Upper extremity trauma is an ordinary chief complain for children presenting to the pediatric emergency room, which underlies the high frequency of elbow fractures in this population, accounting for up to 10% of all pediatric fractures [1]. Elbow radiographs are still the first-choice imaging modality in this condition. However, they are often diagnostically challenging given the complex anatomy of this joint and its multiple cartilaginous ossification centers associated with a number of normal variants in growth plate appearance that can mimic fractures [2]. Furthermore, nondisplaced supra condylar fractures can be difficult to detect directly and manifest as a joint effusion on lateral radiographs, marked with subtle changes in anterior and/or posterior fat pads [3]. On normal radiographs only the anterior fat pad is seen close to the humerus, while the posterior fat pad is hidden in the olecranon fossa.

Artificial intelligence (AI)-based algorithms show promise in real-life clinical triage and could be helpful in the field of pediatric trauma to direct towards either orthopedic evaluation or discharge. This could overcome the growing number of emergency department visits and the lack of trained on-site pediatric radiologists [4], [5], [6], [7]. Indeed, emergency physicians can miss up to 11% of acute pediatric fractures compared to pediatric radiologists, leading to adverse events and changes in management [8], [9], [10]. Furthermore, few studies have recently reported that deep-learning algorithms can accurately detect traumatic pediatric elbow joint effusion [11], [12], [13].

The purpose of this study was to conduct an independent external validation of a commercially available AI solution for the detection of elbow fractures and joint effusions using digital radiographs from a real-life cohort of children presenting to the emergency room.