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
Signify Research has recently updated its AI regulatory database, which tracks the number of AI solutions that are cleared for use by the various regulatory bodies across the world. The number of regulatory submissions has continued to grow rapidly from 2021’s report, with vendors across the globe increasingly readying their solutions for hospital sales, and regulators increasingly acquiescing to their submissions. Since 2016, there were a similar number of CE Mark (189 AI algorithms) as there were US-FDA clearances (192); however, the number of NMPA Class III (China) approvals lagged significantly (17), with the first only issued in early 2020.
The number of CE Mark clearances ramped up faster prior to the advent of the Covid-19 pandemic; the total number of clearances in 2019 was equal to the total number between 2016 and 2019 (54). Since then, a further 81 algorithms have received CE Mark to date. US-FDA clearances have almost doubled since the pandemic (121 AI algorithms) compared with those approved between 2016 and the start of the pandemic (71).
Despite this growth however, most solutions approved continued to cater to familiar sub-specialties, such as chest imaging (e.g., chest x-ray, chest CT) and neuroradiology (e.g., stroke, brain MR) on well-addressed modalities, including CT, and MR. There were several solutions that received approval, for liver imaging and prostate imaging, for example, but these use cases represented only a tiny minority. This could change as drugs to target conditions such as liver cirrhosis or fatty liver disease become increasingly available and make AI tools that can identify and track patients with those conditions more valuable, but at present the demand for such solutions is limited.
Mirroring this is the renewed interest in brain MRI because of new drugs for Alzheimer’s disease being released, with drugs for Parkinson’s disease and multiple sclerosis on the horizon. AI tools that quantitatively analyse brain volume and structure will become even more valuable, especially as many have evolved from classical machine learning to deep learning tools.
A further factor in the limited approval of these solutions is that in some instances imaging is not the first line diagnostic procedure. For these conditions, such as prostate cancer, a diagnostic imaging AI solution fits less easily into existent diagnosis and care pathways, so there is, for the time being, less demand for such tools. Another, simpler reason for the continued focus on the same use cases is that there is growing demand in the market, with existing solutions acting as predicate devices, a factor which can expedite regulatory approval.
AI solutions for MSK imaging also remain relatively scarce compared to other imaging sub-specialties, despite the tremendous potential they may deliver to clinicians. For example, AI solutions will offer significant benefits for speeding up painstaking processes such as segmentation of spinal imaging, or by quantifying conditions such as knee osteoarthritis, supporting clinicians to make more accurate diagnoses, faster, and ultimately improving patient management.
There is, however, nuance to the regulatory process. In the US, for example many vendors are tending towards securing approval for triage solutions rather than diagnostic solutions, with the newer triage (CADt) regulatory pathway more straightforward than the traditional detection (CADe) or quantification route. This has somewhat opened up the market, enabling more tools to be approved than would otherwise be possible.
However, in some instances these CADt tools are more limited in functionality, resulting in them being less valuable to clinicians and less attractive to providers (due to the lack of diagnostic support for clinicians beyond the worklist). This could result in a relative glut of commercially available solutions, for which there is a distinct lack of demand.
One vendor that has been affected by this dynamic is Annalise AI, which recently announced it had secured US-FDA clearance for the triage and notification of pneumothorax on chest X-ray (CADt). In other markets, however, the vendor has full regulatory clearance for its comprehensive solution, covering the detection and quantification of more than 125 findings. While the CADt clearance does give Annalise visibility in the US, its attractiveness as a viable tool has been severely compromised in the US, threatening to hamper the commercial momentum the vendor has built in Europe and Australia.
This regulatory reticence could stem from several factors. One issue could be the lack of clinical validation studies that have been conducted using US data, a factor which, given the necessity of training datasets to reflect target populations could be a legitimate concern. Another potential cause is more philosophically driven, with some modalities, such as chest X-ray in the case of Annalise AI, being seen to reflect a lower value use case than other, more advanced modalities and their more headline-grabbing uses such as CT (e.g., FFR-CT and stroke detection). The US-FDA is, in effect, tacitly shaping the development of medical imaging AI in the US. What’s more, it is also likely to enable the approval of large numbers of solutions, but, given the limitations, still hinder commercial uptake in the US.
Problems with Paperwork
Regulatory factors could also stymie the adoption of AI in Europe, with the shift from MDD to MDR, and the introduction of UKCA set to curb the growth in clearances in the EU. There are reports of a significant backlog for MDR, with purported delays of 12 to 18 months in securing CE approval. This will steady the flow of approvals for AI solutions in Europe, but leave those that acted early to secure MDR in a strong position to capitalise, bestowing them with what almost amounts to a first-mover advantage. These vendors will be able to capitalise on the growing appetite for AI solutions, while other vendors are forced to await approval. They will also be able to establish themselves at providers and become the provider’s go-to vendor of choice, making it harder for competitors to displace them as their own solutions are cleared.
