“The greatest opportunity offered by AI is not reducing errors or workloads, or even curing cancer: it is the opportunity to restore the precious and time-honored connection and trust – the human touch – between patients and doctors. Not only would we have more time to come together, enabling far deeper communication and compassion, but also we would be able to revamp how we select and train doctors.”

Eric Topol, author of Deep Medicine

This manuscript from Nature is a review of the state of the art of AI in health and medicine. While AI has contributed to medical image analysis and reached some degree of adoption in various clinical venues, other areas remain a challenge. This review is partly based on a weekly review of medical AI by the authors, in the form of a podcast.

A reasonably good summary of recent progress in the deployment of AI algorithms in medicine opens the manuscript. Of note is that both the Center for Medicare and Medicaid Services, as well as the Food and Drug Administration, have made progress in AI adoption in healthcare.

A very useful diagram (Figure 1) in this manuscript is the overview of the progress, challenges, and opportunities of AI in health. The progress of mainly medical images in fields such as radiology and pathology, along with CMS reimbursement and FDA approvals, has been steady over the past few years.

The authors are correct in stating the challenges as well: implementation (dataset limitations and building model trust), accountability (regulation and responsibilities), and fairness (equity and bias).

Finally, the opportunities include: non image data sources such as EHR, signals, multimodal, and signals, as well as AI beyond supervised learning (semi supervised learning, unsupervised learning, and image reconstruction). There is a discussion about data sharing solutions such as federated learning, a decentralized methodology that shares model updates but not data.

While there are very few surprising insights in this review for the avid AI in medicine reader, it is nevertheless a good overall summary of the past few years of AI in medicine and healthcare. The concluding statement, “The field of medical AI has made considerable progress toward large-scale deployment, especially through proactive studies such as RCTs and through medical image analysis,” as well as others in the conclusion section are overly optimistic, or at least unrealistic, from the clinicians’ perspective. An even stronger effort to promote strategic use of AI in all venues outside of the elite academic medical centers would have been useful.

Lastly, none of the authors (other than the venerable Dr Topol, of course) is a clinician, so the manuscript is skewed towards data science and AI, as well as published reports, rather than being balanced with equal focus on clinical relevance in the real world. In short, just because a study is validated does not mean there is sufficient clinical utility or significant clinical impact for the clinician and the patient.

Read the full paper here.