radiologists could be very beneficial. This is particularly true given that Fujifilm has adopted a partnership model. This will give the vendor flexibility to offer different tools in different regions (e.g., Fujifilm has already partnered with Lunit in Mexico and across some regions of South America) or abandon partnerships should better alternatives become available elsewhere.

For Annalise on the other hand, its focus remains on its fuller enterprise-based offering. Offering mobile versions of its solution embedded on mobile X-ray systems will help grow its market share and could serve as an introduction to the company for some providers, but, in isolation, such deals will never allow the vendor to live up to its lofty valuation. Instead, it must focus on selling to providers, with any embedded deals merely an additional revenue stream.

Ultimately it is a smart move, but not one that will have a huge market impact. More significantly, it shows how AI can be embedded into solutions, but these best of breed integrations represent only a small part of a market blessed with numerous approaches. Living on the edge remains a choice not a necessity.

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 ResearchTo view other recent Premium Insights that are part of the service please click here

Signify Premium Insight: Autonomous AI Debate Reiterates Importance of Radiologists

This Insight is part of your subscription to Signify Premium Insights – Medical ImagingThis 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 ResearchTo view other recent Premium Insights that are part of the service please click here.

“With the medical imaging AI market still maturing, suggestions of autonomous AI may be a leap of faith radiologists are yet unwilling to take.”

Dr Sanjay Parekh

Lithuanian AI developer Oxipit recently announced that it had secured CE Class IIb certification for its ChestLink autonomous AI solution. According to the vendor, the tool assesses chest X-rays and reports on those scans showing no abnormalities without any radiologist involvement. The vendor claims it is the first such autonomous solution, where a diagnostic evaluation will be carried out solely by an AI tool.

The Signify View

Medical imaging AI has long promised to help relieve the burden faced by overworked radiologists in understaffed hospital networks. Many of the solutions currently available are contributing towards this goal, but do so by enhancing a radiologist, enabling them to do more in the same amount of time. They might, for instance, incrementally accelerate a radiologist’s reading of X-rays, or automatically quantify features on an MR scan, allowing exams to be interpreted more swiftly. However, while these use cases are beneficial, they are not transformative, speeding up existent processes rather than replacing them, or at least complementing them with something new. This additional value is what several of the most successful medical imaging AI vendors have, or at least are attempting, to do.

One application for medical imaging AI, which is arguably the most transformative of all, is autonomous solutions. While not designed to replace radiologists, these tools automate some of the more laborious tasks radiologists undertake, effectively promoting radiologists to a supervisory role and allowing them to focus on more complex and demanding tasks. Oxipit claims its ChestLink solution is the first such tool to have received regulatory approval, therefore becoming the first such tool that can be used commercially in Europe.

The tool is designed to scan chest X-rays and rule out healthy patients, automatically sending results for those that display no indicators of pathology. Any image that the tool does not determine as healthy, is flagged as such for radiologist review. The tool’s use is focused on primary care, where, according to Oxipit, as many as 80% of X-rays are of healthy patients. By ruling out this proportion of patients, ChestLink does have the potential to significantly reduce radiologists’ workloads. With, according to Signify Research’s Diagnostic Imaging Procedure Volumes Database, more than 70m chest X-rays undertaken in Western Europe in 2021, any automation could have a dramatic impact.

Market Malaise

However, while the potential advantages of such autonomous technology is evident, the case for the market being ready for it is less clear. The adoption of medical imaging AI is still very much in its infancy, with many of the most successful tools, even those which are used under close physician supervision, augmenting rather than replacing established clinical workflows. Stroke solutions from the likes of Viz.ai and RapidAI, for example, identify stroke patients more quickly so they can be prioritised. However, if the tool fails to detect an occlusion, the patient will not be prioritised, but will instead be left positioned on the worklist as though the tool had not been used at all. The worst case is the de facto pathway.

The stakes are much higher for an autonomous AI solution, however. If Oxipit’s ChestLink suite is wrong, and incorrectly clears a patient as healthy, the results could be severe. Oxipit has made efforts to assure potential customers of its accuracy, highlighting that prior to its approval the tool has been operating in a supervised manner, in multiple pilot locations for more than a year, and in doing so has processed more than half a million real-world chest X-rays. What’s more the vendor boasts of its 99% sensitivity metric and the “zero clinically-relevant” errors made during pilot studies. Outwardly, these claims are impressive, although, like many young AI start-ups, a lack of published clinical validation makes these figures difficult to fully assess.

