Diagnostic Imaging: AI Performs at a Level Equivalent to Humans...Maybe

By Joe Elia

Edited by David G. Fairchild, MD, MPH, and Lorenzo Di Francesco, MD, FACP, FHM

Deep-learning algorithms (commonly called artificial intelligence or AI) are "equivalent to health-care professionals" in their ability to arrive at a diagnosis via medical imaging, according to a meta-analysis in Lancet Digital Health. That journal's commentator is not convinced, however.

The analysts offer their equivalence statement "cautiously" after detailing the many flaws in the research they examined. The imaging ranged from that done for ophthalmic diseases, to trauma and breast cancer. Additionally, few studies directly compared the performance of health care professionals with AIs on the same sets of patients.

The analysts, taking the 14 best studies into account, came up with a pooled sensitivity 87% for AIs, versus 86% for clinicians; for specificity, the numbers were, respectively, 93% and 91%.

The commentator offers this: "AI is no worse than humans, but the data are sparse and it might be too soon to tell."