The Exchange

Commentary and Observations from
the Medical Front Lines

A Glimpse of What Artificial Intelligence May Bring to Medicine

A Glimpse of What Artificial Intelligence May Bring to Medicine

I’ve always been fascinated by what the intersection of big data, technology, and medicine will bring. While I am a firm believer that IBM's Watson will not replace clinicians for the foreseeable future, the certain application of things such as artificial intelligence and predictive analytics will save lives, reduce health care costs, and improve outcomes. And we have a plethora of data—like chest radiographs—sitting out there, awaiting analysis.

A researcher at Massachusetts General Hospital and Harvard Medical School retrospectively examined two large databases in patients who had chest radiographs on file, and these patients continued to be followed and treated for a period of time. This team of researchers developed a convolutional neural network, colloquially known as deep learning or artificial intelligence (AI), and then trained it to stratify these patients by all-cause mortality using a single CXR in a cohort of 85,000 patients and radiographs. This revealed a graded association between risk score and mortality long-term, defined as between 8 and 12 years. These predictions were statistically significant for both cancer as well as non-cancer deaths, including cardiovascular and respiratory causes.

Now, this is not ready for prime time in part because even if repeated and validated, it begs the question, "What do I do with this information once I know that this specific patient has a dramatically increased risk of dying in the next decade?"

There are numerous medical—and especially ethical—issues these kinds of analytics bring to medicine that smack of science fiction movies. But this does begin to shed light on what AI may bring to medicine: the possibility that individual patients, when presented with information that confidently predicts their demise within a period of time if they continue unhealthy lifestyle and behavior choices, just might change their behavior. The skeptic in me says that a percentage may do the exact opposite or even worse.

Reference
  • Lu MT, Ivanov A, Mayrhofer T, Hosny A, Aerts HJWL, Hoffmann U. Deep learning to assess long-term mortality from chest radiographs. JAMA Netw Open. 2019;2:e197416.

Share this page:

Filed under: Preventive Medicine, Public Health

Development Widget