Regardless, there are other challenges, a specificity rate of 99% implies a false negative rate of 1%, which, on the volume of scans that such a tool could be assessing could be significant. For edge cases and ambiguous scans, radiologists can also use other contextual information which may have a bearing on a diagnosis. AI suites don’t have that luxury, potentially impacting some diagnoses. Oxipit will no doubt ensure there are systems in place to address these shortcomings, but without a human’s reactive, situationally aware cognizance, there is still scope for problems given the range of unprecedented and unforeseen issues that could arise. This could be a particular barrier when it comes to taking responsibility for errors. If ChestLink is wrong and a patient is impacted, who is liable? Presumably either the autonomous solution must offer such a saving that expensive lawsuits can be stomached as an operating cost, or Oxipit must be held liable, otherwise providers would be unlikely to risk such litigation on such a tool.

What is Normal Anyway?

One of the central imperatives in medical imaging AI is the creation of value for radiologists. One of the ways vendors are achieving this is by developing increasingly comprehensive AI tools, which ‘solve’ a modality and body area combination, striving to identify every single finding for a given body area/modality combination. In the future, this could lead to autonomous tools which mimic a radiologist; assessing a scan for any finding and, either reporting that finding, or, if no finding is present, reporting a negative.

In contrast, Oxipit claims its ChestEye solution can identify 75 findings, relating to 90% of potential pathologies, which is technically impressive but still far less thorough than is required for the basis of an autonomous solution. Oxipit’s approach to ensuring that patients with one of these 10% of pathologies, to which ChestLink is essentially blind, are not given a clean bill of health is by effectively making ‘normal’ a positive diagnosis. Images are, in essence, compared to normal images, meaning that anything which does not conform to the trained pattern of normality is considered abnormal, and sent for radiologist review.

This is not necessarily problematic. In practice, what looks ‘normal’ is a lot vaguer than specific findings, but assuming ChestLink can deal with these vagaries, along with the myriad combination of patient demographic, modality manufacturer, inconsistencies in patient positioning etc, it will have value as a rule-out tool. However, this will be its only use, compared to other more sophisticated comprehensive solutions which offer a more nuanced analysis, and in the future could autonomously make positive diagnoses, as well as negative diagnoses based on the absence of findings.

Fighting for Firsts

These are issues that must be considered, both from a fundamental standpoint of whether such a tool can be usefully deployed, and the question of whether providers will choose to deploy it? This latter question may be complicated further by competition in the space. While Oxipit claims its CE Mark Class IIb approval is the first of its kind for such an autonomous solution, in practice, it appears functionally similar to Behold.ai’s Red Dot solution. This tool was awarded Class IIa approval in 2020, an approval Behold CEO Simon Rasalingham claimed was a “first in kind”, with the tool being the “first autonomous AI algorithm to rule out normal chest X-rays”.

Regardless, Oxipit’s attainment of European approval is an achievement. With the vendor last having secured funding in a seed round likely to be looking for more funding soon, especially if it is to avoid losing touch with competitors like Lunit and Annalise.ai, which have raised $137.8m and $117m respectively. Such an achievement, and the publicity it brings will no doubt help in this regard, so the timing is particularly fortuitous.

The ultimate question, however, is whether providers, radiologists and patients will be willing to hand over control to a computer programme. Interim solutions can be implemented; the UK’s Care Quality Commission, for example, insists that all examinations auto-reported by Behold.ai’s tool must be reviewed by a human within 24 hours. Similar stipulations would likely be attached to the use of ChestLink, with the vendor already insisting that the tool’s deployment begins with a phase devoted to retrospective analysis, a second stage in which the AI shadows radiologists, and only then, in the third stage, can autonomous operation begin.

These, and other such solutions to very central challenges are clumsy and inelegant. They are, however, necessary until more sophisticated answers can be found. These fixes rob autonomous AI of some of the potential it holds, mitigating the benefits it can bring providers. Despite this, they mean that the tools can be used in hospitals. This makes Oxipit’s clearance significant, something to be celebrated and a factor that could tip the scales in the fight for VC funding. But, it is also clear that ChestLink isn’t a complete answer to all the questions posed by autonomous AI, that transformative leap remains, for the time being, out of reach.

