3 ways machine learning will disrupt radiology—and the rest of medicine with it

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - MachLearn

Machine learning’s expansive capacity to quickly turn big health data into evidence-based care will challenge all practitioners of medicine to either grow along with the technology or accept getting left behind by it. And radiologists will be among the first to feel its push (if they’re not among the rads who are already working with it).

So predict a pair of medical thought leaders in commentary published online Sept. 29 in the New England Journal of Medicine.

Emergency physician Ziad Obermeyer, MD, MPhil, of Harvard and oncologist/bioethicist Ezekiel Emanuel, MD, PhD, of the University of Pennsylvania note that the AI subfield of machine learning draws out rules from data.

This is distinct from AI “expert systems” algorithms, which apply human-created rules to draw conclusions about specific scenarios.

Machine learning, the authors point out, has by now become ubiquitous in most sciences.

“In astronomy, algorithms sift through millions of images from telescope surveys to classify galaxies and find supernovas,” they write. “The same methods will open up vast new possibilities in medicine.”

In order to perform well, they add, medical machine-learning algorithms need to feed on massive datasets incorporating millions of medical observations.

(For this reason, the first successful adopters in radiology are likely to be big practices like the telerad giant Virtual Radiologic (vRad), which is already working with machine learning and is currently compiling and crunching image data from well over 2,000 imaging facilities across the U.S.).

Obermeyer and Emanuel lay out three ways medicine is likely to be disrupted by the coming data-into-knowledge transformation. 

1. Machine learning will dramatically improve the ability of health professionals to establish a prognosis.

Obermeyer and Emanuel back this up by citing early evidence from their own ongoing work using machine learning to predict death in patients with metastatic cancer.

“We can precisely identify large patient subgroups with mortality rates approaching 100 percent and others with rates as low as 10 percent,” they write. “Predictions are driven by fine-grained information cutting across multiple organ systems: infections, uncontrolled symptoms, wheelchair use and more.”

The authors predict the use of prognostic algorithms to spring up over the next five years, adding as a caveat the likelihood that prospective validation will take several more years of data collection.

2. Machine learning will displace much of the work of radiologists and anatomical pathologists.

These physicians focus largely on interpreting digitized images, which can easily be fed directly to algorithms instead, Obermeyer and Emanuel note.

“Massive imaging data sets, combined with recent advances in computer vision, will drive rapid improvements in performance, and machine accuracy will soon exceed that of humans,” they write. “Indeed, radiology is already partway there: Algorithms can replace a second radiologist reading mammograms and will soon exceed human accuracy.”

The patient-safety movement will increasingly advocate the use of algorithms over humans, the authors predict. “[A]fter all, algorithms need no sleep, and their vigilance is the same at 2 a.m. as at 9 a.m.”

3. Machine learning will improve diagnostic accuracy.

Citing the alarming rate at which medical errors occur, as spotlighted in the Institute of Medicine’s 2015 report “Improving Diagnosis in Health Care,” Obermeyer and Emanuel predict machine-learning algorithms “will soon generate differential diagnoses, suggest high-value tests and reduce overuse of testing.”

They expect this disruption to roll out more slowly than the other two, not least because the standard for diagnosis is often fuzzy and overlapping in many conditions (for example, sepsis and rheumatoid arthritis), where it’s often sharp and binary—malignant vs. benign—in radiology and pathology.

Obermeyer and Emanuel conclude by pointing out that needing to handle large datasets is nothing new to physicians, who have long had to assimilate patients’ physiological and behavioral factors along with findings from pathology and imaging. 

“The ability to manage this complexity has always set good doctors apart from the rest,” they write.

“Machine learning will become an indispensable tool for clinicians seeking to truly understand their patients,” they write. “As in other industries, this challenge will create winners and losers in medicine. But we are optimistic