Prospective Diagnostic Accuracy and Technical Feasibility of Artificial Intelligence-Assisted Rib Fracture Detection...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Emergency Medicine

Background:

• Rib fractures occur in 10%-15% of thoracic trauma cases and are frequently missed on chest radiographs.
• Delayed diagnosis of rib fractures can impede timely treatment.

Purpose of the Study:

• To evaluate the diagnostic accuracy, processing speed, and technical feasibility of an AI-assisted rib fracture detection system.
• To assess the AI system's performance in a real-world, high-volume emergency department workflow.

Main Methods:

• An observational feasibility study utilized a region-based convolutional neural network AI model.
• Analyzed 23,251 chest radiographs collected prospectively in an emergency department setting.
• Compared AI performance against final reports from board-certified radiologists, with CT review for discordant cases.

Main Results:

• AI achieved 74.5% sensitivity and 93.3% specificity for rib fracture detection.
• Median AI inference time was 10.6 seconds, significantly faster than the 3.3 hours for radiologist reports.
• The AI system demonstrated high negative predictive value (99.2%).

Conclusions:

• The AI system is technically feasible for real-time rib fracture detection in emergency departments, offering rapid analysis.
• The AI system functions effectively as a supportive screening tool but should not replace clinical judgment.
• Further studies are recommended to explore clinical integration and impact on diagnostic decision-making.