Automated Association for Osteosynthesis Foundation and Orthopedic Trauma Association classification of pelvic fractures on pelvic radiographs using deep learning

Abstract

High-energy impacts, like vehicle crashes or falls, can lead to pelvic ring injuries. Rapid diagnosis and treatment are crucial due to the risks of severe bleeding and organ damage. Pelvic radiography promptly assesses fracture extent and location, but struggles to diagnose bleeding. The AO/OTA classification system grades pelvic instability, but its complexity limits its use in emergency settings. This study develops and evaluates a deep learning algorithm to classify pelvic fractures on radiographs per the AO/OTA system. Pelvic radiographs of 773 patients with pelvic fractures and 167 patients without pelvic fractures were retrospectively analyzed at a single center. Pelvic fractures were classified into types A, B, and C using medical records categorized by an orthopedic surgeon according to the AO/OTA classification system. Accuracy, Dice Similarity Coefficient (DSC), and F1 score were measured to evaluate the diagnostic performance of the deep learning algorithms. The segmentation model showed high performance with 0.98 accuracy and 0.96–0.97 DSC. The AO/OTA classification model demonstrated effective performance with a 0.47–0.80 F1 score and 0.69–0.88 accuracy. Additionally, the classification model had a macro average of 0.77–0.94. Performance evaluation of the models showed relatively favorable results, which can aid in early classification of pelvic fractures.