Five barriers to the commercialization of medical AI research

Consistent and significant investments in AI research in Canada have resulted in globally recognized centres of academic excellence across the country. The often-repeated challenge moving forward is to translate that academic work into commercial success. Based on my experience spinning off two companies developing technologies using AI in the medical space from the University of Ottawa and the Children’s Hospital of Eastern Ontario, I see five key barriers to such academic commercialization efforts. Commercialization of academic research in general has multiple benefits including the following: It is a faster path to transitioning research results into practice rather than relying on the natural technology transition process from academia. It converts research to valuable IP and equity that can be a source of additional revenue to academic institutions to support their broader educational mandates. It generates beneficial economic activity within Canada that can also attract highly skilled talent from overseas. The barriers to gaining these benefits relate to incentives and data. For incentives, we need to: 1. Create the right incentives for commercialization within academia. Currently in many academic institutions the reward structure is not set up to encourage commercialization. In fact, some have argued the exact opposite—commercialization activities are an inappropriate use of academic resources. If we really want to commercialize medical AI research, the incentives need to be aligned to encourage that. 2. Celebrate our success. Cases where there have been successful commercialization efforts are generally unknown. We are not good at promoting these, telling the stories, and establishing a positive narrative around the conversion of research results into successful businesses. The stories we tell are important as they demonstrate what is possible, set examples, and give room for faculty and students to take some risks. Medical AI is built on data. The ability to get access to health data to train models, and to commercialize these models, is important. This also faces some friction. 3. Restrictive policies on IP. Given that most health data is generated from a public system and that we have a more or less single-payer system, our data are quite valuable relative to other jurisdictions that are much more fragmented. However, the IP rights to the data and to AI models generated from the data are not always clear. This is problematic especially in the context of commercialization. And if there is an assignment of the IP in data, the terms would need to be predictable, reasonable, and not impose conditions that would be unattractive to investors. 4. Stakeholders against the commercialization of health data. There are stakeholders who hold strong views against the commercialization of health data, seeing data as a public resource that should only be used for the public’s benefit, and only entrusted to public sector entities. Profiting from data use is not deemed to be socially acceptable. To the extent that these perspectives take hold in policy and legislative circles, that would be a strong disincentive to commercialization—no one wants to be the enemy of the people. It is difficult to scale the commercialization of research if there isn’t a consensus that we should commercialize research. 5. Consistent privacy laws that enable commercialization. Privacy laws need greater flexibility to safely and responsibly allow health data to be disclosed to commercial entities to train AI models. This is typically done by creating non-identifiable datasets that are shared, using techniques like synthetic data generation, for example. The lack of clarity in many Canadian statutes on the obligations of companies when creating and disclosing these datasets (e.g., obligations with respect to patient consent and purpose limitation) creates uncertainty, and uncertainty results in risk-avoidance, which in turn means we throttle the commercialization of medical AI. These barriers are not independent of each other and are somewhat inter-related. While the barriers above are specific to health data, they may manifest themselves in other verticals as well. If we want to encourage commercialization of medical AI research from our academic medical centres, then a serious effort is needed to address these barriers. Dr. Khaled El Emam is the Canada Research Chair (Tier 1) in Medical AI at the University of Ottawa, where he is a professor in the School of Epidemiology and Public Health. The Hill Times