Development,validation And Artificial Intelligence Exploration Of Nomogram Predicting Femoral Head Osteonecrosis After Cannulated Screws Fixation For Femoral Neck Fractures

Objective: Femoral neck fracture is a common hip fracture.It is estimated that more than 1.6 million femoral neck fractures occur globally each year,and this number continues to rise as a result of an aging society.Osteonecrosis of the femoral head(ONFH)is the most common of the many complications after internal fixation of femoral neck fractures,accounting for 20-37.9% of all complications.Postoperative ONFH resulted in hip joint dysfunction and pain,resulting in a serious decline in the quality of life of the patients,and eventually necessitated joint replacement to maintain basic mobility.The purpose of this study was to assess perioperative variables for ONFH in patients with closed reduction of femoral neck fractures treated with cannulated screws,and to establish and validate an individualized nomogram to predict ONFH for early intervention and treatment.At the same time,we designed and validate an AI(Artificial Intelligence,AI)algorithm model by based on postoperative half-year X rays.Methods: In this study,671 patients with femoral neck fracture admitted to the First Affiliated Hospital of University of Science and Technology of China(N =481)and the Southern District of the First Affiliated Hospital of University of Science and Technology of China(n=190)from January 2009 to June 2017 were retrospectively collected.Postoperative diagnosis of ONFH relied on hip MRI or by three experienced orthopedic surgeons based on hip joint X-rays at the last follow-up.In the end,470 patients were enrolled in this study,divided into ONFH group and ONFH group.In the training cohort,Cox regression models were used to assess the prognostic value of multiple variables during the perioperative period.We established a nomogram for ONFH prediction using a multivariate logistic regression model,and evaluated the consistency and predictive power of the prediction model in the training cohort.We reassessed the nomogram’s performance in a validation sequence and evaluated its clinical application value.In the development of deep learning algorithm,we use MATLAB to develop a CNN model,which is used to calculate the abstract image features from the input image pixel array.This CNN model was trained through postoperative X-ray and output imaging variables as AI prediction indexes.We incorporated this prediction index into the routine clinical Nomogram and trained it to assess its predictive value.Results: Of the 470 patients who met the inclusion criteria,141(30%)were diagnosed with postoperative ONFH.Results showed that alcohol use(OR,1.743,95%CI,1.042-2.901,P=0.033)and cerebrovascular disease(OR,5.357,95%CI,2.318-13.13,P<0.001),injury-preoperative interval(OR,5.273,95% CI,2.724-10.43,P<0.001;Garden classification(OR,23.17,95% CI,6.812-145.3,P<0.001;Garden index(OR,5.935,95% CI,2.670-14.184,P<0.001;OR,5.935,95% CI,2.670-14.184,P<0.001),partial weight-bearing time after surgery(OR,0.053,95%CI,0.006-0.296,P=0.002)and Harris score six months after surgery(OR,0.856,95%CI,0.792-0.919,P<0.001)was identified as an independent predictor of postoperative ONFH.We developed a nomogram based on these variables.Our Nomogram showed good predictive power in both the training queue(AUC=0.865)and the validation queue(AUC=0.877),and proved the nomogram’s good performance in the validation queue.We used the CNN model of postoperative hip joint X-ray training to compare the reading and prediction abilities of an inexperienced orthopedic surgeon and an experienced orthopedic surgeon.The accuracy of CNN model validation set is 0.873,and the AUC value obtained by the algorithm is 0.912.Compared with a single clinical Nomogram,a mixed Nomogram model predicted better alignment of ONFH with actual observations.The AUC obtained by the hybrid Nomogram model based on deep learning was 0.948(95%CI,0.920-0.976),and the difference in area under the single clinical nomogram curve was statistically significant,indicating that the hybrid nomogram had better differentiation and prediction ability for the diagnosis of ONFH.Conclusions: We developed and validated an easy-to-use Nomogram to predict postoperative ONFH.The Nomogram provides patient,surgeon,and rehabilitation physician decision biases for intraoperative intervention and postoperative rehabilitation planning.At the same time,we applied deep learning neural network to ONFH prediction after femoral neck fracture surgery,and the results showed that the introduction of deep learning could improve the predictive ability of the traditional single clinical prediction model and increase the predictive value of the prediction model.