In 2018, Stanford University developed an algorithm, dubbed CheXNeXt, that interprets X-Ray scans and calculates the likelihood for fourteen different pathologies. The AI was reportedly able to beat radiologists at their own game, able to analyze the X-Rays with similar accuracy, but doing so in the span of a few minutes as opposed to the few hours it normally takes a radiologist. Should radiologists should be afraid of losing their jobs to machines like this?
Matthew Lungren, MD, Assistant Professor of Radiology at Stanford, is excited for the algorithm to enter the medical field, but even he cautions against being overly reliant on it. “Would you fly a plane without a pilot?” he asks hypothetically.
CheXNeXt works by looking at X-Ray images that are already confirmed to be positive or negative for a specific disease, say pneumonia. It will then find similarities between the pneumonia-positive X-rays, and find differences between the positive and negative X-rays. It’s then able to look at new X-ray images and determine if they share more commonalities with the positive X-rays in its database or with the negative X-rays get to a result. This is called Deep Learning. CheXNeXt was given over 10,000 images to use in its memory, and the tests left the researchers quite confident.
Part of the community of radiologists advocates for technology’s integration into clinical use. Dr. Constance Lehman, radiologist at Massachusetts General Hospital, is openly fond of AI, “absolutely” hoping that technology replaces what she does. She refers to the one she uses as her student, “training” it with over 200,000 mammograms. But she goes on to say that what the program does is only a narrow slice of the role of a radiologist, and that radiologists “do much more than search for patterns on images.”
Dr. Paul Chang, Radiologist at University of Chicago, voices that while the so-called “Gold Rush” of technology and AI into the medical field is exciting, it’s also mostly hype. He doesn’t fear any large-scale replacement taking place in his field. He believes that over time, technology like the one developed at Stanford, when developed and helpful enough, will be used, “not because it outperforms [radiologists], but it’s going to augment [radiologists].”
Radiologists are still divided over whether AI will help or hurt their profession. Supporters like Dr Lehman are looking forward to technology changing the game of their profession, allowing it to handle parts of their job so they can focus on others. Skeptics like Dr. Paul Chang, while not completely dismissive, hold the idea that practicing radiology will stay the same, at least for a while. The airline pilot eventually had to embrace technology and automation finding its way into the cockpit, but there still remains a real need for a human to be in the seat. It doesn’t seem likely that AI will completely replace radiologists any time soon, but technology like CheXNeXt may find its use one day in enhancing the field of radiology, making it more accessible and helping to ease some of the work for radiologists all over.