摘要
目的:探讨超声联合机器学习模型在产后压力性尿失禁(SUI)风险预测中的应用价值。方法:纳入362例产后女性(SUI组114例,对照组248例)。收集产科临床因素及产后42d盆底超声参数。采用逻辑回归、随机森林、支持向量机(SVM)和极端梯度提升四种算法构建预测模型,通过区分度、校准度和临床实用性多维评估确定最优模型,并分析主要贡献特征。结果:四种机器学习模型中,SVM算法表现最佳(区分度为0.924,校准度为0.103)。特征重要性排序显示,新生儿出生体重、肛提肌厚度、妊娠期体重增加值、膀胱颈移动度和Valsalva时肛提肌裂孔面积是预测产后SUI风险的关键因素。结论:基于产后42d超声检查数据和临床特征构建的机器学习模型能有效预测产后SUI风险,为临床早期识别高风险人群并实施个体化干预策略提供了新工具。
Objective:To investigate the value of ultrasound combined with machine learning models in predicting postpartum stress urinary incontinence(SUI)risk.Methods:A total of 362 postpartum women were included(114 in the SUI group,248 in the control group).Obstetric clinical factors and pelvic floor ultrasound parameters at 42 days postpartum were collected.Four algorithms—Logistic regression,random forest,support vector machine(SVM),and extreme gradient boosting—were used to construct prediction models.The optimal model was determined through multidimensional evaluation of discrimination,calibration,and clinical utility,followed by analysis of key contributing features.Results:Among the four machine learning models,SVM algorithm demonstrated the best performance(discrimination of 0.924,calibration of 0.103).Feature importance ranking showed that neonatal birth weight,levator ani muscle thickness,gestational weight gain,bladder neck mobility,and levator hiatus area during Valsalva maneuver were the key factors in predicting postpartum SUI risk.Conclusion:Machine learning models based on ultrasound data and clinical characteristics at 42 days postpartum can effectively predict postpartum SUI risk,providing a new tool for early identification of high-risk populations and implementation of individualized intervention strategies.
作者
王彬
多涛
徐秀丽
吴红梅
马娟
马丽萍
俞学卉
WANG Bin;DUO Tao;XU Xiu-li;WU Hong-mei;MA Juan;MA Li-ping;YU Xue-hui(Department of Ultrasound,Yinchuan Maternal and Child Health Hospital,Gansu Yinchuan 750001)
出处
《中国医疗器械信息》
2025年第18期8-11,109,共5页
China Medical Device Information
基金
宁夏回族自治区卫生健康系统科研课题(项目名称:三维/四维超声技术对产后盆底功能障碍康复的价值评估,项目编号:2022-NWKY-069)。
关键词
压力性尿失禁
盆底超声
机器学习
风险预测
支持向量机
stress urinary incontinence
pelvic floor ultrasound
machine learning
risk prediction
support vector machine
