ObjectiveTo construct a novel clinical-radiomics model of kidney stone for predicting the stone-free rate (SFR) of percutaneous nephrolithotomy (PCNL) and validate it.

MethodsThe clinical and imaging data from 113 patients undergoing PCNL were retrospectively collected, and divided into stone removal group and stone residue group according to the results of the CT scan or X-ray of kidney-ureter-bladder review after operation.The clinical and radiomics data of patients were collected, the volume of interest was drawn on the CT images and 120 radiomics features were extracted using image analysis software and computer programming language tool.The variables of the training group were selected to obtain the best feature selection.A novel clinical-radiomics model was established by multivariate logistic regression analysis, and the area under the curve (AUC) was used to evaluate the predictive effect of the model.

ResultsThe 113 patients underwent a follow-up examination of urological imaging one month after surgery, among them, 68 cases had stones removed and 45 cases had residual stones, and the overall SFR was 60.2%.There were statistically significant differences in gender, postoperative blood white blood cell (WBC), hospital stay, stone length, stone width, and Guy's score (GSS) of patients between the two groups (P < 0.05 to P < 0.01).Univariate logistic analysis showed that there were statistically significant differences in gender, GSS, postoperative blood WBC, stone length, and stone width (P < 0.05 to P < 0.01).Multivariate logistic analysis showed that gender, GSS, and postoperative blood WBC were independent predictors of postoperative SFR of PCNL (P < 0.05 to P < 0.01).The 14 meaningful radiomics features selected by Lasso regression were analyzed using univariate and multivariate logistic analysis, the results showed that there were statistically significant differences in maximum three-dimensional diameter and sphericity (P < 0.05 and P < 0.01).The maximum three-dimensional diameter and sphericity were used to construct a clinical-radiomics prediction model, the AUC in the training set was 0.923, which in the validation set was 0.876, and both was better than GSS.

ConclusionsThe constructed novel clinical-radiomics model combined with clinical features can help to predict the SFR in PCNL patients, and provide reference for urologists in preoperative communication and selection of surgical methods.