Will machine learning turn radiologists into losers?

By Brian Casey, AuntMinnie.com staff writer

October 5, 2016 -- Machine learning will "displace much of the work of radiologists" and other physicians who rely on interpreting digitized images, according to an opinion article published September 29 in the New England Journal of Medicine.

The article paints a picture of a brave new world in which big data, machine learning, and other artificial intelligence technologies combine to deliver healthcare that's more accurate -- to the benefit of patients (NEJM, September 29, 2016, Vol. 375:13, pp. 1,216-1,219).

But machine learning will also create "winners and losers in medicine," with radiologists apparently among the losers as their jobs are replaced by machines. The authors of the article are Dr. Ziad Obermeyer of Harvard Medical School and Brigham and Women's Hospital and Dr. Ezekiel Emanuel, PhD, of the University of Pennsylvania.

Obermeyer and Emanuel discuss how the emerging breed of machine-learning tools differs from current algorithms used in healthcare, which they call "expert systems." The latter are rule sets that encode knowledge on a particular topic and are applied to draw conclusions about clinical scenarios, such as assessing the appropriateness of an imaging exam, the authors wrote.

On the other hand, machine-learning approaches problems by learning rules from data, much as a physician progressing through residency might. Algorithms can analyze vast numbers of variables, looking for combinations that can predict outcomes. The value of machine learning is that it is capable of handling an enormous number of predictors and combining them in a variety of ways.

Obermeyer and Emanuel use the example of a chest radiograph, in which the "digital pixel matrices" underlying the radiograph become "millions of individual variables" that can be analyzed by algorithms that combine lines and shapes and learn the contours of fracture lines and other examples of pathology.

Ultimately, machine learning will disrupt three main areas of medicine, according to the authors:

  1. It will dramatically improve the ability of healthcare providers to establish a prognosis.
  2. It will "displace much of the work of radiologists and anatomical pathologists."
  3. It will improve diagnostic accuracy.

On the second point, Obermeyer and Emanuel predict that in the future digitized images will be fed directly to algorithms instead of physicians, as advances in computer vision drive performance improvements. Patient safety advocates will even get into the fray by recommending the use of algorithms over humans, the authors believe.

The radiology community has already been struggling with the arrival of artificial intelligence and machine learning, with some observers portraying the technology as a boon to radiologists, while others believe that artificial intelligence is so many years away from replicating the work of radiologists that it doesn't represent a threat to any radiologist's job.

Obermeyer and Emanuel are not so sure, however, claiming that the time scale for the disruptions caused by machine learning will be "years rather than decades." In a related article published September 18 in the Journal of the American College of Radiology, Emanuel and another co-author predicted that machine learning will become a "powerful force" in radiology in the next five to 10 years.

"As patients' conditions and medical technologies become more complex, the role of machine learning will grow, and clinical medicine will be challenged to grow with it," Obermeyer and Emanuel concluded. "As in other industries, this challenge will create winners and losers in medicine. But we are optimistic that patients, whose lives and medical histories shape the algorithms, will emerge as the biggest winners as machine learning transforms clinical medicine."

The 3 biggest threats to radiology
Radiology has long been an attractive field for physicians-in-training, with its generous salaries, (relatively) low rates of malpractice claims, and...
Machine learning may aid reading of thyroid ultrasound
Can machine-learning techniques perform better than a radiologist in differentiating between malignant and thyroid nodules on ultrasound? It depends on...
How soon will deep learning affect radiology?
For radiologists concerned about the arrival of deep learning, the future might be here sooner than you think. Deep-learning algorithms could begin producing...
Will AI soon put radiologists out of a job?
Are radiologists really "wasted protoplasms" -- as allegedly described recently by a CEO of a medical imaging start-up company -- who can be easily replaced...
Could machine learning diagnose Alzheimer's disease?
Could techniques like machine learning help radiologists diagnose Alzheimer's disease? Dutch researchers used machine-learning software to analyze arterial...

Copyright © 2016 AuntMinnie.com

Last Updated np 4/14/2017 9:55:27 AM

47 comments so far ...
10/5/2016 1:51:35 PM
One of the authors of the original NEJM paper gave bit more in-depth interview:
The key statement is IMHO this: 
"In 20 years, radiologists won’t exist in anywhere near their current form. They might look more like cyborgs: supervising algorithms reading thousands of studies per minute and zooming in to inspect and adjudicate ambiguous cases; or they might transform into “diagnosticians” like Dr. House, that go out and have more contact with patients and integrate that into their diagnostic judgments."
There are two possibilites ahead it seems, build on the already inherent geekiness of the profession and become basically machine learning experts, or get out of the dark and increase patient contact. The way I see if we want to preserve radiology as a medical profession then we should go for the second option.

10/5/2016 2:18:05 PM
Don't drink zeekie's kool aid. Radiology will be find. It may evlove, as it has done with PACS, CAD, etc, but it will not cease to exist. Though I am sure that is exactly what people like Zeekie wants.

10/5/2016 2:24:05 PM
I am not familiar with the authors as but similar, albeit a bit more reserved views have been shared by many. The question is will pattern recognition remain the holy grail of the radiologists skillset? On a long term the answer is likely to be a no.

10/5/2016 2:30:10 PM
Ezekiel Emmanuel is head of CMS, formulator of ACA and created the idea of so-called "death panels". For whatever reason, he has a blatant bias against radiology. He has said specifically on another panel discussion that radiology is in his "hitlist". Sounds like a true professional who values his colleagues. He clearly has a bullseye out on us. I would take ANYTHING he says with a grain of salt.
I would argue that primary care can be just as nice for machine learning. IBM watson able to assemble all kinds of data from many sources from EMR, labs, and stuff in the literature and can formulate a diagnosis and beyond. You can have a midlevel do the rest. 
All of this nonsense is a political attack on a specialty.

10/5/2016 2:37:56 PM
I see. I'm practicing in the EU so again I have no information about these people but there are many others including tech experts and futurists like Martin Ford or Erik Brynjolfsson who say that pattern recognition and therefore radiology is a low hanging fruit for machine learning.