Dr. Alan Kadish, President of Touro College and University System, joined fellow cardiologist Dr. Eric Topol to discuss the transformative changes that are coming to healthcare, thanks to Artificial Intelligence. Topol, director of the Scripps Research Translational Institute in La Jolla, California, is a leading voice on digital health. His newest book Deep Medicine is regarded by Dr. Kadish as a must-read—“As important to healthcare as Future Shock was to the computer age.”
Dr. Kadish: Patients and their physicians are accustomed to computers in medicine running things like MRIs, CT scans and medical records. I think they need to understand how artificial intelligence in medicine differs from just using computers.
Dr. Topol: The term artificial intelligence has been around for over 50 years, but what’s new is the subset Deep Learning, which is what’s having a remarkable and likely transformative impact on healthcare because you can take medical scans of images of any type—speech, voice and text—and train machines to do things that humans would never be able to do.
Dr. Kadish: Please tell us in lay terms what Deep Learning is.
Dr. Topol: Basically it’s a story of inputs and outputs. You have massive amounts of data to train and you put it through a neural net. These are layers of artificial neurons which can distinguish features progressively, and there’s no human bias about the algorithm because the data flows through these layers are determined by the data itself. We’ve learned that you can put these data sets, from this insatiable hunger for data, and get remarkable outputs that are accurate and are superior to what humans can do.
This is the main form of AI that has really ignited the world. Just recently, the Turing Award—the most important prize in computer science—was awarded to AI pioneers Geoffrey Hinton, Yann LeCun and Yoshua Bengio. But it’s going to have its biggest implication, I believe, in medicine.
Dr. Kadish: Please give us an example of how Deep Learning is already being used, or how you see it being used in the very near future.
Dr. Topol: It’s being used with respect to scans. Lots of different medical scans, such as X-rays and CT scans, are now in hospitals—not just the U.S. but throughout the world—are first being read by a Deep Learning algorithm, and then teed up for the radiologist. On the patient or consumer level, there’s the smart watch that diagnoses heart rhythm abnormality. It senses that for you; it offers Deep Learning for you. What is your resting heart rate? Your physical activity heart rate? When something is off track, it tells you on your wrist that you should do a cardiogram. It also gives you the interpretation of that cardiogram. Those are examples that are already out there and being used on both the physician’s and patient’s side.
Dr. Kadish: One of the fascinating things that you wrote about in Deep Medicinewas the way you believe that AI will actually improve the doctor-patient relationship. How do you see that happening?
Dr. Topol: That was not my premise in doing a few years of work on this topic, but then it became abundantly clear. If we do this right—and that’s a big if—we can restore the patient-doctor relationship, which is precious and has been largely eroding over decades. The reason being is that time—the gift of time—is essential. There’s so little time between patients and doctors, and during that limited time, there’s so many mistakes, so many unnecessary tests that are ordered because there is not enough time. And that goes on with prescriptions and procedures. There’s so much waste and unnecessary things done. We can override that by giving the gift of time back to both clinicians and to patients. That’s really what the remarkable potential is here because you have so much of the work being outsourced to machines. And then you enhance the human side, which is the context, the judgment and the oversight of that work, with the communications, guidance, trust and presence—all the good things that people can do. The human-human bond. Deep Empathy is the state that we have to get back to.
Dr. Kadish: A recent article in Fortunepoints out that there’s been a lot of data that electronic medical records, which was thought to have a lot of the same promises that you are mentioning, have actually not, on balance, improved things in terms of patient care or the doctor-patient relationship. If you agree that the electronic medical record hasn’t been done right, how do we do it differently with the implementation of AI?
Dr. Topol: The article that you’re referring to “Death by 1000 clicks” examines what it now takes to be a doctor or a nurse. It was the electronic health record that made clinicians into data clerks, and it was all done for business purposes. It’s horrendous! That was because of big business—it has done nothing in terms of caring for doctors or patients. So basically, what we’re talking about now with AI, is everything that’s just the opposite. This is to make doctor’s lives better, to restore trust to patients when go to a doctor—to have that precious relationship again.
I don’t believe that the current electronic health record is tenable because each person needs to own their data. The only way you can have deep learning of your data is if you have all your data. No one has it today. It sits in various health systems, doctor’s offices, hospitals; a lot of it is being generated by the person through their sensors. It could be their genome or parts of their genome. So all these different pieces of data are not assembled, but that’s what needs to happen in order for us to go into this ability for deep learning.