CT影像组学机器学习模型预测逆行输尿管软镜碎石术后泌尿系结石清石率  

CT radiomics machine learning model for predicting stone free rate of urinary calculi after retrograde intrarenal surgery

作　　者：周聪 王亚洲[1] 吴青霞 朱永月 廖雯欣 王道清[1] ZHOU Cong;WANG Yazhou;WU Qingxia;ZHU Yongyue;LIAO Wenxin;WANG Daoqing(Department of Radiology,the First Affiliated Hospital of Henan University of CM,Zhengzhou 450046,China;Department of First School of Clinical Medicine,Henan University of Chinese Medicine,Zhengzhou 450046,China;Beijing United Imaging Research Institute of Intelligent Imaging,Beijing 100089,China)

机构地区：[1]河南中医药大学第一附属医院放射科,河南郑州450046 [2]河南中医药大学第一临床医学院,河南郑州450046 [3]北京联影智能影像技术研究院,北京100089

出　　处：《中国介入影像与治疗学》2025年第1期52-57,共6页Chinese Journal of Interventional Imaging and Therapy

摘　　要：目的观察CT影像组学机器学习(ML)模型预测泌尿系结石经逆行输尿管软镜碎石术(RIRS)后清石率(SFR)的价值。方法 回顾性纳入216例接受RIRS的泌尿系结石患者并将其分为残余组(n=73)及无残余组(n=143)。以单因素及多因素logistic回归分析临床资料及结石CT表现,筛选RIRS后SFR独立预测因素。分别利用窗宽窗位归一化联合最大最小归一化(记为方法 a)、最大最小归一化(记为方法 b)、窗宽窗位归一化(记为方法 c)及无归一化(记为方法 d)对RIRS前腹部CT进行预处理,基于结石最佳影像组学特征建立ML模型[包括支持向量机(SVM)、逻辑回归(LR)和随机梯度下降(SGD)模型]并筛选其中最佳者;行RUSS及改良S. T. O. N. E评分预测RIRS后泌尿系结石SFR;联合独立预测因素及最佳ML模型构建联合模型。评估各模型及评分系统的预测效能。结果 结石数量、最大结石CT值及体积均为RIRS后SFR的独立预测因素(P均<0.05)。以方法 b预处理后图像构建SVM模型的曲线下面积(AUC)最高(0.861),高于RUSS及改良S. T. O. N. E总评分(AUC=0.750、0.759,P均<0.05)而与联合模型的AUC差异无统计学意义(AUC=0.853,P=0.775)。结论 基于最大最小归一化法预处理CT图像构建的影像组学SVM模型可有效预测泌尿系结石经RIRS后SFR。Objective To observe the value of CT radiomics machine learning(ML)model for predicting stone free rate(SFR)of urinary calculi after retrograde intrarenal surgery(RIRS).Methods Totally 216 patients with urinary calculi who underwent RIRS were retrospectively enrolled and divided into residual group(n=73)and non-residual group(n=143).Univariate and multivariate logistic regression(LR)were performed to analyze clinical data and CT manifestations of stones to screen independent predictors of SFR after RIRS.Window width and window level normalization combined with max-min normalization(denoted as method a),max-min normalization(denoted as method b),window width and window level normalization(denoted as method c)and non-normalization(denoted as method d)of pre-RIRS abdominal CT were performed,respectively,and the best radiomics features of stones were extracted and screened to establish ML models,including support vector machine(SVM),LR and stochastic gradient descent(SGD)models,and the best ML model was screened.RUSS and modified S.T.O.N.E scores were evaluated based on pre-RIRS CT for predicting SFR of urinary calculi after RIRS.A combined model was then constructed with the independent predictors and the best ML model.The predictive efficacy of each model and scoring system were assessed.Results The number of stones,CT value and volume of the maximum stone were all independent predictors of SFR after RIRS(all P<0.05).The area under the curve(AUC)of SVM model constructed with images preprocessed by method b was the highest(0.861),higher than that of the total scores of RUSS and modified S.T.O.N.E(AUC=0.750,0.759,both P<0.05)but not different from that of combined model(AUC=0.853,P=0.775).Conclusion Radiomics SVM model based on max-min normalization preprocessed CT could effectively predict SFR of urinary calculi after RIRS.

关 键 词：尿石症 体层摄影术 X线计算机 机器学习 影像组学 逆行输尿管软镜碎石术 

分 类 号：R691.4[医药卫生—泌尿科学] R814.42[医药卫生—外科学]