What do patents tell us about the nature and trajectory of medical AI?

Inter-CeBIL is the International Collaborative Biomedical and Innovation Law program involving the law faculties at Copenhagen, Cambridge, Harvard and select other partners. PHG Foundation is supporting the Cambridge Faculty of Law.

With funding from the Novo Nordisk Foundation, Inter-CeBIL has developed into a leading international hub for research-based analysis of legal regulation relating to current and future health and life science innovation.

This blog was co-written by Inter-CeBIL Cambridge leads, Professors Kathy Liddell and Mateo Aboy, and PHG’s Dr Colin Mitchell.

What do patents tell us about the nature and trajectory of medical AI?

The hype around artificial intelligence (AI) has become almost inescapable—both the promise of this technology to transform all aspects of our lives, even humanity itself, and the dangers that it may pose. It is widely predicted that AI is poised to reshape medicine in significant ways. Policymakers and regulators are therefore seeking to incentivise and streamline the development and implementation of high-quality AI tools, in the hope of leveraging significant clinical and economic benefits.

One potential lever for encouraging such innovation is the granting of patents. A patent gives an inventor strong rights to exclude other people from using their inventive idea for up to twenty years. This enables them time to recoup R&D costs by commercialising and marketing their discoveries, granting licences for their use, or selling the patent rights. However, if the system proves challenging for AI innovators there is a danger that difficulty obtaining a patent could stifle innovation. Layered onto this is the potential that some jurisdictions—like the U.S.—could be more amenable to AI patent applications while others—like Europe—could present more challenges. This could distort the innovation landscape. 

So, are inventors struggling to obtain patents for medical AI or machine learning? Is it easier or harder to obtain such patents in the EU or the US? If you are a developer, what do you need to be aware of and what can patents tell policymakers and the rest of us about the trajectory of medical AI?

Until recently, it wasn’t easy to answer some of these questions with hard evidence but a recent paper published in Nature Biotechnology shines a light on the patent system and medical machine learning (MML) in both the US and Europe. The three authors and experts in patent law, Mateo Aboy, W. Nicholson Price II and Seth Raker, carefully combed through the databases of the US and European Patent Offices. They searched for and evaluated 20 years of patent applications on both sides of the Atlantic and, in doing so, have provided some interesting answers. 

The big picture—robust and rising patenting of medical AI

The first thing this study does is to dispel concerns that patents might be unavailable for medical AI or ML inventions. In fact, over 3,479 AI/ML patents with a medical application were granted by either the European or US Patent Office between 2001-21. And there has been massive growth in annual applications since 2013, going from around 250 in that year to over 2,500 in 2021. 

So far so good for those seeking the patents but is this skewed to one jurisdiction over another? There had been real concerns this might be the case following a series of Supreme Court decisions in the United States between 2012-14—Mayo v. Prometheus, Myriad v. AMP and Alice v. CLS Bank. This is because some things, like abstract ideas (e.g. mathematical formulas), are not in themselves patentable and, through these decisions, the US Court raised significant uncertainty about whether many software innovations were eligible for patenting.¹ What the authors actually found is that, despite this ‘doctrinal instability’, the US Patent and Trademark Office has been the patent office of choice over the European Patent Office by a very significant margin. And this isn’t just a consequence of US organisations dominating patent ownership in this field. In fact, Siemens (headquartered in Germany) is identified as the leading MML patent owner. 

What sort of medical AI or machine learning inventions are being patented?

The position of Siemens, closely followed by Philips at the top of the MML patent tree, is perhaps unsurprising given their status as one of the largest manufacturers of MRI scanners and the dominance of radiology and imaging thus far in the roster of approved MML devices. However, it is wrong to think that all the innovation is occurring in this area. In fact, the patents show that a wide range of medical AI/ML applications are being invented and patented for purposes spanning, treatment, diagnosis, measurement, technology improvement and decision support (among others). 

One thing that is potentially surprising given the general hype around AI is that the patent landscape is largely free of this hoopla. Instead, this study finds that AI/ML elements are most often integral to the invention—not thrown in as a buzzword or token element—and that, so far, the weight of patented innovation has been toward relatively ‘conservative’ AI/ML applications such as measurement, analysis and classification tools, not fully automated diagnosis systems. 

What should we expect in the near future?

For the time being then, we should not anticipate a tsunami of wholly automated MML systems breaking across healthcare. Or at least, the evidence of the patent landscape does not bear this out. That said, the authors of this study acknowledge its retrospective nature and that claim-drafting practice is evolving with the technology. For example, terms such as ‘deep learning’ are being increasingly used in recent years. They suggest that the current focus on tasks that are significantly removed from ultimate clinical decision making reflects current capabilities and state of knowledge. They suspect that this pattern will change as more advanced MML systems gain acceptance in policy debates. 

For policymakers and developers

For policymakers and developers this comprehensive review of the patent landscape should provide some reassurance. There do not appear to be pressing reasons to panic that the patent system is failing to accommodate medical AI and machine learning properly. Instead patent attorneys and inventors are finding ways around potentially difficult doctrinal hurdles when necessary and—as far as the authors can tell—patenting is working as an important driver to incentivise diverse and largely practical (as opposed to hype-infused) innovation. 

Perhaps one of the most interesting insights from this type of study is that radical impacts of AI/ML on medical decision-making appear to be some way from being realised. What the patents show is that most innovators have had their head down focused on less glamorous—but arguably more valuable—uses of AI or ML to enhance the myriad measurement, detection or classification tasks that currently absorb precious clinical time and effort.

Reference

1: Aboy M, Liddell K, Crespo C, Cohen IG, Liddicoat J, Gerke S, Minssen T. How does emerging patent case law in the US and Europe affect precision medicine?. Nature biotechnology. 2019 Oct;37(10):1118-25.