| Purpose:Obstructive nephropathy is a common chronic kidney disease,and urinary calculi are more common cause in adults.Surgical treatment is the main way to save the residual renal function of patients with obstructive nephropathy.At present,there are no specific molecular biological markers or predictive models,which can quickly and accurately predict the recoverability of renal function in patients with obstructive nephropathy after surgical treatment.Radiomics is a new and cutting-edge image processing technology,which can extract image features that can’t be recognized by human eyes and transform them into quantitative data.Models which are established by using radiomics data alone or in combination with demography,metabolism,genome or proteome data can provide a new entry point for disease diagnosis,differentiation and treatment.Therefore,this study intends to establish a radiology-based model through clinical data combined with traditional CT imaging data to predict the recovery of renal function after relieving of obstruction in patients with obstructive nephropathy,and evaluate its feasibility and accuracy to find the optimal prediction model in this study by comparing different models.At the same time,the radiological features and clinical features were analyzed by logistic regression analysis to explore the risk factors affecting the recovery of renal function in patients with obstructive nephropathy.Methods:A total of 275 patients with obstructive nephropathy caused by urinary calculi were searched in Haitai Electronic Medical record system of Xiangya Hospital of Central South University from April 2020 to October 2021.65 patients were selected according to the inclusion and exclusion criteria.The demographic data,laboratory results,CT imaging data within 3 months before operation and serum creatinine levels within3 months after operation were collected retrospectively.The postoperative decrease of serum creatinine ≥ 20% compared with the preoperative group was defined as the renal function improvement group,and the postoperative serum creatinine decreased less than 20% or did not decrease or even increased compared with the preoperative group,which was defined as the group with no improvement of renal function.The region of interest(ROI)on CT imaging was drawn manually by ITPSNAP software,then the features of manual radiology(HCR)were analyzed.The radiological and clinical features screened by lasso regression are used to construct machine learning models.The training set and test set are randomly assigned according to the proportion of 7:3.Eight different machine learning algorithms are used to establish the models.The area under the curve(AUC)is calculated by receiver operating characteristic(ROC)curve,and the accuracy and precision degree of models are evaluated by 50% discount cross-validation.Finally,the prediction effects of different algorithms are compared to find the best prediction model.At the same time,the radiological and clinical features were analyzed by univariate logistic regression analysis,and the significant indexes in univariate logistic regression were analyzed by multivariate logistic regression analysis to explore the risk factors affecting the recovery of renal function in patients with obstructive nephropathy after relief of obstruction.Results:A total of 65 cases were included in this study,including 37 cases in remission group and 28 cases in non-remission group.46 cases of training set and 19 cases of test set were randomly assigned at 7:3,and 8 different machine learning algorithms were used to establish prediction models.In the training set,the model constructed by Gradient boosting(GB)algorithm showed the best prediction performance in all classifier modeling(AUC= 0.83 ±0.22),with good accuracy(Accuracy=0.94)and precision degree(Precision degree=0.96),and also had good prediction performance(AUC=0.80)in the training set.In univariate logistic regression analysis,four imaging features including wavelet＿HLL＿glrlm＿Run Entropy＿1,wavelet＿HHL＿glcm＿Cluster Prominence＿1,original＿glszm＿Size Zone Non UniformityNormalized＿1 and wavelet＿HHL＿glrlm＿Run Entropy＿1 were the prognostic factors of obstructive nephropathy,all of which were texture features.Gender,serum creatinine and blood urea nitrogen in clinical data were statistically significant in univariate logistic regression.Further multivariate logistic regression showed that wavelet＿HLL＿glrlm＿Run Entropy＿1(each additional unit)(OR=0.11,P=0.001)was an independent risk factor for prognosis after obstructive nephropathy was relieved.Gender(male)and preoperative creatinine were statistically significant in multivariate logistic regression.Conclusion:This study proved that it is feasible and effective to predict the recovery of renal function after relieving obstruction in patients with obstructive nephropathy by combining radiological features based on CT with clinical data and using machine learning method.The model constructed by GB algorithm has the best prediction performance in all classifier modeling,and also has better accuracy and precision degree.Radiological features(wavelet＿HLL＿glrlm＿Run Entropy＿1)can be used as an independent risk factor to affect the recovery of renal function in patients with obstructive nephropathy,which is helpful for clinicians to make reasonable treatment decisions for patients with obstructive nephropathy.Figures:5,tables:9,references:64... |
PDF Full Text Request