Radiologists don’t need to be experts in AI—but they should still study the basics
As the relationship between radiology and artificial intelligence (AI) continues to evolve, radiology trainees may find themselves wondering what, exactly, they should know about these groundbreaking technologies. Do they need to be AI experts? Can they just avoid the subject altogether?
A new commentary published in the Journal of the American College of Radiology explored this line of thinking. Trainees don’t learn much about AI or informatics during medical school, the authors explained, but they should still know some basics if they want to keep up with the specialty.
“Expecting all radiologists to run a gradient descent algorithm from scratch or navigate the intricacies of ResNet, a 152-layer deep convolutional neural network, would be unrealistic,” wrote authors Gerard K. Nguyen, MD, and Anup S. Shetty, MD, of the Mallinckrodt Institute of Radiology at Washington University in St. Louis. “Lack of advanced programming skills should not be a barrier to active engagement with AI. However, familiarity with the language of shared concepts in the field of data science would be a first step.”
Some of it, Nguyen and Shetty explained, isn’t quite as complex as one might think. And taking the time to learn about these concepts can make a world of difference to a young radiologist preparing to enter the workforce.
“For example, our term for sensitivity is their term for recall; our positive predictive value is their precision,” the authors wrote. “Other concepts in image-processing techniques such as thresholding or region-based segmentation may require only minimal explanation for those not already familiar. Looking toward the future, it may not be far-fetched to expect radiologists to recognize and understand reasons underlying an overfitting error of a deep learning malignancy classification model, much the same way one may recognize and understand the reasons underlying an aortic pulsation artifact to refrain from describing an artifactual lesion in the liver. The better we can understand and communicate the same language with back end developers, the smoother the integration will be.”
To gain this knowledge, Nguyen and Shetty suggested trainees begin within their own training program. Find the faculty member who seems the most focused on informatics and go from there. Also, they added, radiology trainees should seek out some of the resources developed by groups such as the American College of Radiology and Society for Imaging Informatics.
Collaboration is also key. The authors pointed out that AI startups have a need for someone with a background in radiology. By working with such a company, a trainee can provide a service while also gaining valuable experience working with advanced technologies.
“The main value a radiology trainee can provide for a startup is ensuring a diverse, reliable, and meaningful data set from which to train models,” the authors wrote. “As the saying goes, garbage in is garbage out, and curating a quality data set requires meticulous attention and proficiency. A radiologist can easily offer abnormality detection and segmentation, classification labeling, and ground truth verification.”
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.