Machine Versus Man: Artificial Intelligence Diagnostic Accuracy in Fracture Diagnosis

Abstract

Introduction: Yearly around 21 thousand adult patients visit our tertiary hospital’s emergency department after suffering from high or low energy trauma. Skeletal radiographs, being inexpensive and widely available, are the first‐line imaging mo‐ dality. Recent studies are showing encouraging results of the use of artificial intelligence in the detection of bone fractures. The main objective of this study is to compare the diagnostic accuracy between a medical‐grade artificial intelligence (AI) software (BoneView®, Gleamer) and orthopaedic surgeons of various levels of expertise for the detection of bone fractures in a tertiary hospital’s emergency department.
Methods: Retrospective analysis of a series of posttraumatic radiographic examinations, including only adult patients with plain radiographs of limbs or pelvis obtained after a recent trauma. Exclusion criteria were patients with cast control radiographs, images with inadequate radiographic quality, and examinations showing only obvious fractures. The diagnostic performance of the AI software and six orthopaedic surgeons was measured by sensitivity, specificity, and area under the receiver operating characteristic curve (AUC).
Results: The AI software had 91.3% sensitivity (95% CI: 82.03‐96.74) and 97.3% specificity (95% CI: 93.22‐99.26), with 0.95 AUC (95% CI: 91.3‐98.8; p <0.001). All six readers had inferior results in every measure obtained, with slight differences between them.
Conclusion: Our study demonstrated that the BoneView® software has a high diagnostic capacity for fractures and, in this regard, can be considered a useful tool in the emergency department.