AI is Joining the Radiology Workforce: Efficiency Gains Observed So Far.

Nine years ago, a prominent artificial intelligence scientist picked an at-risk profession.

“Individuals should stop pursuing a career as a radiologist now,” stated Jeffrey Hinton, asserting that AI would undoubtedly surpass human performance in this area within five years.

Currently, radiologists—medical imaging specialists diagnosing and treating diseases—are still in significant demand. Recent studies indicate a steady workforce growth projected by the American College of Radiation until 2055.

Dr. Hinton, who earned a Nobel Prize in physics for his groundbreaking AI research last year, has indeed had a monumental influence on technology.

This is evident at Mayo Clinic, one of the nation’s premier healthcare systems, with its primary campus located in Rochester, Minnesota. In recent years, Mayo Clinic has embraced AI technology to analyze images, automate everyday tasks, detect medical issues, and forecast diseases. AI also acts as a “second opinion.”

“But will it replace radiologists? We don’t believe so,” said Dr. Matthew Colestrom, chairman of radiology at Mayo Clinic. “We understand how challenging this work is and its interrelations.”

Computer scientists, industry experts, and policymakers have long debated the future of AI in the workforce. Will it serve as a smart assistant, enhance human performance, or be a robotic agent that displaces millions of workers?

The conversation intensifies as the cutting-edge technology behind chatbots appears to be advancing more quickly than anticipated. Leaders from companies like OpenAI and others forecast that AI will automate most cognitive tasks within a few years. Conversely, numerous researchers predict a more gradual transformation, akin to the introduction of electricity and the Internet, consistent with historical technological disruptions.

The potential obsolescence of radiologists serves as an illustrative example. Thus far, AI has proven to be a robust medical asset that enhances efficiency and augments human abilities, rather than replacing them.

Radiology has been a primary focus in the development and implementation of AI in healthcare. Of the more than 1,000 AI applications approved by the Food and Drug Administration for medical purposes, approximately 75% pertain to radiology. AI excels in identifying and assessing specific abnormalities, such as lung lesions and breast tumors.

“While there have been remarkable advancements, these AI tools mainly focus on general cases,” remarked Dr. Charles E. Kern Jr., a radiology professor at the University of Pennsylvania’s Perelman School of Medicine and editor of the journal. Radiology: Artificial Intelligence.

Radiologists do much more than merely examine images. They provide consultations to other physicians and surgeons, engage with patients, compile reports, and scrutinize medical histories. After detecting potential tissue anomalies, they interpret the implications for individual patients based on their unique medical backgrounds, drawing from years of expertise.

David Ortl, a labor economist at the Massachusetts Institute of Technology, stated that AI “underestimates the intricacy of work performed by humans.”

At Mayo Clinic, AI tools are being researched, developed, and customized to align with the hectic schedules of physicians. Since Dr. Hinton’s prediction, the radiology staff has expanded by 55%, now exceeding 400 radiologists.

Prompted by concerns and advancements in AI-related image recognition in 2016, radiology leaders assembled a team to evaluate the potential effects of the technology.

“Our initial thought was to leverage this technology for our betterment,” recalled Dr. Callstrom. “That was our primary objective.”

A decision was made to invest. Today, the Department of Radiology boasts a 40-member AI team, featuring AI scientists, radiation researchers, data analysts, and software engineers. They have created a diverse suite of AI tools, from tissue analysis instruments to disease prediction models.

The team collaborates with specialists like Dr. Theodora Pototzke, who focuses on the kidneys, bladder, and reproductive organs. She regards the radiologist’s role as that of a “secondary physician,” clearly conveying imaging findings and providing guidance.

Dr. Pototzke employs AI tools to gauge kidney volume. Growth in the kidneys, when coupled with cysts, can signal a decline in function even before changes are detectable in blood tests. Previously, she measured kidney volume mainly by hand, akin to using an on-screen ruler, resulting in variable outcomes and lengthy processes.

Serving as a consultant, end user, and tester for the department’s AI team, Dr. Pototzke assisted in designing software with color coding for various conditions and evaluating measurements.

Now, she can simply retrieve an image on a computer, click an icon, and instantly see the kidney volume measurements. This saves her 15-30 minutes with each kidney scan and consistently yields accurate results.

“This is a fantastic example of effectively utilizing AI for increased efficiency and accuracy,” Dr. Pototzke commented. “AI can augment, enhance, and quantify processes, but I am not in a position to relinquish interpretative duties regarding technology.”

In the hall, staff radiologist Dr. Francis Buffer elaborated on the various AI applications prevalent in the field, often operating behind the scenes. He stated that manufacturers of MRI and CT scanners incorporate AI algorithms to expedite image acquisition and enhance quality.

AI also autonomously identifies images with the highest likelihood of abnormal findings, effectively informing the radiologist, “focus here first.” Another application scans for heart or lung clots, even when the medical emphasis lies elsewhere.

“AI is currently integrated throughout our workflow,” noted Dr. Buffer.

In total, Mayo Clinic implements over 250 AI models, both developed in-house and sourced from vendors. The Radiology and Heart Disease divisions are the largest consumers of these technologies.

In some circumstances, emergent technologies unveil insights surpassing human capabilities. One AI model analyzes ECG data to forecast patients likely to develop cardiac fibrillation.

Research initiatives in radiology utilize AI algorithms to detect subtle transformations in pancreatic shape and texture, potentially identifying cancers up to two years before conventional diagnoses. The Mayo Clinic team is collaborating with other healthcare organizations to further validate these algorithms with more data.

“Mathematical modeling enables us to perceive what the human eye cannot,” mentioned Dr. John Haramka, president of the Mayo Clinic Platform, overseeing the digital initiatives of the health system.

Dr. Halamka, an advocate for AI, is confident that this technology will revolutionize medicine.

“In five years, failing to use AI will be considered a form of medical malpractice,” he suggested. “However, this means that humans and AI will collaborate closely.”

Dr. Hinton concurs. Reflecting on his previous statements, he believes he was overly broad in 2016, clarifying that his remarks were solely about image analysis, and while he may have misjudged the timeline, he maintains his original stance.

Over the years, most medical imaging interpretations are made through a partnership between AI and radiologists, which not only enhances accuracy but also significantly increases radiologists’ efficiency, according to Dr. Hinton.