Last week, I had the opportunity to speak at the Artificial Intelligence in Medicine Symposium sponsored by Telemed.edu and the Michigan State Medical Society. It was great to be back in my home state at the Troy Marriott. The only downside to the meeting was the weather didn’t exactly cooperate resulting in a delayed departure and an additional hour of flight time, and Troy/Oakland airport runway was closed, so I decided to land at Ann Arbor, visit my parents, and drive. You can see my actual flight, which took a little longer than anticipated posted below.
Despite the travel issues, the meeting was surprisingly good. This was the first meeting I’ve spoken at where AI in medicine was the unifying theme. Typically, I find myself at anesthesiology or critical care meetings or healthcare IT meetings where AI isn’t the central focus. This was distinctly different.
What became clear to me was that AI has already invaded much of medicine. It is not just pushing the envelope but becoming a solid part of some practices. This appears to be particularly true in imaging where AI is moving into the diagnostic realm. This was demonstrated with ophthalmology where technologies like those from Digital Diagnostics can diagnose diabetic retinopathy, and the use of the technology is actually compensated at a higher value than human generating the diagnosis.
The realm of assisted diagnosis, where AI helps to identify and/or risk stratify patients, appears to be moving ahead at a rapid pace. Again this is present in the imaging space where AI appears to be impacting radiology and anatomic pathology, identifying “normal” vs “abnormal” images allowing the clinician to see the likely abnormal images first in their workflow and then reviewing normal images to validate the are indeed without pathology. This is similar to the area I discussed with patient deterioration and prediction of deterioration in the inpatient setting. Again, clinicians can use the technology to focus their attention on patients that are most likely to decompensate and attempt to prevent the event.
An area that didn’t occur to me was where AI is being explored below the direct clinician/patient layer. There were a couple of excellent talks discussing how AI was used simply to improve the display of data by smoothing images or figures. These made things simply more readable and less complex rather than actually providing some form of risk stratification or diagnosis. This is where many of us are using AI today in our daily lives and we don’t even know it. Many of the images we see or interactions we have with our phones or computers are AI assisted today.
The last area of focus at the conference was AI and liability. There was an excellent discussion regarding the profound complexity of this topic. There are questions of liability for using such technology and potentially making an incorrect diagnosis, but that is just the start of the discussion. With different laws in all 50 states governing the practice of medicine and nursing, a technology in one state may be actively practicing medicine where it would not be doing so in another state. Essentially, the amount of case law is minimal at this point and legal guidance is speculative at best regarding the implications of using AI based technologies and the liability to the practitioner.
By the end of the conference, I was convinced we are at the start of a 15-20 year journey into the integration of AI in medical practice. While the use of such technologies will continue to accelerate in the coming years, the medical and legal impact of that use will not be known for probably the better part of a decade or two. Clearly, we are living in interesting times and a new age of medicine is ahead of us.