摘要
目的:探讨多参数MRI的瘤内及瘤周影像组学预测乳腺影像报告与数据系统(BI-RADS)4类病变良恶性的应用价值。方法:回顾性分析2020年1月-2024年12月行MRI检查且被评估为BI-RADS 4类的144名患者的临床影像资料,采用7:3的随机分配比例,将所有患者划分为训练组(n=100)及验证组(n=44)。在DCE-MRI和DWI图像上逐层绘制病灶的瘤内感兴趣区(ROI),并通过自动向外扩张3 mm的方式获取瘤周ROI。采用pyradiomics软件提取影像组学特征,并逐步筛选出最优的影像组学特征,采用逻辑回归算法分别构建DCE-MRI序列、DWI序列以及DCE-MRI+DWI的瘤内模型、瘤周模型及瘤内联合瘤周模型。采用Logistic回归分析筛选出与病变良恶性鉴别相关的临床独立危险因素,与最佳预测模型的影像组学评分(Rad-score)相结合,构建临床影像组学融合模型诺模图(Nomogram)。采用受试者操作特征(ROC)曲线、校准曲线及决策曲线分析(DCA)评估模型的预测效能。结果:多参数MRI的的瘤内联合瘤周的影像组学模型对BI-RADS 4类病变良恶性展现出较高的预测效能,训练组和验证组的AUC值分别为0.835和0.843。单因素及多因素分析结果显示,毛刺征是良恶性病变鉴别的临床独立危险因素,最终构建的诺模图进一步提升了BI-RADS 4类病变良恶性预测效能,训练组和验证组的AUC达0.903和0.878。结论:多参数MRI瘤内联合瘤周的影像组学可以有效地鉴别乳腺BI-RADS 4类病变的良恶性,为临床诊疗决策提供可靠参考。
Objective:To explore the value of multiparametric magnetic resonance imaging radiomics based on intratumoral and peritumoral radiomics methods in differentiating benign from malignant of BI-RADS 4 lesions.Methods:A retrospective analysis was conducted on the clinical and imaging data of 144 patients with BI-RADS 4 lesions from January 2020 to December 2024.Patients included in this study were subdivided into the training and validation cohort with a ratio of 7:3 randomly.The intratumoral ROI of the lesions were delineated layer by layer on DCE-MRI and DWI images,and the peritumoral ROI was obtained by expanding outward by 3mm;Subsequently,using the pyradiomics software to extract radiomics features and gradually select the most optimal features.The radiomics models of DCE-MRI,DWI,DCE-MRI+DWI based on intratumoral,peritumoral and intratumoral combined with peritumoral regions were then respectively constructed by using the Logistic Regression.Univariate and multivariate logistic regression analysis were performed to select the independent risk factors,the fusion model was constructed by integrating the radiomics signature with the clinical risk factor.The differential diagnostic performance of the radiomics models were analyzed by using the receiver operating characteristic curve,calibration and decision curves.Results:The radiomics model combining intratumoral and peritumoral regions based on DCE-MRI and DWI has the best predictive efficacy for differentiating benign from malignant of BI-RADS 4 breast lesions.The AUC values of the training group and validation group were 0.835 and 0.843,respectively.The spiculation sign was an independent risk factor for the differentiation.Finally,the nomogram further enhanced the predictive efficacy yielding an AUC of 0.903 in the training and 0.878 in the validation cohort.Conclusions:The multiparametric MRI radiomics based on intratumoral and peritumoral model demonstrates strong capability in differentiating benign from malignant BI-RADS 4 lesions,offering reliable imaging guidance for clinical decision-making.
作者
李若铭
孙敏
王雨薇
赵威宁
刘毓昕
李晗僡
LI Ruo-ming;SUN Min;WANG Yu-wei(Department of Magnetic Resonance Imaging,Medical University Affiliated Cangzhou Central Hospital,Hebei 061000,China)
出处
《放射学实践》
北大核心
2026年第3期265-273,共9页
Radiologic Practice
基金
河北省医学科学研究课题计划(20241572)。
关键词
乳腺病变
乳腺影像报告与数据系统
影像组学
瘤周
磁共振成像
Breast lesion
Breast imaging reporting and data system
Radiomics
Peritumor
Magnetic resonance imaging
