Purpose or Case Report: To assess whether a convolutional neural network (CNN) can be trained via transfer learning to accurately diagnose metaphyseal corner fractures on long bone radiographs by comparing its performance to that of a board-certified pediatric radiologist. Secondary endpoints will include the ability of the CNN to detect the type of long bone presented on the radiograph as well as identify the chronicity of the fracture.
Methods & Materials: In this IRB approved retrospective study, we reviewed records from our children hospital’s radiology database. Consecutive patients, 12 months in age or younger, from January 2009 to September 2019 were screened for bone surveys in the evaluation of suspected non-accidental trauma performed at our institution. Patients with inadequate or incomplete radiographic examinations due to poor patient positioning or overlying artifact were excluded.

A total of 132 frontal radiographs of the femur were retrospectively utilized for the training and testing of a CNN binary classifier. Radiographs were obtained from 61 patients who were found to have femoral metaphyseal corner fractures and from 71 patients with normal femoral metaphyses and no metaphyseal corner fractures. Twenty cases were randomly allocated from the total for validation and an additional 20 for testing. We utilized transfer learning (VGG16) and data augmentation in our algorithm due to the low sample size. The model was assessed using the test dataset with area under the Receiver Operating Characteristic curve and accuracy.
Results: Using a small training dataset of 82 images, we achieved an accuracy of 95% with both validation (n = 20) and test (n = 20) datasets. Our model achieved an area under the Receiver Operating Characteristic curve of 0.94.
Conclusions: A CNN trained on a small set of images with data augmentation is able to detect metaphyseal corner fractures on long bone radiographs with relatively high accuracy. We predict that further analysis with segmentation using a larger dataset will further improve accuracy.