Artificial intelligence is one of the most promising digital technologies in medicine. We’ve written about AI diagnostic apps with a diverse set of diseases and conditions. Earlier this year, we wrote about researchers at Radboud University Medical Center in the Netherlands training an AI system to determine the aggressiveness of prostate cancer. Scientists at UT Southwestern published a study of deep-learning AI that classifies tumors on the brain. In these studies, researchers used large numbers of images from existing patients to train AI engines to differentiate image contents quickly and accurately. So far, none of the studies we have seen suggest that AI algorithms can completely replace physicians. Several studies found AI performed better than novice specialists and nearly on par with experts in the respective fields.
An international group of researchers from universities and hospitals recently published a study in Nature Medicine that measured the diagnostic accuracy of AI support for dermatologists and primary care physicians with varied experience and expertise. The question examined in the study was not whether AI is better than generalists or specialists in diagnostic accuracy, but the extent to which AI-support can benefit physicians with varying levels of expertise.
The study included 169 expert dermatologists, 77 dermatology residents, and 38 general practitioners. The participants diagnosed dermoscopy images first without and then with AI-based decision support. The group wrote about several findings in the report. The significant outcome was that physicians who used good-quality AI support had greater accuracy overall than AI or the full group of physicians. AI support was more effective with less experienced specialists and generalists than with the experts. Bad quality AI support mislead all experience levels, which underscores the importance of high-quality AI models.
In our view, the concept of AI as a decision support tool to aid healthcare professionals is likely to have much better acceptance among the medical community and patient populations than the idea that AI would replace physicians.
