Real-life benefit of artificial intelligence-based fracture detection in a pediatric emergency department

Comprehensive Summary

In this study, researchers evaluated the performance of RBFracture, an AI-based software, in fracture detection in a pediatric emergency department. A real-life dataset of 1672 radiographs was identified, with 1061 cases being upper-extremity and 611 being lower-extremity. The RBFracture software evaluated each case, highlighting potential fractures in the radiographs. All radiographs were also evaluated by two pediatric surgical residents and a radiology resident, before they were presented with the AI assessment. The AI software achieved sensitivity ranging from 74-98%, specificity ranging from 61-100%, and accuracy ranging from 80-94%. Additionally, the resident physicians were able to accurately diagnose 87.5% of patients without AI assistance. The AI tool was also able to reduce frequency of missed fractures from 13.8% to 10.7%, improve sensitivity from 83.7% to 87.3%, and improve specificity from 90.7% to 92.4%. However, despite being able to improve initially incorrect diagnoses in 4.1% of cases, AI also led to incorrect alterations of initially correct diagnoses in 1.5% of cases. Overall, findings show that AI software displays sufficient stand-alone performance and can improve confidence in less-experienced residents. However, further work is required to decrease inaccuracies and increase applicability to more real-world settings.

Outcomes and Implications

The implementation of AI software for fracture detection in an emergency department setting offers several benefits for both providers and patients. First, as described in the study, support from AI offers training and extra support for resident physicians or providers with less experience. This can lead to improved diagnostic accuracy, which in turn produces better patient outcomes. Additionally, AI can improve timeliness of diagnosis, which is extremely important in a fast-paced environment such as an emergency department. Continued research and development on AI software in this context can help improve patient care and fracture detection long-term.