摘要
目的:探讨基于自动乳腺全容积成像(ABVS)冠状面图像影像组学联合声触诊组织定量(VTQ)技术术前预测浸润性乳腺癌淋巴血管侵犯(LVI)的价值。方法:收集经病理及免疫组化证实的浸润性乳腺癌患者138例,LVI阳性43例,LVI阴性95例。基于ABVS冠状面图像提取影像组学特征,采用LASSO回归降维,筛选最优特征,构建影像组学标签评分(Rad-score)。Rad-score联合超声特征采用Logistic单因素和多因素回归分析筛选独立预测因子,基于超声特征构建影像学模型,影像学模型结合Rad-score构建联合模型,并绘制联合模型列线图和校准曲线。Hosmer-Lemeshow检验评价联合模型的拟合优度,受试者操作特征(ROC)曲线下面积(AUC)评判各模型效能,Delong检验比较各模型AUC,2折交叉对模型行交叉验证,临床决策曲线(DCA)评判模型临床适用性。结果:共筛选8个最优特征构建Rad-score,多因素Logistic回归显示肿瘤长径(LD)、剪切波速度(SWV)及Rad-score为独立危险因素。校准曲线显示预测值与观测值一致性较高,C-index=0.828。Hosmer-Lemeshow检验显示模型拟合较好(χ^(2)=11.469,P=0.177),影像学模型、Rad-score及列线图AUC分别为0.735(95%CI:0.652,0.818)、0.768(95%CI:0.691,0.845)、0.828(95%CI:0.756,0.901)。Delong检验表明所有模型中列线图预测效能最高(均P<0.05)。2折交叉验证fold1 AUC为0.807(95%CI:0.693,0.922)、fold2 AUC为0.771(95%CI:0.636,0.906),平均AUC为0.789。DCA显示模型临床有较高应用价值。结论:基于ABVS冠状面图像影像组学联合VTQ术前可有效预测浸润性乳腺癌LVI状态。
Objective:The purpose of this study was to explore the value of preoperative prediction of lymphatic vascular invasion(LVI)in invasive breast cancer based on automated breast volume scanner(ABVS)coronal imaging Radiomics combined with virtual touch tissue quantification(VTQ).Methods:138 patients with invasive breast cancer confirmed by pathology and immunohistochemistry were collected.Among them,43 cases were LVI positive and 95 cases were LVI negative.Based on ABVS coronary image extraction of omics features,LASSO regression dimensionality reduction was used,optimal features were screened,and radiomic tag score(Rad-score)was constructed.The independent predictors were screened by Logistic univariate and multivariate regression analysis.The imaging model was constructed based on the ultrasonic features.The imaging model was combined with Rad-score to construct the joint model,and the joint model nomogram and calibration curves were drawn.The goodness of fit of the combined model was evaluated by Hosmer-Lemeshow test,the efficiency of each model was evaluated by the area under the receiver operating characteristic(ROC)curve(AUC),the AUC of each model was compared by Delong test,the model was cross verified by 2-fold cross validation,and the clinical applicability of the model was evaluated by clinical decision curve(DCA).Results:A total of 8 optimal features were selected to construct Rad-score,and multivariate Logistic regression analysis showed that the long diameter of tumor,shear wave velocity(SWV)and Rad-score were independent risk factors.The calibration curve showed that the predicted value was in high agreement with the observed value,C-index=0.828.The Hosmer-Lemeshow test showed that the model fit well(χ^(2)=11.469,P=0.177),and the imaging model,Rad-score,and line chart AUC were 0.735(95%CI:0.652,0.818),0.768(95%CI:0.691,0.845),and 0.828(95%CI:0.756,0.901),respectively.Delong test showed that nomogram had the highest prediction efficiency among all models(all P<0.05).The 2-fold cross-validation fold1 AUC
作者
范莉芳
黄磊
赵劲松
吴艺敏
徐争元
徐晓燕
傅雨晨
FAN Li-fang;HUANG Lei;ZHAO Jin-song(Department of Medical Imaging,Wannan Medical College,Anhui 241002,China)
出处
《放射学实践》
CSCD
北大核心
2023年第3期342-348,共7页
Radiologic Practice
基金
国家级大学生创新创业训练计划项目(202210368049)
教育部产学合作协同育人项目(H202207,220603731205201,220603731205157)
皖南医学院校级重点研究项目(WK2021Z15)。
关键词
乳腺肿瘤
自动乳腺全容积成像
影像组学
声触诊组织定量
浸润性乳腺癌
Breast neoplasms
Automated breast volume scanner
Radiomics
Virtual touch tissue quantification
Invasive breast cancer
