Objective To explore the value of deep learning(DL)models based on digital radiography(DR)in pre-dicting the efficacy of non-vascularized fibular grafting(NVFG).Methods A total of 339 patients(432 hips)with os-teonecrosis of the femoral head(ONFH)who underwent NVFG between June 2009 and June 2021 were included.They were divided into a training set(n=324)and a test set(n=108)in a ratio of 3:1.Standard anteroposterior and frog-lat-eral DR images were taken before surgery for both hips and distributed within the same dataset.The success or failure of preservation was defined based on the follow-up results at 2 years postoperatively.A DL model was constructed using Res-Net-50 as the backbone network.The model was trained and optimized in the training set,and its predictive performance was evaluated in the test set using metrics such as the area under the curve(AUC),accuracy(Acc),precision(Pre),re-call(Rec),and F1-score.Results As of June 2023,a total of 309 hips had excellent follow-up results,with a success rate of 71.52％for preservation.The model constructed using combined anteroposterior and frog-lateral DR images showed the best predictive performance for NVFG efficacy(P＜0.05),with an AUC of 0.780,Ace of 0.789,Pre of 0.787,Rec of 0.960,and F1-score of 0.865.The model based on anteroposterior DR had an AUC of 0.660,Acc of 0.662,Pre of 0.703,Rec of 0.900,and F1-score of 0.790.The model based on frog-lateral DR had an AUC of 0.710,Acc of 0.761,Pre of 0.762,Rec of 0.960,and F1-score of 0.850.Conclusion The DL model based on DR can accurately predict the efficacy of NVFG,which is worthy of application to clinical practice.
Deep learningOsteonecrosis of the femoral headNon-vascularized fibular graftingPrecision medicineDigital radiography
NETL
NSTL
万方数据
