I am a pediatric cardiologist and Chief Intelligence and Innovation Officer of CHOC Sharon Disney Lund Medical Intelligence, Information, Investigation and Innovation Institute (Mi4). I am the founder of Artificial Intelligence in Medicine (AI-Med), the Medical Intelligence Society (MIS), the American Board of Artificial Intelligence in Medicine (ABAIM), the Alliance of Centers in Artificial Intelligence in Medicine (ACAIM) and the Pediatric Centers of Artificial Intelligence in Medicine (PCAIM). My book, Intelligence-Based Medicine is the first of its kind textbook on AI in medicine and is used at colleges and universities around the globe.
“AI winters were not due to imagination traps, but due to lack of imaginations. Imaginations bring order out of chaos. Deep learning with deep imagination is the road map to AI springs and AI autumns.”
Amit Ray, author of Compassionate Artificial Superintelligence AI 5.0 – AI with Blockchain, BMI, Drone, IOT, and Biometric Technologies
I have spoken and written several times about how we need to be very careful not to over hype artificial intelligence in healthcare so as to create an artificial winter, and this article resonates with my personal sentiment. The paper, with authors from mainly the Netherlands, reminds us that the majority (well over 90%) of AI models remain in the testing and prototyping environment, and that the development and implementation trajectory of these clinical AI models are complex and unsuccessful. In order to change the trajectory of these AI models and lower the risk of an artificial intelligence winter, the authors propose a step-by-step overview in order to both enhance clinicians’ understanding of these models, as well as promote the quality of medical AI research.
The authors suggest that we focus on the AI-related risks at four different levels: data, technology, process and people (“DTP2”). In addition, regulatory processes and guidelines have emerged to promote the quality of clinical AI research. Overall, these elements have created a fragmented medical AI landscape so the authors feel compelled to present a step-by-step approach to AI implementation in healthcare.
Phase 0 begins with defining the clinical problem (this is spot on) and engaging the stakeholders. Phase I involves the data preparation and model validation processes. Phase II starts with externally validating the model, while phase III focuses on design and execution of a clinical study. The final phase IV of this structured overview is the legal approval, as well as safety and governance of the data and the model.
There are several weaknesses to the proposed strategy. The phases seem relatively arbitrary and not easy to remember; this limitation will render the routine execution of this strategy exceedingly difficult, if not nearly impossible. In addition, my major disagreement with this strategy is that as many of the stakeholders as possible should be involved throughout the entire process, rather than at certain parts of this continuum. Lastly, this esteemed group of authors may already realize that, as good as this attempt to have a structured approach to medical AI is, there is no mandate from a body of leaders in AI in healthcare with authority to leverage for this strategy to be adopted.