AI for Automated Identification of Carpal Instability on Radiographs

Overview

This study developed and validated an AI system to automatically measure and detect signs of carpal instability on conventional wrist radiographs. The AI demonstrated accurate assessments of scapholunate joint distance, scapholunate and capitolunate angles, and carpal alignment, comparing favorably to clinicians across specialties.

Background

Carpal instability occurs when carpal bones lose normal alignment under physiological loads, often due to acute trauma such as ligament ruptures or fractures. Early identification is critical to prevent progression to scapholunate dissociation and advanced collapse. Conventional radiography is the first-line imaging modality but signs of instability are frequently overlooked due to measurement variability and clinician unfamiliarity. Automated AI-based assessment could improve detection and clinical decision-making.

Data Highlights

Key Findings

• AI system accurately measured scapholunate (SL) joint distance, SL and capitolunate (CL) angles, and carpal alignment on standard radiographs.
• Thresholds for abnormal SL joint distance were set at 3.0 mm (neutral view) and 3.7 mm (ulnar-deviated view), with age-adjusted values for children.
• AI performance was validated on a balanced dataset including normal and abnormal cases, demonstrating reliability across multiple radiographic views.
• Observer study showed AI measurements were comparable to or exceeded accuracy of clinicians from various specialties.
• Automated detection addresses human variability and unfamiliarity with carpal measurements, potentially reducing overlooked instability signs.

Clinical Implications

The AI system provides a reliable, automated tool for early detection of carpal instability on conventional radiographs, which are commonly the first imaging modality after wrist trauma. This can assist clinicians, including non-specialists, in identifying subtle signs of ligament injury and abnormal carpal alignment, facilitating timely management to prevent progression to advanced collapse.

Conclusion

Automated AI assessment of carpal instability indicators on standard wrist radiographs is feasible and accurate, offering a valuable adjunct to clinical evaluation. Integration of such tools may improve detection rates and patient outcomes in wrist trauma care.

References

1. Dornberger et al 2020 -- Thresholds for Scapholunate Joint Distance
2. Kaawach et al 2019 -- Age-Adjusted Normal Values for SL Joint Distance
3. Radboud University Medical Center et al -- Study on AI for Carpal Instability