Artificial Intelligence (AI) has increasingly integrated into both everyday life and professional practices, particularly as a support tool for radiological diagnosis in emergency departments.

Objective

This study aimed to assess the diagnostic performance of AI-assisted radiographic software (Deep Unity Gleamer BoneView) in detecting fractures and dislocations of the metacarpals and phalanges.

Methods and Materials

The study was diagnostic, retrospective and monocentric. The population consisted of any patient admitted to the emergency department with hand trauma for whom a hand radiograph was acquired.

Each radiograph was independently analyzed by two senior hand surgeons to determine the presence or absence of fractures/dislocations. Two senior hand surgeons independently reviewed each radiograph In cases of discordant findings, a consensus gold standard (GS) was established after a re-evaluation incorporating clinical records. The diagnostic performance of the AI was compared against the GD, and was expressed as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) with 95% confidence interval (CI).

A total of 1,915 hand X-ray records were analyzed (1,892 patients or 4,738 X-rays analyzed) between December 2022 and January 2024. The mean patient age was 40.6 years [16-99]. The GD identified 457 fractures/dislocations and 1458 radiographs without lesions. AI detected 608 fractures/dislocations and 1307 radiographs without lesions. The AI achieved a sensitivity of 97.6% (95% CI, 95.7-98.7), specificity of 88.9% (87.2-90.4), PPV of 73.4% (69.8-76.9) and NPV of 99.2% (98.7-99.7). Upon re-evaluation, 11 false negatives were identified.

Results

AI has high diagnostic performance with excellent sensitivity in the detection of metacarpal and phalangeal fracture/dislocation in the adult population, as has also been shown in the literature for carpal and distal radius lesions.

These findings demonstrate that AI-assisted radiographic software exhibits excellent sensitivity and overall diagnostic performance in detecting metacarpal and phalangeal fractures/dislocations, consistent with its established utility for carpal and distal radius injuries.

Conclusions

While AI proves to be a valuable tool in emergency settings, it should serve as an adjunct to, rather than a replacement for, clinical expertise and thorough patient evaluation.

The full article is available in French only.