RSNA 2017: The Race for Medical Imaging Analytics Ecosystems Takes New Turns

At the HIMSS17 show last March we announced the race for medical imaging analytics ecosystems, and profiled seven vendors making early moves. Eight months later at RSNA 2017, this race continued to pick up steam in a big way.

With a momentum that seems to mirror the rapid proliferation of medical imaging artificial intelligence companies, more and more of the larger informatics vendors are looking to capitalize on these diverse applications to augment their own offerings. As more AI algorithms obtain regulatory approval, the big opportunity now is to channel them to market while delivering them as an integral part of the daily clinical workflows. Indeed, experience has shown that even the smartest computer-aided detection or diagnosis (CADx) algorithm, if not well integrated into the routine workflow, is likely to be ultimately ignored by most clinical end-users.

Therefore, while these algorithms may become validated for market use, they still need a pathway to market adoption and “a place to live” once in the field. It is not reasonable to think that their creators, often startups, have the wherewithal and the market access to channel them into the field. In fact, if they had to switch gears towards the marketing and sales function, they would quickly lose their technological edge and inevitably slow down their product portfolio development. Not counting the fact that, at a time when integrated delivery networks (IDN’s) and health systems are trying to consolidate their vendor relationships and standardize their procurement mechanisms, no one is going to enter into dozens of new vendor contracts to deploy various niche AI applications.[[{"type":"media","view_mode":"media_crop","fid":"64768","attributes":{"alt":"Nadim Michel Daher, Industry Principal Medical Imaging and Informatics, Frost & Sullivan","class":"media-image media-image-right","id":"media_crop_198633661643","media_crop_h":"0","media_crop_image_style":"-1","media_crop_instance":"8297","media_crop_rotate":"0","media_crop_scale_h":"0","media_crop_scale_w":"0","media_crop_w":"0","media_crop_x":"0","media_crop_y":"0","style":"height: 213px; width: 170px; float: right;","title":"Nadim Michel Daher, Industry Principal Medical Imaging and Informatics, Frost & Sullivan","typeof":"foaf:Image"}}]]

Unlocking Access to Curated Datasets

Deep learning frameworks provide an efficient way to develop software algorithms, at much higher speed than traditional coding methods. Therefore, the proliferation of medical imaging AI solutions that we saw at RSNA 2017, with a massive number of new players and new algorithms from existing players, is bound to continue. Whether these frameworks are utilized in a supervised or unsupervised (or hybrid) way, what it really boils down to is assembling a robust dataset of images, associated findings, annotations, and outcomes, then let the framework work its magic to identify the relationships between them, and derive the right software code.

That’s where few vendors have identified the opportunity to help AI developers gain cost- and time-efficient access to large, well-curated medical imaging datasets. DICOM Systems with their Imaging Data Supply Chain, and Medexprim with their now open-sourced Radiomics Enabler, are two such vendors that promoted this value proposition at RSNA. This commercial data access is meant to complement the two strategies for accessing good quality image datasets that early-moving imaging AI vendors have had to rely on: the imaging resources and biobanks available for R&D projects such as these (list not exhaustive) on one hand, and on the other hand, one-to-one R&D collaborations between individual hospital organizations and individual vendors.

Just as interesting to see in this context of capitalizing on data, is the growing vendor interest for delivering ways for academic researchers themselves to collaborate, promote and validate their own home-grown advanced algorithms. Flywheel, with a focus on neuroimaging, was created to do just that.

Enter the New Imaging AI Marketplaces

Our seven early movers are all making some good progress, either by adding native applications to their advanced imaging analytics ecosystem, or by expanding their portfolio through new partnerships with  independent software vendors (ISV’s). Notably, Google Health made it as a first-time exhibitor at this year’s RSNA, while NTT Data augmented its AI partner portfolio by integrating  a promising newcomer, MD.ai. In parallel, few imaging AI vendors that started out with a single niche application are gradually expanding organically by adding new native modules. A good case in point is Arterys who announced at RSNA it had received FDA approval for its Web-based platform, MICA, where it will soon roll out its new clearance-pending oncology lung CT and liver CT and MR applications, complementing the FDA-cleared cardiac MR application.

Meanwhile, much of the attention on the RSNA showfloor was drawn towards few new contenders setting out to build their own AI marketplace or “app store”, the top two of which would be:

• Nuance: Nuance is probably the highest-profile new entrant in this playing field, due to the vast market footprint of the radiology reporting giant. Nuance launched its imaging AI Marketplace at RSNA, dubbed the first “open” marketplace which brings on-board a number of imaging AI algorithms. Concurrently, Nuance also announced two new significant related partnerships, with Philips Healthcare, and with the GPU market leader NVIDIA, both of which strengthen its market entry. Things have shaped up very nicely for Nuance in the imaging analytics area since its 2014 acquisition of image sharing vendor Accelarad (re-branded PowerShare Network).

In fact, Nuance was able to capitalize on two other related acquisitions that were much less promoted publicly, both radiology analytics vendors: that of Montage Healthcare Solutions (re-branded as PowerScribe 360 MONTAGE and mPower), and that of Primordial Design. Now, the combination of these image management and analytics capabilities, along with the new partnerships, put Nuance in an ideal position to power-up, populate and operationalize its new imaging AI marketplace.

• EnvoyAI: EnvoyAI is the new name for McCoy Medical Technologies, which advanced visualization vendor TeraRecon acquired last June but decided to maintain as an independent entity. EnvoyAI’s new Exchange platform is an imaging AI distributorship, with over 15 partner ‘apps’ and 35 algorithms, and counting, already accessible through the platform. EnvoyAI also acts as a channel for some native TeraRecon ‘apps’ it contributes to the platform, since two are already part of EnvoyAI’s offering. Synergies could also work the other way for the two companies, since on its side TeraRecon also pre-launched its new Northstar viewer, dubbed the ‘world’s first AI-results viewer’, which will allow for closer interaction between image viewing and these advanced imaging analysis tools.

The company is dedicated to making the use of medical imaging AI algorithms practical for end-users, which means they will be able to experiment with the various algorithms and pay per-use. These prices, determined by each AI vendor, already vary widely across the portfolio. For example, few algorithms are priced under $1 per study, while others, which can be billed for against a CPT reimbursement code, at over $40 per study.

FDA Approvals… and Workarounds

As of the RSNA 2017 timeframe, only a handful of imaging AI algorithms seeking regulated use cases, already have U.S. FDA market approval. While the process is time- and cost-intensive, most vendors seem to have the ball rolling on the process, yet everyone is eagerly waiting for FDA regulation on how to deal with the continuous learning of algorithms, rather than having to re-submit regular “Maintenance 510k” applications for new algorithm versions, as is being done by some right now.

The marketplaces offering access to both FDA-approved and non-FDA approved algorithms raises the question of whether we are bound to enter an interim phase of off-label, experimental use of the regulated yet non-approved ones, as has been largely the case, for example, with image study interpretations in non-diagnostic grade mobile viewers.