Will Radiologists Be Replaced by Computers? Debunking the Hype of AI
5 reasons why the future of radiologists is secure.
Dr. Eliot Siegel, Professor and Vice Chair Research Informatics, University of Maryland School of Medicine.
There has been quite a bit of “trash talk” about radiologists being replaced by computers in both thepopular press and in the medical journals lately. Experts in artificial intelligence and machine learning such as Andrew Ng at Stanford have suggested that radiologists might be easier to replace than their executive assistants.
A well-funded startup’s CEO recently suggested that he would love to replace the “wasted protoplasm” that represents the radiology profession with a machine learning system.
Ezekiel Emanuel, principal “architect” of the Affordable Care Act, has gone so far as to suggest that radiologists might be replaced by computers in the next four to five years. He made his comments about artificial intelligence in healthcare in his keynote address at the ACR this spring. He repeated the comments in recent articles in the New England Journal of Medicine “Predicting the Future – Big Data, Machine Learning, and Clinical Medicine”, and in the Journal of the American College of Radiology “The End of Radiology? Three Threats to the Future Practice of Radiology”.
As a result of all the unfounded hype, I’ve been getting letters from trainees and colleagues. They are concerned about the potential threat of ‘automated radiologists’ and ask whether they should drop out of radiology or avoid it as a career.
I, For One, Do Not Expect Our New Computer Overlords to Arrive Anytime Soon
This prediction about radiologists being replaced by computers any time soon or in our careers is way premature. In fact, it’s nonsense. My conclusion is based on my perspective as a researcher in this area for the past 25 years and my involvement with major commercial “Artificial Intelligence” (AI) systems in medicine.
Here are my top 5 reasons why the future of radiologists is secure:
- Machine learning/convolutional neural networks have been successful with very small (e.g. 220 x 240 pixel) images. But they have not been applied to the far more complex images on a radiograph, much less a volumetric CT or MRI study. No one is anywhere close to having general success applying current techniques to medical images. In order to create a system to make radiology observations, one would need to combine thousands of algorithms developed over the past 25 years. Even these would only cover a tiny fraction of the diseases and diagnoses made by radiologists.
- No one has developed a system for general AI in any applications. To date, there are only narrow applications such as speech recognition or automated driving. These narrow artificial intelligence applications are currently limited. We are nowhere near a general AI system that could learn how to understand radiology in general as radiology residents do during their training.
- Even if a person from an advanced planet or our future traveled with a system that could interpret radiographs, it would take years to find the databases and interpretations to test it for the more than 20,000 diagnoses, findings, diseases, and modalities that a radiologist is trained to discern.
- The FDA currently does not have a system to “clear” a general system that would make findings and interpret radiographs. It would need to have evidence to test the many thousands of individual claims that would need to be made for such a general system.
- Lastly, radiologists do many, many more things than just making findings and diagnoses and recommendations. Their scope includes communication, image quality assessment, image optimization, education, procedures, policy making, and more. Being proficient at these essential skills would require a level of “general AI” that is at least 20 years and maybe far longer away.
So whether you’re a practicing radiologist, trainee, work in an imaging department, or are a part of the medical imaging community, rest assured that the future of radiology is secure.
My advice is that you embrace the exciting developments in machine learning that will make us safer, more efficient, and smarter, especially with regard to applications outside of image interpretation itself. But realize that computers will evolve slowly as an ever smarter, watchful, and loyal assistant that is eager to please. But machine learning and artificial intelligence will continue to lack the wisdom and knowledge needed to replace radiologists, at least in our lifetimes.
Want to learn more about the application of big data in radiology? Read the blog by Carestream Health.
Dr. Eliot Siegel is Professor and Vice Chair Research Informatics at the University of Maryland School of Medicine, Department of Diagnostic Radiology; and Chief of Radiology and Nuclear Medicine for the Veterans Affairs Maryland Healthcare System, both in Baltimore, MD. He is also adjunct Professor of Computer Science and Biomedical Engineering at the University of Maryland undergraduate campuses. He pioneered the world’s first hospital-wide filmless radiology department and has written over 300 publications on topics related to digital imaging, big data and high performance computing, and artificial intelligence applications in medicine.
Dr. Siegel is on Carestream’s Medical Advisory Board and on Carestream’s Board of Directors.