FFDM影像组学在预测乳腺癌腋窝淋巴结转移中的价值

The Value of FFDM-Based Radiomics in Predicting Axillary Lymph Lode Metastasis in Breast Cancer

目的探讨基于全视野数字化乳腺X线摄影(FFDM)的影像组学在预测乳腺癌患者腋窝淋巴结(ALN)转移中的价值。方法回顾性分析407例乳腺癌患者的FFDM图像和临床资料,将患者以7∶3的比例随机分为训练集(n=285)和测试集(n=122)。对FFDM图像的肿瘤病灶进行分割和影像组学特征提取,通过LASSO回归进行特征筛选并建立影像组学模型,计算影像组学评分。通过Logistic回归分析ALN未转移组(n=242)与转移组(n=165)间的临床危险因素并建立临床模型。根据临床危险因素和影像组学评分构建联合预测模型,并绘制列线图。绘制ROC曲线及校准曲线评价模型的效能。结果经筛选得到6个影像组学特征和3个临床特征,分别建立影像组学、临床以及二者的联合模型,3种模型在训练集中的曲线下面积分别为0.740、0.812、0.859,在测试集中的曲线下面积分别为0.749、0.794、0.803。联合模型列线图包含了影像组学评分、肿瘤大小、淋巴结触诊状态和FFDM观察淋巴结状态。结论基于FFDM影像组学特征与临床危险因素构建的列线图,可以作为一种无创性的预测工具,帮助临床医师在术前评估乳腺癌患者的ALN状态。

Objective To investigate the value of FFDM-based radiomics in predicting axillary lymph node(ALN)metastasis in breast cancer.Methods The FFDM images and clinical data of 407 breast cancer patients were retrospectively analyzed,and the patients were randomly divided into training set(n=285)and test set(n=122)in a ratio of 7:3.The tumor lesions were segmented on FFDM and radiomics features were extracted.LASSO regression was used for feature selection to construct the radiomics model,and radiomics score was calculated.The clinical risk factors were analyzed by Logistic regression between the ALN non-metastasis group(n=242)and metastasis group(n=165)to construct the clinical model.The joint predictive model was constructed by combining clinical risk factors and radiomics score,and the nomogram was drawn.The ROC curve and the calibration curve were plotted to evaluate the effectiveness of the model.Results Six radiomics features and three clinical features were screened to establish the radiomics model and the clinical model,and a combined model of both was established.The area under curve(AUC)of the three models were 0.740,0.812 and 0.859 in the training set,and 0.749,0.794 and 0.803 in the test set,respectively.The nomogram includes radiomics score,tumor size,lymph node palpation status and FFDM-reported lymph node status.Conclusion The nomogram based on the radiomics features of FFDM and clinical features can be used as a non-invasive predictive tool to help clinicians assess the ALN status of breast cancer patients before surgery.