基于DCE-MRI影像组学和常规MRI特征的列线图预测乳腺癌淋巴血管侵犯

目的：研究基于动脉对比增强磁共振成像（DCE-MRI）影像组学和常规MRI特征的列线图用于术前预测浸润性乳腺癌（IBC）患者的淋巴血管侵犯（LVI）状态。方法：回顾性分析2016 年7月至2021年5月间，经术后病理证实的300例IBC患者，并按照8:2比例随机分为训练组（n =238）和验证组（n =62）。对所有患者第二期DCE-MRI图像中的病灶区进行手动分割，并提取影像组学特征。采用方差阈值、select k best、LASSO回归进行影像组学特征筛选并计算影像组学评分（rad-score)。使用Logistic回归分析筛选常规MRI特征建立常规特征模型，并结合影像组学和常规MRI特征中的独立危险因素构建联合预测模型，并绘制列线图。使用受试者工作特征（ROC）曲线和校准曲线评估模型的效能，使用决策曲线分析评估模型的临床价值。结果：共提取到1 409个影像组学特征，经筛选得到15个影像组学特征与乳腺癌LVI状态相关，参与计算rad-score值。所有常规MRI特征中，最大直径（OR =1.743，P ＜0.001）和毛刺征（OR =6.304，P ＜0.001）是预测LVI阳性的独立危险因素。在训练组及验证组中，影像组学模型的ROC曲线下面积（AUC）分别为0.831和0.811；常规特征模型的AUC分别为0.779和0.770；联合预测模型的AUC分别可提高到0.889（95%CI =0.844～0.934）和0.856（95%CI =0.759～0.952）。校准曲线表明列线图预测值和实际值吻合较好，决策曲线显示列线图有较高的临床应用价值。结论：基于DCE-MRI影像组学和常规MRI特征构建的列线图用于术前预测IBC患者LVI状态具有良好的应用价值，可为临床治疗提供参考。

Nomogram for prediction of lymphovascular invasion in breast cancerbased on DCE-MRI radiomics andconventional MRI features

Objective: To develop a nomogram based on arterial contrast-enhanced magnetic resonance imaging (DCE-MRI) radiomics and conventional MRI features for preoperative prediction of lymphovascular invasion (LVI) status in patients with invasive breast cancer (IBC). Methods: A retrospective analysis was made on 300 patients with IBC confirmed by postoperative pathology between July 2016 and May 2021, who were randomly divided into the training (n=238) and validation group (n=62) according to ratio of 8:2. Lesion areas in second-stage DCE-MRI images of all patients were manually segmented and radiomics features were extracted. Variance threshold, select k best, and LASSO regression were used to screen radiomics features and calculate the radiomics score (rad-score). Logistic regression analysis was used to screen conventional MRI features to establish conventional feature model, while a joint prediction model was built based on radiomics and independent risk factors in conventional MRI features, and a corresponding nomogram was drawn. The performance of the model was evaluated using receiver operating characteristic (ROC) and calibration curves, and the clinical value of the model was assessed using decision curve analysis. Results: A total of 1409 radiomics features were extracted,and 15 radiomics features were screened to correlate with the LVI status of breast cancer, and were involved in the calculation of rad-score values. Among all conventional MRI features, the largest tumor diameter (OR=1.743,P<0.001) and burr sign (OR=6.304, P<0.001) were independent risk factors for LVI positive. In the training and validation group, the area under the ROC curves (AUCs) of the radiomics model was 0.831 and 0.811,respectively; the AUCs of the conventional feature model was 0.779 and 0.770, respectively; and the AUCs of the joint prediction model was improved to 0.889 (95%CI=0.844-0.934) and 0.856 (95%CI=0.759-0.952). The calibration curve showed that the predicted value of the nomogram was in good agreement with the actual value, and the decision curve showed that the nomogram had a higher clinical value. Conclusion: The constructed nomogram based on DCE-MRI radiomics and conventional MRI features has good application value for preoperative prediction of LVI status in IBC patients, and can provide reference for clinical treatment.