While those vendors who moved quickly in securing the new MDR approval should have a relatively smooth ride, there are still difficulties to consider. The clearest regulatory obstacle for any vendor that wishes to trade in the UK from July 2023 is the new UKCA approval. Introduced after the UK left the EU, the new UK Conformity Assessment prevents vendors with European approval selling into the UK, until they have been specifically cleared for the British market. Similarly, vendors that have won UKCA approval will be unable to sell in Europe without undertaking a separate clearance.
While this will require more resource from all vendors targeting both the EU and the UK, some vendors who have undertaken testing in the UK could be particularly affected. The availability of data in the UK has made it an attractive location for vendors to conduct pilot studies, but the value of these studies could be diminished in the eyes of the EU, which may prefer pilot sites within European Union countries. This could also impact which vendors are able to tender for contracts. European hospitals may prefer algorithms trained or validated on local patient data, and whether the UK continues to be considered ‘local’ remains to be seen.
These regulatory headwinds point to stagnation in Western Europe. While the region has seen growth over recent years, this is set to plateau, with more challenges for local vendors and less incentive for foreign vendors to try to make their mark on European soil, a factor compounded by the already fragmented nature of European healthcare. Combined, vendors may increasingly avoid the region to focus on more welcoming markets, such as Latin America and the Middle East.
More of the Same
One market that continues to grow is China. Although the country saw significantly fewer approvals (Class III NMPA) than Europe or the US, the overall total of approvals since the start of 2021 almost doubled compared to the previous year. This highlights the potential in the market, although vendors that are not China-native are unable to access this potential, with the country highly focused on nurturing its domestic capability.
Despite the near doubling in approvals over the past year, solutions that have received NMPA Class III clearance have continued to focus on several specific use cases, including FFR-CT, lung nodule detection, pneumonia, and bone fractures. As detailed last year, the reason for the focus on several specific use cases pertains to the availability of datasets for those tools, with approvals likely to be granted for more use cases as datasets become available. Additionally, further approvals could also be granted utilising successful international tools as predicate devices, as is understood to have occurred in the case of Keya Medical’s FFR-CT Class III approval, given that the NMPA doesn’t appear to harbour a relevant dataset.
Interestingly, there have been few approvals for X-ray and none for ultrasound. In a move that has shades of the US-FDA’s focus on advanced modalities, Chinese NMPA appears to also have a focus on MRI and CT. These modalities enjoy greater reimbursement and are more heavily used in diagnosis (whereas X-ray and ultrasound tend to be used more for screening purposes), as such, solutions centred around these modalities promise greater clinical value than many others.
One vendor that does stand out in China is Shanghai United Imaging Intelligence. It is one of the fastest growing modality vendors in China’s medical imaging market, but it has also secured four Class III approvals, giving it more clearances than any other vendor in the region, despite its broad focus. This includes the company receiving the first-ever Class III approval for its ICH stroke solution. This competence highlights the growing opportunity AI presents for United Imaging beyond modality sales, illustrating the vendor’s ambition to become an all-round competitor in the medical imaging market for both hardware and software, in China and beyond.
Combined, these regions overwhelmingly represent the bulk of AI approvals, with the US-FDA and CE Mark at present, far outperforming China. Some tools from outside these regions have also received approval, with limited approvals in South Korea, despite the country’s strong health tech sector. The same is true in Japan, where a less prominent start-up culture and tendency to focus on hardware rather than software has meant that aside from Fujifilm and Canon, which have focused on AI development, there is only one native independent software vendor which has received approval.
More broadly, the regulatory headwinds discussed above suggest that outside of China, the rate of approvals in Europe, or the sophistication of approved tools in the US could begin to slow after several years of acceleration. Innovation will not be held back, but the speed at which solutions can be commercialised in Europe and the US will suffer. While this could help adoption outside these established strongholds, with vendors sensing potential in other markets, it could prove frustrating for vendors looking to capitalise on the growing interest in AI tools. This is particularly true given other contemporary challenges such as the lack of clinical validation, and the often-unclear return on investment AI tools promise (especially given the lack of reimbursement). For AI start-up vendors with shortening funding runways, this must be a concern.
Regulation is one of the few tools governments that want to encourage AI adoption have at their disposal. As such, particularly given the other headwinds, more efforts may be put into streamlining the process. Of course, standards must be upheld, and the safety of tools must be beyond question, but there is no doubt that some facilitation by these bodies to help AI’s enormous potential to be realised sooner would be of great significance.
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