Following last year’s RSNA, there was a lot of discussion regarding the possible future of artificial intelligence being used in the field of radiology. As a year’s time has passed since RSNA 2017, there has been some interesting developments. Of course, the big question remains, “Will AI replace radiologists?” In the short term, the answer appears to be, “No. Not just yet.”
According to a recent Stanford University Medicine study, a new algorithm can read chest x-rays in a matter of seconds, screening more than a dozen types of diseases. The algorithm, CheXNeXt, is reportedly the first algorithm to “simultaneously evaluate x-rays for a multitude of possible maladies and return results that are consistent with the readings of radiologists.”
After scientists on 14 different pathologies, they were put to the test and performed at the same level as (and in one case, better than) radiologists. Previously, AI algorithms have been successful in performing in narrow, single use cases, whereas this particular study shows that it could eventually be possible for “algorithms to reliably and quickly scan a wide range of image-based medical exams for signs of disease without the backup of professional radiologists.” Despite that potential end, the ultimate goal is not to replace radiologists. Rather, the technology could serve as a resource for high quality digital consultations to resource deprived regions of the world without access to radiologists.
There would also be benefits in full developed healthcare systems as well. As of November 2018, the algorithm had been trained on 112,000 x-rays but the research will continue to explore performance of CheXNeXt with a much larger data set. Thus far, the algorithm’s training has included identifying masses, enlarged hearts, and collapsed lungs. This particular algorithm is also notable because its performance was compared to that of a group of radiologists. Taking this into consideration, the algorithm could potentially triage the x-rays for doctors to review or even work beside primary care doctors for consultation purposes.
There is more work to be done. We leave you with the words of Matthew Lungren, MD, MPH, associate professor at Stanford University and member of Stanford’s Bio-X, Child Health Research Institute, and Center for Artificial Intelligence in Medicine & Imaging. “We should be building AI algorithms to be as god or better than the gold standard of human, expert physicians. Now, I’m not expecting AI to replace radiologists any time soon, but we are not truly pushing the limits of this technology if we are just aiming to enhance existing radiologist workflows…Instead, we need to be thinking about how far we can push these AI models to improve the lives of patients anywhere in the world.”