Convolutional neural network for detecting rib fractures on chest radiographs: A feasibility study

Abstract

Abstract Background: The application of artificial intelligence for the detection of rib fractures on chest radiographs is limited by image quality control and multi-lesion screening. We aimed to create a model for multiple rib fracture detection using a convolutional neural network (CNN) based on quality-normalised chest radiographs.Methods: A total of 1,080 radiographs with rib fractures were obtained and randomly divided into training (918 graphs, 85%) and testing (162 graphs, 15%) sets. An object detection CNN, you only look once (YOLO) v3, was adopted to build the detection model. Receiver operating characteristic (ROC) and free-response ROC (FROC) were used to evaluate model performance. A joint testing group of 162 radiographs with rib fractures and 233 radiographs without rib fractures was used as the internal testing set. Furthermore, additional 201 radiographs, 121 with rib fractures and 80 without rib fractures, were independently validated to compare the CNN model performance with the diagnostic efficiency of radiologists.Results: The sensitivity of the model in the training and testing sets was 92.0% and 91.1%, respectively, and the precision was 68.0% and 81.6%, respectively. FROC in the testing set showed that the sensitivity for whole-lesion detection reached 91.3% when the false-positive of each case was 0.56. In the joint testing group, the case-level accuracy, sensitivity, specificity, and area under the curve were 85.1%, 93.2%, 79.4%, and 0.92, respectively. In the independent validation set, at the fracture level, the sensitivity of the CNN model (87.3%) was higher than that of the senior (80.3%) and junior radiologists (73.4%), while the precision (80.3%) was slightly lower than that of the latter two (82.4% and 81.7%, respectively). At the case level, the accuracy and sensitivity of the CNN model (91.5% and 96.7%, respectively) were both higher than those of the junior radiologist (85.1% and 77.7%, respectively) and close to those of the senior radiologist (94.0% and 96.7%, respectively). Conclusions: The CNN model based on YOLOv3 is sensitive for detecting rib fractures on chest radiographs and shows great potential in the preliminary screening of rib fractures.