Value of clinical-radiomic model to the prediction of severe bleeding after percutaneous nephrolithotomy for renal ureteral stones
Objective To construct a clinical-radiomic model for predicting severe bleeding after percutaneous nephrolithotomy(PCNL)in patients with renal ureteral stones,and to explore its predictive value.Methods Totally 130 patients underwent PCNL in the Second Affiliated Hospital of Nanchang University from April,2021 to August,2022,and were divided into bleeding group(n=39)and non-bleeding group(n=91)according to the presence or absence of severe bleeding 24 h postoperatively.The clinical data as the location,number and size of stone were compared between two groups.Multivariate logistic regression was employed to identify the influencing factors of severe bleeding after PCNL in patients with renal ureteral stones,and a clinical prediction model was constructed.All 130 patients underwent CT scans,and non-contrast enhanced CT images were processed to delineate the stone region of interest.Radiomic features were extracted following preprocessing steps such as filtering,log transformation and resampling,lasso regression was used for feature selection,and multivariate logistic regression analysis was performed to identify the radiomic influencing factors of severe bleeding after PCNL in patients with renal ureteral stones,and to construct a radiomic prediction model.A combined clinical-radiomic prediction model was constructed by integrating clinical and radiomic influencing factors,and ROC curves were plotted to evaluate the predictive performance of both models for severe bleeding after PCNL in patients with renal ureteral stones.The Delong test,decision curve and calibration curve were employed to compare the predictive efficacies of two models.Results(1)The cast stone percentage,and number and diameter of stones were higher in bleeding group[87.2％,4(3,5),2.7(2.3,3.0)cm]than those in non-bleeding group[24.2％,3(1,4),1.7(1.3,2.3)cm](P＜0.05),and the percentages of stones locating in the ureter and kidney,primary surgery and single-channel surgery were lower in bleeding group(10.3％,51.3％,79.5％,87.2％)than those in non-bleeding group(20.9％,63.7％,98.9％,98.9％)(P＜0.05).(2)The stone shape(OR=25.481,95％CI:5.408-120.054,P＜0.001)and surgical stage(OR=19.486,95％CI:1.210-313.687,P=0.036)were the clinical influencing factors of severe bleeding after PCNL in patients with renal ureteral stones.The clinical model logistic(Pi)=-2.888+3.109 X cast stone+3.297 X second-stage surgery.(3)Lasso regression analysis revealed seven radiomic features which were mostly highly correlated with severe bleeding after PCNL in patients with renal ureteral stones as exponential_firstorder_Median,lbp-2D_firstorder_90Percentile,wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis,wavelet-LLH_ngtdm_Busyness,wavelet-LHH_glszm_SmallAreaEmphasis,wavelet-HLH_firstorder_Skewness and wavelet-HHL_glrlm_LongRunHighGrayLevelEmphasis.Wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis(OR=24.970,95％CI:1.351-461.611,P=0.031)and wavelet-HLH_firstorder_Skewness(OR=10.671,95％CI:1.192-95.509,P=0.034)were the radiomic influencing factors of severe bleeding after PCNL in patients with renal ureteral stones.The radiomic model logistic(P2)=-2.058+2.086 × wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis+1.547 × wavelet-HLH_firstorder_Skewness.(4)The clinical-radiomic model logistic(P3)=-5.383+3.695 × wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis+2.812 × wavelet-HLH_firstorder_Skewness+3.613×cast stone+3.180× second-stage surgery.When the optimal cut-off values of the clinical-radiomic model and the clinical model were 0.416 and 0.304,the AUCs for predicting severe bleeding after PCNL in patients with renal ureteral stones were 0.900(95％CI:0.839-0.961,P＜0.001)and 0.844(95％CI:0.770-0.918,P＜0.001),respectively.Compared with the clinical prediction model,the clinical-radiomic prediction model had better clinical net benefits.The clinical-radiomic prediction model was superior to the clinical model in predicting severe bleeding after PCNL in patients with renal ureteral stones.The clinical-radiomic prediction model had good fitting and calibration capabilities.Conclusion The second-stage PCNL,cast stone,wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis,and wavelet-HLH_firstorder_Skewness increase the risk of severe bleeding after PCNL in patients with renal ureteral stones,and the clinical-radiomic prediction model based on clinical characteristics and radiomic features has a high predictive value.
urinary tract stonespercutaneous nephrolithotomysevere bleedingclinical modelradiomic modelclinical-radiomic model
NETL
NSTL
万方数据
