Research on the Application Value of Artificial Intelligence Assisted Detection of Calcaneal Traumatic Fractures

Background: A deep learning model was constructed to detect traumatic calcaneal fracture based on radiographs, and its performance was evaluated according to the level of orthopedics experts. Method: 15800 lateral radiographs of the calcaneus were included, which include12300 normal images and 3500 fracture images. The detection performance of the model was evaluated, including the identification of fractures, the morphological measurement of Bohler's angle, Gissane's angle, length and height, the judgment of fracture types based on the Essex-Lopresti fracture classification, the evaluation of fracture lines, the the identification of fracture area and other capabilities, and the study of generalization capabilities of the model. Results: The mean angle errors(MAE) of BA and GA are 3.90° and 5.37°, and the mean distance errors(MDE) of length and height are 2.44mm and 1.53mm. The accuracy rates of fracture recognition in groups A, B and C are 95.10% and 94.67%, respectively. , 86.92%, AUC values were 99.48, 98.18, 97.47, the accuracy of fracture type judgment was 73.20%, 73.24%, 73.03%, and the average accuracy of fracture line or fragment evaluation was 81.94%, 81.50%, 74.71%, respectively ; the accuracy rates of fracture accounted for regional identification were 94.41%, 88.51% and 87.26% respectively. Interpretation: Deep learning can use ordinary plain radiographs to detect and evaluate traumatic calcaneal fractures, and the performance is close to the level of orthopedic experts, and it has good generalizability. Funding Statement: The research is supported by the school-enterprise cooperation fund. Declaration of Interests: The authors declare that they have no conflicts of interests. Ethics Approval Statement: The study was approved by the Ethics Committee of the Affiliated Hospital of Jinzhou Medical University.