Clinicians are getting smart new virtual partners to improve care.
U.S. News & World Report interviewed Keith Dreyer, chief data science officer of Partners HealthCare and vice chairman of radiology at Massachusetts General Hospital and Brigham and Women’s Hospital, to get his perspective on the extraordinary impact artificial intelligence will have on health care in coming years.
What exactly is artificial intelligence, and what will it mean to health care?
The formal term we use is data science. That includes artificial intelligence and machine learning – the science of getting computers to act without being programmed by humans. Currently, scientists can pick a specific type of machine-learning algorithm and then train it to handle a certain task. One such algorithm is called a neural network because it can learn and improve performance on its own like the human brain, but it can work much faster. Collectively, these powerful tools will one day help us find disease almost before a patient is symptomatic, treat it early, and achieve a higher survival rate with much less patient suffering and at far less cost.
Want to publish your own articles on DistilINFO Publications?
Send us an email, we will get in touch with you.
How is AI being used now?
One key way is with diagnostic and imaging tools, like MRIs and CT and PET scans. Algorithms can be trained, for instance, to accurately measure all of the lymph nodes from a cancer patient’s CT scan to see if they’re changing size. It’s a huge job that algorithms can do much more quickly than humans. The clinician can then take the results and decide whether a therapeutic regime is working or needs to be adjusted or changed. We now also use machine-learning tools in stroke detection and classification. We can train an algorithm to learn to assess thousands of data points covering the range of strokes and how each can be characterized. Over time, the algorithm learns how to read images with a high degree of accuracy. For example, the computer might say the patient had a hemorrhagic stroke. Then the neuroradiologist and neurologist will determine whether surgery or medication is needed to treat it.
Where else is data science having an impact?
Population health management is one area. You can look at electronic health records for many thousands of people and identify biomarkers or other data to make predictions, for example, about patients’ likelihood of getting a disease. Patients with no prior history of diabetes, say, might have certain characteristics that put them on a path to the disease. So you could steer them to preventative care. Another area where AI can benefit patients is precision medicine. For example, it turns out diabetes has five or more types. With data science you can tease out a lot of information about how people with a specific type might react to different therapeutics and tailor treatments that will work best for each patient. Data science will also help in the laboratory in areas like pathology and genomics – anything that requires large amounts of data to be analyzed for discovery.
Should doctors and nurses be worried about their jobs?
You can’t replace the comfort of human-to-human interaction, but in the near future doctors using AI will win out over those that don’t in terms of delivering the best care. AI and machine learning will be critical in helping clinicians by aggregating and analyzing maybe thousands of data points for a particular patient (like lab results, genomics, imaging) to identify key conditions the doctor needs to manage, from pulmonary disease to congestive heart failure. So instead of clinicians being overwhelmed by data, they now have an AI partner to process and interpret the information and even advise them on treatment options.
How fast will these changes happen?
Incrementally. To put an algorithm into wide clinical practice, you have to collect and structure data to train the algorithm. Then you need to get FDA regulatory approval. Finally you need to figure out how to deploy that solution in clinical practice to providers using different electronic health records. And that’s just one algorithm. We might eventually need thousands in radiology alone. For now, we focus on solving narrow problems like detecting lung or breast cancers. We’re going to see big successes, but change won’t seem dramatic at first. People in the field all say that in time we will no longer talk about artificial intelligence, but rather a smarter “something” – a smarter cellphone, a smarter CT scanner, a smarter stethoscope – or a smarter physician.
Date: October 11, 2018