Comprehensive Summary
This study evaluated a deep learning (DL) model based on a convolutional neural network (CNN) for classifying elbow fractures using the 2018 AO/OTA classification system. Radiographs were first manually classified by orthopedic surgeons, then used to train the model. Performance was evaluated against a test set independently reviewed by orthopedic surgeons. The model achieved an AUC of 0.88 and performed well in identifying and classifying fractures, though performance declined with more subtle fractures.
Outcomes and Implications
DL models for elbow fracture classification can improve diagnostic accuracy and expedite intervention. Delaying treatment or misclassifying elbow fractures may result in a lower range of motion and a higher risk for post-traumatic arthritis. Such models have the potential to reduce missed fractures and ultimately lead to enhanced clinical outcomes. The model was tested on non-standard radiographic images, reflecting real-world clinical variability and enhancing clinical relevance. Prior to clinical implementation, more external validation and clinical assessments must be performed to establish reliability across clinical settings and imaging variability.