摘要
目的探讨基于MRI的影像组学模型对穿透性胎盘植入(placenta percreta,PP)的鉴别能力。材料与方法回顾性分析2021年1月至2023年12月在郑州大学第三附属医院放射科行胎盘MRI平扫且MRI提示为PP的80例孕产妇的临床及影像资料,其中以术中所见为标准诊断为PP 48例,非PP 32例。所有患者按7∶3的比例随机划分为训练集与测试集。在轴位、冠状位及矢状位的T2WI序列上手动勾画感兴趣区,并提取影像组学特征。对提取的影像组学特征首先进行Z-score正则化操作,再通过t检验进行特征筛选,随后计算Pearson相关系数,最后采用最小绝对收缩和选择算子算法对组学的特征进行筛选及降维,并计算影像组学分数。从7种不同的机器学习算法中选取最优算法进行影像组学模型的构建。对临床信息和影像组学评分分别做单因素逻辑回归分析,将差异有统计学意义的因素纳入到多因素分析,得到独立危险因素(临床信息同时用于构建临床模型),并将其可视化(列线图)建立联合预测模型(影像组学-临床模型)。绘制受试者工作特征曲线,并通过曲线下面积(area under the curve,AUC)、敏感度、特异度和准确率等指标比较模型的效能,运用校准曲线评价模型的校准程度,决策曲线分析评估模型的临床实用价值。结果筛选出的独立危险因素为孕次和影像组学评分,其优势比分别为0.272[95%置信区间(confidence interval,CI):0.151~0.492]和1934.105(95%CI:118.985~31445.149)。影像组学模型与临床模型在训练集中的AUC值分别为0.948(95%CI:0.884~1.000)和0.723(95%CI:0.596~0.850),在测试集中的AUC值分别为0.828(95%CI:0.601~1.000)和0.676(95%CI:0.474~0.878)。在训练集中影像组学-临床模型的AUC值为0.962(95%CI:0.906~1.000)。DeLong检验结果表明在训练集中临床模型与影像组学模型间及临床模型与影像组学-临床模型间差异均具有统计学意义(P<0.05),但影像组学模型与影像组学-临床模型间差异无统计学意义(P>0.05)。影像组学模型与影像组学-临床模型均具有较好的校准度及临床应用价值。结论影像组学-临床模型具有较好的诊断效能,可作为对PP鉴别的方式,有助于临床医师对妊娠终止时机和方式的制订提供可靠的依据。
Objective:To explore the ability of a MRI based imaging histologic model to identify placenta percreta(PP).Materials and Methods:A retrospective study was conducted to collecting data from 80 cases of pregnant women who underwent placental MRI scanning and MRI indications pointing to PP in the Department of Radiology of the Third Affiliated Hospital of Zhengzhou University from January 2021 to December 2023,with surgical findings as the standard,including 48 cases of PP and 32 cases of non-PP.The region of interest was manually outlined on the axial,coronal and sagittal T2WI sequences,and the features of imaging histology were extracted.All patients were randomly divided into training and test sets in the ratio of 7∶3.The extracted imaging histology features were firstly subjected to Z-score regularization,then feature screening by t test,followed by calculation of Pearson correlation coefficients,and finally the least absolute shrinkage and selection operator algorithm was used.Selection operator algorithm for screening and dimensionality reduction of the features of the histology,and calculate the radiomics score.The optimal algorithm was selected from 7 different machine learning algorithms and used to construct an radiomics model.Univariate logistic regression analysis was performed on both clinical data and radiomics scores,revealing statistically significant differences.Subsequently,factors demonstrating significant differences were incorporated into multivariate analysis to identify independent risk factors(clinical information was used to construct clinical models).These factors were then visualized to construct a predictive combined model(nomogram).The receiver operating characteristic curve was plotted,and the efficacy of the model was compared by the indicators of area under the curve(AUC),sensitivity,specificity,and accuracy,and the calibration curve was used to evaluate the calibration degree of the model,and the decision curve analysis was used to assess the effectiveness of the model.The calibration curve was used to evaluate the calibration degree of the model,and the decision curve analysis was used to assess the clinical utility value of the model.Results:The multivariate analysis identified two independent risk factors:parity and radiomics score.Parity demonstrated a protective effect with an odds ratio of 0.272 [95% confidence interval(CI):0.151 to 0.492],while the radiomics score showed a strong positive association with an exceptionally high odds ratio of 1 934.105(95%CI:118.985 to 31 445.149).The AUC values for the imaging histology model and the clinical model in the training set were 0.948(95%CI:0.884 to 1.000) and 0.723(95% CI:0.596 to 0.850),respectively,and in the test set were 0.828(95% CI:0.601 to 1.000) and 0.676(95% CI:0.474 to 0.878).The AUC value of the imaging histology-clinical model in the training set was 0.962(95% CI:0.906 to 1.000).The results of DeLong test showed that there were significant differences in the training set,both between the clinical model and the imaging histology model as well as between the clinical model and the imaging histology-clinical model(P < 0.05),but the differences between the imaging histology model and the imaging histology-clinical model were not statistically significant(P > 0.05).Both the radiomics model and the radiomics-clinical model had good calibration and clinical application value in the test set.Conclusions:Imaging histology-clinical modeling has better diagnostic efficacy and can be used as a modality for the identification of PP.It provides a reliable foundation for clinicians in determining the timing and method of pregnancy termination,thereby aiding in the formulation of informed clinical decisions.
作者
冯刘娟
张灵洁
程美英
张小安
李思柯
鲁钰
刘世鹏
杨金泽
赵鑫
FENG Liujuan;ZHANG Lingjie;CHENG Meiying;ZHANG Xiaoan;LI Sike;LU Yu;LIU Shipeng;YANG Jinze;ZHAO Xin(Department of Medical Imaging,the Third Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China;Department of Ultrasound Medicine,the Third Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China)
出处
《磁共振成像》
北大核心
2025年第3期83-89,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
2020年河南省重大公益专项(编号:201300310800)
郑州市科技局协同创新重大专项(编号:18XTZX12009)。
关键词
胎盘植入性疾病
穿透性胎盘植入
影像组学
磁共振成像
鉴别
placenta accreta spectrum disorders
placenta percreta
radiomics
magnetic resonance imaging
differentiate
