A machine learning model based on CT texture features for predicting the efficacy of initial ESWL treatment for ureteral stones
Objective To investigate whether a machine learning model based on CT texture features can accurately assess the therapeutic efficacy of the first extracorporeal shock wave lithotripsy(ESWL)in patients with urinary calculi.Methods A total of 317 patients with urinary calculi were included in this study.These patients underwent multidetector CT(MDCT)examination 2 weeks before the first ESWL and were divided into an internal cohort(Shenzhen People's Hospital,250 patients)and an external cohort(Shenzhen Luohu District People's Hospital,67 patients)according to their origin.An automated semantic segmentation model was built using the 3D-UNet algorithm,and the degree of overlap between the predicted segmentation and the real segmentation was quantified by calculating the Dice coefficient.A machine learning model for the therapeutic efficacy of ESWL for urinary calculi was built based on texture features and XGBoost algorithm.Strict five-fold cross-validation and multi-center external testing strategies were employed to verify the stability and generalization performance of the model.Additionally,the SHAP algorithm was used to explore the contribution of each feature to the model decisions.Results The average Dice coefficient of the external cohort was(0.88±0.08).For the prediction model of ESWL efficacy for urinary calculi(Success VS Failure),the area under receiver operating characteristic curve(AUROC)values of the five-fold cross-validation were 0.91,0.89,0.87,0.88,and 0.92,with accuracies of 0.84,0.78,0.76,0.81,and 0.84.The AUROC values of the validation set and test set for the optimal model were 0.92 and 0.84,respectively,both falling within the 95%CI.Whether morphological features,first-order statistical features,or textural features,all the features input to the model played a role in the decision making of the model,and the gray scale correlation matrix with high gray scale dependence(gldm_LDHGLE),first-order Minimum,and shape Elongation played a dominant role in the prediction model of the efficacy of the first ESWL performed for urinary tract stones(Success VS Failure)played a dominant role in the prediction model.Conclusion The prediction model based on CT texture features can accurately assess the therapeutic efficacy of the first ESWL for urinary calculi.
NETL
NSTL
万方数据
