Artificial intelligence, machine learning, medicine and you

Dr. Gadget

March 20, 2018

Dr. Wesley D. Jackson

Recently, I learned that my favorite dermatology app, VisualDx, introduced an add-on feature called DermExpert which allowed the user to take a photograph of the skin lesion “utilizing Apple’s Core ML Technology to get personalized skin answers in seconds.” This app and website is one of the perks of a faculty appointment at the University of Calgary and access to the U of C library; it is otherwise available for purchase for around US$400 per year. The DermExpert add-on is another US$100 per year.

For this kind of money, I was expecting the add-on to somehow match the photo I take to the large database of dermatologic lesions on file to help me more quickly arrive at a diagnosis with the patient. Unfortunately, I was disappointed when the app would only guess at the morphology (and not so accurately to start) and didn’t seem to even attempt to access the aforementioned database. I expressed my concerns and disappointment to the company; they were very gracious, open to feedback, and readily refunded the money invested. At the same time, I was advised that the tool would improve over time and was also given an instructional tour of the website which made me appreciate even more its potential as a point-of-care diagnostic tool.

This experience prompted further questions on exactly what artificial intelligence (AI) and machine learning (ML) are, how they are related, and in particular, what their applications are in medicine. Briefly, AI is the broader concept of machines being able to carry out tasks in a way that we would consider “smart” and ML is a current application of AI based around the idea that we should be able to give machines access to data and let them learn for themselves. In other words, ML is a subset of AI, using a pioneering technology called the neural network, which mimics (to a very rudimentary level) the pattern recognition abilities of the human brain by processing thousands or even millions of data points. Pattern recognition is pivotal in terms of intelligence. AI encompasses other areas apart from ML, including knowledge representation, natural language processing/understanding, planning and robotics.

AI and ML are currently used in several applications in medicine that extend beyond simply identifying dermatologic morphology. For example, in the spring of 2017, a study conducted by the University of California, San Francisco (UCSF) suggested that the Apple Watch could detect an abnormal heart rhythm with 97% accuracy when paired with an AI-based algorithm called DeepHeart. There was also some suggestion in the same study that the sensors present in the watch could also accurately detect sleep apnea and hypertension.

In November 2017, the US Food and Drug Administration (FDA) approved the AliveCor Kardiaband ECG reader as the first ever medical device accessory for the Apple Watch in the US. This US$200 accessory, available in Europe for quite some time, clicks into a slot on the watch band and can be used to detect abnormal heart rhythms including atrial fibrillation. Users can obtain a monitor reading continuously and discretely by simply touching this sensor.

While these advancements are exciting, they do not come without some disappointment and some risk. While the Kardiaband or a similar device may be helpful in an identified well-informed high-risk individual, there may be significant harm in releasing such a tool to the masses.

Some concerns include:

  • How accurate will this technology be “in the wild?”
  • Even if it is accurate, how will this affect “the worried well” when they get information about “funny” heartbeats?
  • How does one react in a coherent way to deal with and act on this data?
  • Will reported normal heart rhythms falsely reassure individuals causing them to delay or avoid an appropriate visit to their physician, such as normal rhythms with anginal chest pain?

In November 2017, Apple, in partnership with Stanford Medicine, launched a first-of-its-kind research study using the heart rate sensor contained in the Apple Watch to look specifically at atrial fibrillation. The Apple Heart Study, available to US residents, uses an app which passively monitors heart rate and notifies the user if an irregular heart rhythm is observed. This notification is followed by a free video consultation on the iPhone with the study’s medical professionals for further analysis after which additional testing may be recommended. User privacy is protected as heart rate sensor data will be collected and analyzed by Stanford Medicine with no individual access by Apple. The results of this analysis will be used to further understand and improve the complex algorithms required for this type of individualized machine learning as well as advance our knowledge of this medical condition.

This Stanford/Apple approach seems to answer many of my concerns and illuminates some of the great potential for the use of AI and particularly ML specifically in cardiology, but also in many areas of medicine. For example, will we be able to predict and prevent falls based on identified gait abnormalities or is the development of a truly effective artificial pancreas for diabetics becoming a reality?

While these advances have great potential, the real benefit, as illustrated by the Stanford approach, will only happen after personal contact with a trusted professional – you!

The AMA advances patient-centered, quality care by advocating for and supporting physician leadership and wellness.