Application And Reliability Analysis Of Artificial Intelligence Diagnosis System In The Classification Of Femoral Intertrochanteric Fractures

Objective: To evaluate the effectiveness of the self-developed artificial intelligence diagnostic system based on CT images for the classification of femoral intertrochanteric fractures,and evaluate its clinical application value.Methods: A total of 658 patients with femoral intertrochanteric fractures who were hospitalized from January 1,2017 to June 1,2020 were collected,of which 516 patients were included in the study,and they were exported one by one using the PACS(Picture Archiving and Counication System)system The CT images of the hip and 516 cases of femoral intertrochanteric fractures were classified into three types by a chief imaging physician and two orthopedic chief physicians under joint consultation.Consistent conclusions were obtained and the results were recorded as diagnostic criteria.The artificial intelligence diagnosis system used 436 cases of intertrochanteric fracture CT images(training set and validation set)for deep residual network training and verification,and the remaining 80 cases of femoral intertrochanteric fracture CT images were used as the test set to test the trained Model performance,and then an orthopedic resident and an orthopedic chief physician will independently perform three types of diagnosis under the condition of unknown diagnostic criteria,and then the orthopedic resident and the orthopedic chief physician will once again be assisted by the artificial intelligence diagnosis system.Diagnose,record the diagnosis result and diagnosis coincidence rate separately for statistical analysis.Then use orthopedic resident alone or assisted by artificial intelligence diagnosis system to perform six-part classification on the CT images of femoral intertrochanteric fractures,and conduct consistency test through Kappa value.Results: In the six-part classification of femoral intertrochanteric fractures,the diagnostic coincidence rate of the artificial intelligence diagnostic system was 77.6%,which was significantly lower than that of orthopedic residents(P<0.05)and 94.7% of orthopedic chief physicians(P<0.05).Under the system-assisted diagnosis,the diagnosis coincidence rate of the orthopedic resident increased from 88.2% to 94.7%(P<0.05),and the diagnosis coincidence rate of the chief orthopedic physician increased from 94.7% to 97.4%(P>0.05).In the Jensen-Evans classification,the diagnostic coincidence rate of the artificial intelligence diagnostic system was 82.5%,which was lower than 87.5% of orthopedic residents(P>0.05),and significantly lower than 95% of orthopedic chief physicians(P<0.05)Under the system-assisted diagnosis,the diagnosis coincidence rate of the orthopedic resident increased from87.5% to 95%(P>0.05),and the diagnosis coincidence rate of the chief orthopedic physician increased from 95% to 97.5%(P>0.05);in AO In the classification,the diagnosis coincidence rate of the artificial intelligence diagnosis system is 92.5%,which is not significantly different from 92.5%(P>0.05)of orthopedic residents,and lower than 97.5% of orthopedic chief physicians(P>0.05).The diagnostic coincidence rate of the orthopedic resident increased from 92.5% to 96.3%(P>0.05),and the diagnosis coincidence rate of the chief orthopedic physician increased from 95% to 98.8%(P>0.05).When the observer independently classifies the six parts of intertrochanteric fracture,the inter-observer reliability of the classification is highly consistent with the observer’s own reliability,and the Kappa coefficients are 0.715 and 0.714,respectively.When typing with the assistance of an artificial intelligence diagnosis system,the inter-observer reliability of the typing and the observer’s own reliability are the strongest,with Kappa values of 0.898 and 0.883.Conclusion: The artificial intelligence diagnosis system for femoral intertrochanteric fracture classification based on deep learning algorithms can realize automatic classification prediction and auxiliary diagnosis of AO classification,Jensen-Evans classification,and six-part classification.Orthopedic residents assist in the artificial intelligence diagnosis system.The following evaluation of the six-part classification of femoral intertrochanteric fractures can improve credibility and has certain clinical application value.