基于超声新技术联合聚类分析对肝纤维化患者肝心一体化预测方法的研究  

Study of new ultrasound technology combined with cluster analysis on prediction method of liver-heart integration in patients with liver fibrosis

作　　者：张巍[1] 李钦策 周康 卢天祺 蒋健 杨秀华[1] Zhang Wei;Li Qince;Zhou Kang;Lu Tianqi;Jiang Jian;Yang Xiuhua(Department of Abdominal Ultrasound,the First Affiliated Hospital of Harbin Medical University,Harbin 150001,China;School of Computer Science and Technology,Department of Computing,Harbin Institute of Technology,Harbin 150006,China)

机构地区：[1]哈尔滨医科大学附属第一医院腹部超声科,哈尔滨150001 [2]哈尔滨工业大学计算学部计算机科学与技术学院,哈尔滨150006

出　　处：《中华超声影像学杂志》2024年第6期482-488,共7页Chinese Journal of Ultrasonography

基　　金：黑龙江省省属高等学校基本科研业务费科研项目(2022-KYYWF-0272)。

摘　　要：目的基于超声弹性及血流向量成像(VFM)技术联合无监督聚类分析和有监督机器学习方法,实现肝心一体化的早期评估。方法采用观察性研究设计,在无任何干预的条件下,选取2021年12月至2022年9月哈尔滨医科大学附属第一医院收治的肝硬化患者45例,肝纤维化患者43例,健康志愿者42例。应用肝脏联合弹性技术及VFM技术分别获取受检者肝脏和心脏信息。对获取的数据进行标准化处理,借助拓扑数据分析(TDA)技术对处理后的数据进行聚类,再基于统计学分析评估聚类结果,最后通过机器学习方法实现有监督的多分类任务。结果依据受检者相似性网络分为5类,各类维度的平均表现为:与其他类别相比,类别1的整体肝心状况最差,类别2、3次之,而类别4则均为肝心正常的健康对照组;类别5较为特殊,此类受检者的肝脏状况不佳,但根据心脏舒张功能评估指南,其心脏表现并无明显的异常,只有少数指标显示出较大的偏差。应用支持向量机(SVM)、随机森林(RFT)、多层感知机(MLP)在测试集多分类任务中准确性平均值分别为70%、81%、84%。结论通过结合肝脏联合弹性、心脏VFM技术及TDA技术构建的患者相似性网络,成功识别出常规心脏指标未显现异常,而可能存在潜在心功能异常的肝纤维化患者,对于指导临床干预措施的选择和优化患者管理分层具有重要意义。Objective To evaluate the early assessment of hepatocardiac integration based on ultrasonic elasticity and blood flow vector imaging(VFM)technology,in conjunction with unsupervised cluster analysis and supervised machine learning methods.Methods An observational research design without any intervention was adopted from December 2021 to September 2022,45 patients with liver cirrhosis,43 patients with liver fibrosis,and 42 healthy volunteers were selected from the First Affiliated Hospital of Harbin Medical University.Liver combined elasticity technology and VFM technology were used to obtain information on the liver and heart of the subjects,respectively.The acquired data were standardized,and then clustered using topological data analysis(TDA)technology on the processed data.Subsequently,the clustering results were evaluated based on statistical analysis,and finally,supervised multi-classification tasks were realized through machine learning methods.Results Patients were stratified into five distinct groups based on a network of patient similarities.The average characteristics of each group were as follows:Group 1 exhibited the most severe hepatocardiac conditions relative to the other groups.Groups 2 and 3 displayed moderately severe conditions.In contrast,Group 4 comprised entirely of healthy controls,all of whom presented with normal hepatocardiac function.Group 5 presented a unique case among the categories.Participants in this group showed poor liver conditions.However,according to the guidelines for cardiac diastolic function assessment,their heart function was generally unremarkable,with only a minority of indicators deviating significantly.Support Vector Machine(SVM),Random Forest Tree(RFT),and Multilayer Perceptron(MLP)were employed for multi-classification tasks on the test dataset.The average accuracies achieved by these models were 70%,81%,and 84%,respectively.Conclusions By combining liver combined ultrasonic elasticity,cardiac VFM technology and TDA technology to construct a patient similarity netwo

关 键 词：血流向量成像 联合弹性成像 肝纤维化 拓扑数据分析 机器学习 

分 类 号：R445.1[医药卫生—影像医学与核医学] R575.2[医药卫生—诊断学]