Study On AI Rapid Diagnosis Method Of Human Femoral Neck Fracture Image

Objective:Aiming at the deficiency of traditional manual diagnosis methods for femoral neck fracture and work needs to realize the automatic assessment of femoral neck fracture,artificial intelligence technology is used to identify femoral neck fracture,and a visual deep learning model for automatic detection of femoral neck fracture is constructed,and the diagnostic efficiency of the model is evaluated,so as to realize automatic detection of femoral neck fracture and improve the diagnostic efficiency of femoral neck fracture.Methods:The CT image data samples of 1018 cases of human femoral neck fracture in Beijing Jishuitan Hospital were collected.According to the ratio of 7:1:2,1018 cases of CT image samples were divided into training set,verification set and test set,including 676 cases of training set,112 cases of verification set and 230 cases of test set.The construction of femoral neck fracture evaluation model mainly includes three parts: the first step is to develop an automatic image preprocessing module,and to avoid the influence on the effectiveness of fracture diagnosis model due to the inconsistency of image metadata information such as slice thickness and slice ordering through image format conversion,resampling,bone window conversion and normalization,and the second step is to build a femoral neck fracture diagnosis model based on Dense Net.By taking the cross-sectional image of femoral neck as the input of Dense Net network model,and taking the fracture situation of each femoral neck marked by orthopedic experts in Jishuitan Hospital as the label,an intelligent diagnosis model of femoral neck fracture based on Dense Net is constructed.The third step is to evaluate the performance of the diagnosis model of femoral neck fracture.After training the model on the training set,the diagnostic ability of the model is tested on the testing set.The output results of the tested model are compared with the results judged by orthopedic experts,and four test results of the model are obtained: true positive(TP),true negative(TN),false positive(FP)and false negative(FN).Through calculation,several evaluation indexes such as Precision,Recall,Accuracy and F1-Score are obtained to evaluate the model performance.Results:After independently testing the image data of 230 cases in the test set,by comparing the output results of the model with the results judged by orthopedic experts,TP: 71 cases,TN: 110 cases,FN: 41 cases and FP: 8 cases were obtained.After calculation,the Precision was 89.87%,Recall was 63.39%,Accuracy was 78.70% and F1 Score was 74.34%.In the test set containing 230 samples,there are 49 samples with incorrect model judgment,including 19 cases with incorrect fracture position judgment,8 cases with normal fracture judgment and 22 cases with normal fracture judgment.After that,200 samples were randomly selected from 230 samples in the test set and entrusted to a third-party testing institution for further verification.After that,200 samples were randomly selected from 230 samples in the test set and entrusted to a third-party testing institution for further verification.After many tests,the average accuracy of fracture detection in this workstation was 72.0%.Conclusion:The established intelligent evaluation model of femoral neck fracture has reached the level of experienced artificial femoral neck fracture,and it may be used to assist the identification of femoral neck fracture after further optimization.