JMIR Publications

ABSTRACT

Background:

Rib fractures are present in 10–15% of thoracic trauma cases but are often missed on chest radiographs (CXRs), delaying diagnosis and treatment. Artificial intelligence (AI) may improve detection and triage in emergency settings.

Objective:

The objective of this study was to evaluate the real-world feasibility of an AI-assisted rib fracture detection system deployed within a high-volume emergency department. Specifically, we aimed to assess the system’s diagnostic performance, inference speed, operational scalability, and its potential role as a triage support tool embedded in routine clinical workflows. Additionally, we explored sources of false positives and examined infrastructure resilience as critical components for sustainable clinical integration of AI technologies.

Methods:

We conducted a prospective feasibility study of a Faster R-CNN–based AI model deployed in the emergency department to analyze 23,251 real-world CXRs (22,946 AP; 305 oblique) from April 1 to July 2, 2023. The AI operated passively, without influencing clinical decision-making. Ground truth was based on board-certified radiologist reports. A subset of discordant cases underwent post hoc CT review for exploratory analysis.

Results:

The AI achieved 74.5% sensitivity (95% CI, 0.708–0.780), 93.3% specificity (95% CI, 0.930–0.937), 24.2% positive predictive value (PPV), and 99.2% negative predictive value (NPV). Median inference time was 10.6 seconds versus 3.3 hours for radiologist reports (p < 0.001). Analysis revealed peak imaging demand between 08:00–16:00 and Thursday–Saturday evenings. A 14-day GPU outage underscored the importance of infrastructure resilience.

Conclusions:

The AI system demonstrated strong feasibility for real-time rib fracture triage in emergency care, offering potential benefits in case prioritization and resource optimization. Despite the need to reduce false positives from imaging artifacts, these findings support the role of AI-assisted triage in minimizing diagnostic delays, improving radiologist efficiency, and enhancing patient safety in high-volume clinical environments.