基于阴囊超声参数及机器学习的中重度生精功能障碍风险评分系统的构建与验证  

Construction and validation of a risk scoring system for moderate to severe spermatogenic dysfunction based on scrotal ultrasound parameters and machine learning

作　　者：吴萌 李学峰[2] 李鑫焱 刘菲菲[1] 杨洁[1] 崔广和[1] Wu Meng;Li Xuefeng;Li Xinyan;Liu Feifei;Yang Jie;Cui Guanghe(Department of Ultrasound Medicine,Affiliated Hospital of Binzhou Medical University,Binzhou 256603,China;Department of Reproductive Medicine,Affiliated Hospital of Binzhou Medical University,Binzhou 256603,China)

机构地区：[1]滨州医学院附属医院超声医学科,滨州256603 [2]滨州医学院附属医院生殖医学科,滨州256603

出　　处：《中华生殖与避孕杂志》2024年第7期723-727,共5页Chinese Journal of Reproduction and Contraception

摘　　要：目的建立基于阴囊超声参数及机器学习的中重度生精功能障碍风险评分系统并探讨其价值。方法回顾性队列分析2021年6月至2022年12月期间在滨州医学院附属医院生殖医学科确诊无精子症、中重度少、弱精子症患者112例及同期在本院因不孕而就诊的116例生精功能正常男性的阴囊超声参数。分别使用随机森林、支持向量机、逻辑回归、K-最近邻算法、XGBoost构建模型。综合各模型各参数的平均沙普利可加性模型解释方法值构建风险评分系统。通过受试者工作特征曲线评估模型的预测效能,临床决策曲线评估模型的临床应用价值。结果评分系统包括双侧睾丸总体积、睾丸回声是否均匀,右侧精索静脉内径及精索静脉曲张血液反流时间。风险评分系统训练集的曲线下面积(area under the curve,AUC)为0.757,测试集AUC为0.718,决策曲线显示该评分系统具有较高的临床价值。结论基于阴囊超声参数及机器学习建立风险评分系统可有效预测中重度生精功能障碍,对此类患者的早发现有着积极意义。ObjectiveTo establish a risk scoring system for moderate to severe spermatogenic dysfunction based on scrotal ultrasound parameters and machine learning,and to explore its value.MethodsA retrospective cohort analysis was conducted on 112 patients diagnosed with azoospermia,moderate to severe oligospermia,and asthenospermia in the Department of Reproductive Medicine at Binzhou Medical University Affiliated Hospital from June 2021 to December 2022.Scrotal ultrasound parameters of these patients were compared with those of 116 normal male patients who visited the same hospital during the same period for reproductive assistance.Models were constructed using Random Forest,Support Vector Machine,logistic Regression,K-nearest neighbor algorithm,and XGBoost.A risk scoring system was established based on the average SHapley Additive exPlanation values of each model.The predictive performance of the model was evaluated using the receiver operating characteristic curve,and the clinical application value of the model was evaluated using the decision curve analysis.ResultsThe scoring system included bilateral testicular total volume,whether the testicular echo was uniform,the inner diameter of the right spermatic vein,and the reflux time of varicocele.The area under the curve(AUC)of the risk scoring system for the training set was 0.757,and the AUC for the test set was 0.718.The decision curve showed that this scoring system had a high clinical value.ConclusionA risk scoring system based on scrotal ultrasound parameters and machine learning can effectively predict moderate to severe spermatogenic dysfunction,which is of positive significance for early detection of such patients.

关 键 词：生精功能障碍 机器学习 SHAP 阴囊超声 

分 类 号：R698.2[医药卫生—泌尿科学]