Radiology SWARM: Novel Crowdsourcing Tool for CheXNet Algorithm Validation

Introduction Researchers at Stanford University School of Medicine and Unanimous AI conducted a study in which a “swarm” of radiologists (i.e. a group connected by Swarm AI algorithms) reviewed a set of 50 chest x-rays and for each predicted the likelihood that the patient has pneumonia. The predictive accuracy of the Swarm AI system was then compared to that of the machine learning program CheXNet, which has been shown in prior studies to significantly outperform individual human radiologists in pneumonia screening tasks. Thus, while previous research shows that a software-only solution like CheXNet can outperform individual radiologists, the current study explores if small groups of radiologists, when networked together as a real-time collaborative system moderated by AI algorithms, can amplify their collective accuracy to levels that rival or exceed the current state-of-the-art in purely algorithmic diagnosis.