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 ResearchTo view other recent Premium Insights that are part of the service please click here 

Signify Premium Insight: Just Getting Started? Harrison.ai Raises $100m

This Insight is part of your subscription to Signify Premium Insights – Medical ImagingThis 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 ResearchTo view other recent Premium Insights that are part of the service please click here.

In recent weeks Australian AI developer Harrison.ai joined an exclusive club. After securing AUD129m ($94m) in Series B funding, the outfit has become one of the handful of well-funded medical imaging AI vendors that have raised more than $100m.

The funding, which brings Harrison.ai’s total raised in the last two years to over $120m, was led by returning investor Horizons Ventures and also saw participation from Blackbird Ventures and Skip Capital. These investment firms were also joined by Sonic Healthcare and I-MED Radiology network two Australian providers which have deployed Harrison’s AI offering, lending the round an unusual level of consumer, as well as financial, weight.

Beyond merely investing in the firm, these provider partners will also help Harrison.ai target new areas of healthcare, with the vendor announcing plans to target pathology, among others.

The Signify View

As medical imaging AI success stories go, HeartFlow’s is hard to beat. As discussed in a previous Premium Insight when the heart health developer listed, it set a new financial benchmark. When it first launched on the New York Stock Exchange, the vendor had a pro forma enterprise value of $2.4bn, becoming medical imaging AI’s first unicorn.

Another of medical imaging AI’s financial flyers is Infervision. This Chinese vendor was itself the subject of a Premium Insight when it received $139m in Series D funding in July, bringing its total funding to more than $210m (despite an undisclosed Series C funding round).

Look back a few years however and these vendors’ series B funding rounds pale in comparison to Harrison.ai’s with Infervision securing $47m in 2018, while HeartFlow’s series B was only $20.4m in 2011. Of course, changing markets and changing VC strategies mean that these figures aren’t directly comparable to the nigh-on $100m that Harrison.ai has just secured for itself, but it does indicate the kind of rarefied company that the vendor is joining. It also begs the question of how such a sum has been achieved.

Comprehensive Valuation

There are a number of factors that have gone into establishing its valuation, but at the core is Harrison.ai’s central product, its Annalise.ai diagnostic imaging AI. Key to this product is its comprehensive approach to diagnostic radiology. Most solutions automatically identify a number of findings on an X-ray, but still rely on a radiologist to identify those not covered. AI vendors are addressing these gaps using various methods including partnering with other developers to add additional capability or creating platforms and bundling individual algorithms into suites which address particular clinical requirements.

Annalise.ai instead aims to ‘solve’ a particular scan type (its focus so far has been chest x-ray) and automatically identify all possible findings on any given image. So far, its solution identifies over 125 findings. In doing so it aims to make the selection, deployment and use of AI easier for providers. Further value could also be added to the solution in future as additional workflow tools are included, such as structured reporting, for example.

This approach looks to be effective, with the vendor’s own validation studies, which were published in The Lancet Digital Health in July, showing that radiologists assisted by the tool performed better in the vast majority of cases than those that weren’t assisted. What’s more the model’s AUC was also found to be statistically superior to unassisted radiologists for almost all findings.

Beyond published research, however, real world indications also show the value of the tools, with several providers choosing to use the tools in their own hospitals, including Sonic Healthcare, and I-MED, which have gone on to invest in Harrison’s Series B funding round. The fact that customers have quickly become investors is quite the endorsement.

The company’s ambitions, however, do not stop at chest x-ray, and they are looking to develop comprehensive solutions to other high turn-over scan types. In the long run, the company wants to address most of the high turnover scan types via its potential portfolio of comprehensive AI solutions. Early on, this was viewed as a potentially risky approach, such is the breadth of competition that has homed in on higher-volume scan types like chest X-ray. However, the comprehensive findings approach in a singular offering has allowed Harrison to stand-out from the crowd of aspirant vendors, most of which are offering a singular or a limited number of findings.

Ambitions in Pathology

The performance of Harrison’s radiology AI offering is only half the story, however, with the vendor’s stated ambition in pathology also having an impact on its prospects.

AI applications in pathology do, after all, hold significant potential, but the conditions for this potential to be realised are not yet in place. The most significant challenge is the general under adoption of digital pathology. However, this is starting to change with several factors such as regulation changes in the US, and the turbulence created by Covid-19 highlighting the lack of digitisation in pathology and giving impetus for change.

