Object Detection AI Model Accurately Identifies Pediatric Upper Extremity Fractures

Background: Pediatric fractures can be quite challenging to identify, given the different response of the pediatric skeleton to injury compared to adults, and positioning challenges. Most artificial intelligence (AI) fracture detection studies and programs have focused on adults. Objective: To create and transparently share an AI model capable of detecting a range of pediatric fractures of the upper extremity. Methods: 58,846 upper extremity radiographs from 14,873 pediatric and young adult patients were used. These cases were divided into training (n=12,232 patients), tuning (n=1307), internal test (n=819), and external test (n=515) data sets. Fracture was determined by manual inspection of all test radiographs and the subset of train/tune radiographs with positive reports (classified by a rule-based natural language processing [NLP] algorithm). The train/tune data were used to train an object detection model (“strongly supervised”) and an image classification model (&ld more...