AI for Detecting Scaphoid Fractures on X-ray: Systematic Review & Meta-analysis

Overview

This systematic review and meta-analysis evaluated artificial intelligence (AI) algorithms, specifically convolutional neural networks (CNNs), for detecting scaphoid fractures on wrist radiographs. The analysis of ten retrospective studies demonstrated that AI can improve diagnostic accuracy and sensitivity compared to traditional radiographic interpretation, potentially reducing missed fractures.

Background

Scaphoid fractures are the most common carpal bone fractures and require early detection to prevent complications such as avascular necrosis and osteoarthritis. Conventional wrist X-rays have limited sensitivity (66–81%) and are prone to missed diagnoses, especially when interpreted by less experienced clinicians. Advanced imaging like CT and MRI offer higher sensitivity but have drawbacks including radiation exposure and cost. AI, particularly CNNs, has shown promise in medical image analysis and may enhance scaphoid fracture detection on plain radiographs.

Data Highlights

Key Findings

• AI models, primarily CNNs, demonstrated good to excellent diagnostic accuracy with AUC values mostly above 0.8.
• AI improved sensitivity for detecting scaphoid fractures compared to standard radiographic interpretation, addressing the issue of occult fractures.
• Studies included retrospective datasets ranging from 356 to 11,838 wrist radiographs, reflecting diverse sample sizes.
• AI algorithms can assist less experienced clinicians and reduce diagnostic errors caused by fatigue or limited radiologist availability.
• Despite promising results, heterogeneity exists among studies regarding AI architectures, labeling methods, and reference standards.
• AI offers a cost-effective adjunct to radiographs, potentially reducing the need for expensive or radiation-intensive imaging modalities like MRI or CT.

Clinical Implications

Integrating AI into clinical workflows could enhance early detection of scaphoid fractures on plain radiographs, improving patient outcomes by enabling timely immobilization and treatment. AI support may reduce diagnostic errors in emergency settings, especially during off-hours or in facilities with limited specialist availability. Adoption of AI tools should consider validation in diverse clinical environments and integration with existing diagnostic protocols.

Conclusion

AI-based CNN models show promising diagnostic accuracy for detecting scaphoid fractures on wrist X-rays, potentially improving sensitivity and reducing missed fractures. Further prospective studies and clinical implementation research are warranted to optimize AI integration into routine practice.

References

1. Langerhuizen et al. 2023 -- AI for Scaphoid Fracture Detection
2. PRISMA Guidelines 2020 -- Systematic Review Reporting Standards
3. QUADAS-2 Tool 2016 -- Quality Assessment of Diagnostic Accuracy Studies