基于MR-T2WI纹理分析在预测子宫肌瘤高强度聚焦超声消融术后疗效的价值  

The Value of Texture Analysis Based on MR-T2WI in Predicting the Efficacy of High Intensity Focused Ultrasound Ablation for Uterine Fibroids

作　　者：罗锦文[1] 何月明[2] 张婷[2] 关玉宝[1] 戚莹莹 LUO Jinwen;HE Yueming;ZHANG Ting;GUAN Yubao;QI Yingying(Department of Medical Imaging,the Fifth Affiliated Hospital of Guangzhou Medical University,Guangzhou 510700,China;Department of Gynecology,the Fifth Affiliated Hospital of Guangzhou Medical University,Guangzhou 510700,China)

机构地区：[1]广州医科大学附属第五医院医学影像科,广州510700 [2]广州医科大学附属第五医院妇科,广州510700

出　　处：《中国医学计算机成像杂志》2024年第2期250-258,共9页Chinese Computed Medical Imaging

基　　金：广州市基础研究计划基础与应用基础研究项目(202102080250);广东省医学科学技术研究基金项目(A2024300)。

摘　　要：目的:探讨基于MR-T2WI纹理分析在预测子宫肌瘤高强度聚焦超声(HIFU)消融术后疗效的价值。方法:回顾性分析本院接受HIFU消融术治疗前后的100例子宫肌瘤患者MR图像及临床资料,进行T2WI全瘤定量纹理分析,以子宫肌瘤非灌注体积比率(NPV)达80%及以上为消融显著,分析肌瘤术前全瘤纹理参数与术后消融效果的相关性,使用3D-slicer软件勾画容积感兴趣区(VOI),PyRadiomics提取867个组学特征,采用MaZda软件内置FPM(Fisher系数、交互信息、分类错误概率组合平均相关系数联合)法保留30个最优特征,使用最小绝对选择与收缩算子(LASSO)降维得到最优的纹理特征,并将患者按照7∶3比例随机分为训练集及验证集,采用logistic回归构建预测模型,绘制受试者工作特征(ROC)曲线用于评价预测模型诊断性能,绘制决策曲线评估预测模型的临床应用价值。结果:LASSO最终选择得到9个最优特征建立的logistic回归模型在训练集的ROC曲线下面积(AUC)、特异度、灵敏度分别为0.833、0.926、0.651,验证集中对应的AUC、特异度、灵敏度分别为0.829、0.667、0.944;决策曲线在组学特征训练集、验证集中的净收益分别为14.1和13.5。结论:采用MR放射组学构建的logistic回归模型在术前预测子宫肌瘤HIFU消融术后疗效有较大价值,预测效果好,可为子宫肌瘤HIFU消融术后疗效评估及术前患者选择提供个性化及量化参考依据。bjective:To evaluate the value of MR-T2WI texture analysis in predicting the efficacy of high intensity focused ultrasound(HIFU)ablation for uterine fibroids.Methods:MR images and clinical data of 100 patients with uterine fibroids before and after HIFU ablation were retrospectively analyzed.Quantitative texture analysis of the whole tumor on T2WI was performed.The significant ablation was defined as the non-perfusion volume ratio(NPVR)up to 80%or more.The correlation between the texture parameters of the whole tumor before operation and the ablation effect after operation was analyzed.Volume of interest(VOI)was delineated by 3D-slicer software,.and 867 histological features were extracted by PyRadiomics extracted 867 histological features.The MaZda software built-in FPM(combinition of Fisher coefficient,interactive information,and classification error probability combined average correlation coefficient)method was used to retain 30 optimal features.The optimal texture features were obtained using least absolute selection and shrinkage operator(LASSO)dimension reduction to obtain the optimal texture features,and patients were randomly divided into a training set and a validation set at a ratio of 7∶3.Logistic regression was used to build predictive models.The diagnostic performance of predictive models were evaluated by receiver operating characteristics(ROC)curve analysis,and the clinical application value of predictive models were evaluated by decision curves analysis.Results:The area under the ROC curve(AUC),specificity and sensitivity of the logistic regression model established by LASSO with 9 optimal features were 0.833,0.926 and 0.651,respectively in the training set,and the AUC,specificity and sensitivity of the logistic regression model in the validation set were 0.829,0.667 and 0.944,respectively;the net benefit of the decision curve in the training set and validation set of the group characteristics was 14.1 and 13.5,respectively.Conclusion:The logistic regression model constructed by MR radiology ha

关 键 词：磁共振成像 纹理分析 子宫肌瘤 高强度聚焦超声 

分 类 号：R445.2[医药卫生—影像医学与核医学]