As these and other catalysts continue to grow in significance, the adoption of digital pathology will increase. As evidenced at RSNA, this is also a trend among imaging IT vendors which will increasingly incorporate pathology into enterprise imaging platforms. Against this backdrop, pathology AI will be able to find a footing.

The quantitative nature of many tasks in pathology as well as the shortage of pathologists (which is even more acute than the shortage of radiologists) means it is an opportune discipline for AI to have a significant impact, especially as the breadth and complexity of pathology diagnostic findings is a multitude higher than in radiology. This could be particularly true for a vendor such as Harrison, which has been especially thorough with its approach to its comprehensive chest X-ray solution. Frankly, singular point applications will have limited traction in pathology.

Cohesive Competence

Harrison.ai is looking to take this cohesive approach further, expanding out of radiology and addressing another slice of the diagnostic workflow. Longer term this digital pathology tool, the chest X-ray tool and potential future tools could all be integrated, leaving solutions that are more complete in both individual areas, but also along the entire workflow. This cohesion could be particularly useful in areas like oncology, as the broader remit of such solutions would see the vendor providing a service rather than a technology solution. This would enable it to prompt purchasing decisions to be made at a more executive level (e.g., C-suite), tapping into a larger budget pool. However, multi-disciplinary convergence in diagnosis is only just gaining traction in care settings, so in the near and mid-term, Harrison should remain focused on serving each individual diagnostic sector to ensure continued success.

The fact that Harrison is also looking to develop its pathology tool alongside recent customer Sonic is also an advantage. Data is obviously one of necessities for vendors looking to develop AI solutions, but, for pathology in particular, this data is scarce. By partnering with Sonic, Harrison will have access to an abundance of clinical data for algorithm training and refinement, as well as a large user base on which to conduct pilot deployments and validation studies. These are all essential for the successful development of a digital pathology AI tool, and having a route to achieve these already in place will give Harrison an edge over some of its competitors.

Looking to develop a pathology solution was also shrewd from a commercial, as well as a clinical, perspective. While increasing numbers of medical imaging AI vendors are securing ever higher funding rounds, pathology vendors have recently tended to fare better as investors have noted that a surge to adoption is pending, with for example Paige securing $100m in a series C round in January, and PathAI netting $165m for series C in July. This disparity is in part a result of the applicability of some solutions to drug discovery, a market which harbours the greatest returns near-term, but also relates to the relative upside of tackling a pathology market that is still heavily analogue and therefore ripe for disruption.

Of Value and of Worth

In receiving $94m in series B funding, Harrison AI has joined a very exclusive group of medical imaging AI vendors funded over $100m. What’s more impressive is that it has achieved this at an earlier stage than any of its peers. The road ahead is long, and the money will be quickly allocated to address its often quite expensive priorities. Continued commercialisation of its chest X-ray solution will be the first order of business; securing US-FDA regulatory approval and selling into and supporting providers will also require significant funds. Looking further ahead, investing in product development for comprehensive solutions that address other high volume scan types will undoubtedly follow. In pathology, Sonic will provide a short-term commercialisation base, but in the more analogue pathology sector, the firm will also have to take on a degree of market education and evangelism, a process that can have a substantial cash-burn rate.

If these priorities can be achieved, and Harrison.ai can begin generating sizable revenues, then the trajectory for future funding rounds and potential listings could be unprecedented. Moreover, the vendor could have a profound influence on the direction of AI. Many of Harrison’s peers are trying to add value in different ways, such partnering to create suites and developing end-to-end solutions that address entire clinical workflows. Harrison.ai offers another way, creating truly comprehensive solutions for specific use cases and then expanding into other adjacent areas. If the vendor is able to achieve commercial success on a par with its funding success, the developer will no doubt sit alongside HeartFlow as a posterchild of the segment. This could be particularly true if the vendor decides to list in the future.

There are, of course, challenges ahead. A lack of standardisation in pathology could make things harder than the DICOM-based world of radiology, while looking to split focus, as well as investment, between different areas, particularly when the vendor is still so young, could prove to be detrimental to both. Doubly so as it begins to compete with more established competition on both fronts.

These are proportionately minor worries, however. Harrison.ai has progressed carefully and methodically and to the pain of its competitive peers, very quickly. Now, bolstered by extra cash, and guided clinically by its customer partners, the precocious vendor is ready to demonstrate that its worth extends far beyond its valuation.



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 ResearchTo view other recent Premium Insights that are part of the service please click here