De-Hyping Artificial Intelligence and Machine Learning in Health Care

By DocWire News Editors - Last Updated: October 20, 2018

Prior to the annual meeting, the ACR Clinical Research Conference focused on technology advances that are changing the landscape of medicine and rheumatology. Beyond electronic health records, there is a broad range of digital and mobile technologies.

Advertisement

One of the sessions addressed artificial intelligence (AI). Geoffrey Tison, MD, MPH, a cardiologist at the University of California San Francisco, framed the discussion by acknowledging how physicians and other healthcare providers are overworked. They are being asked to do more with less time and fewer resources, he said. So, can AI help? “The topic is surrounded by a fair amount of hype,” he said. “And I’d like to demystify that somewhat today.”

Simply put, he said, machine learning has taken the best of computer science and statistics and married the two. To perform machine learning, one defines a task, takes a large amount of data, puts it into a model, and then allows the machine to provide output.

For example, in his research applying AI to identify atrial fibrillation (AF), Dr. Tison and his team entered many electrocardiogram results (EKGs)—some that showed AF and some that did not—and labeled them as such, so that the machine could learn to recognize future EKGs as positive or negative for AF.

Although Dr. Tison described his work using applications of machine learning in cardiology, he believes the model is applicable to rheumatology and many other specialties within medicine. He cited examples of image recognition and machine learning to identify skin cancer and detect diabetic retinopathy.

Despite the possibilities, Dr. Tison cited limitations as well. Such models are very “data hungry,” needing thousands or tens of thousands of data points. Data must be labeled by humans first, which is cost prohibitive and time consuming. In addition, there is a potential lack of interpretability.

He concluded, “Humans no longer have a monopoly on complex problem solving, but in the age of machine learning, epidemiology remains as important as ever.”

Advertisement