Application of artificial intelligence in the diagnosis of scaphoid fractures: impact of automated detection of scaphoid fractures in a real-life study

Comprehensive Summary

This study evaluates whether two commercially available AI tools (BoneView and RBfracture) are able to help non-specialist radiologist (NSR) detect scaphoid fractures using standard wrist X-ray scans. 764 radiographs from 264 patients with hand or wrist trauma were gathered. The patients were divided into 2 groups (definite and complex cases). Group 1 included the definite cases with clear or no fracture on X-ray, confirmed by specialist radiologist. Group 2 included the complex cases which required CT scans for final diagnosis. The radiographs were analyzed by AI and two NSRs independently and together. For Group 1, the AI and NSRs performed similarly with no significant improvement when the AI was used by the NSRs. For Group 2, even though the overall performance of AI and NSRs were lower, the use of AI by the NSRs significantly improved their performance. These findings highlight AI tools’ usefulness in assisting NSRs, particularly in ambiguous scaphoid fracture cases.

Outcomes and Implications

Scaphoid bone is one of the most important carpal bones, which plays an important role in motion and stability of the wrist joint. On top of being the most common type of carpal fracture, scaphoid fractures are difficult to diagnose, potentially leading to misdiagnosis and eventually severe complications. Artificial intelligence research has lead to development of AI programs that assist radiologist in diagnosing fractures using radiographs. Even though they have been integrated into clinical practice, they still require much testing under specific conditions such as scaphoid fracture scenarios.