基于FFDM的影像组学在预测三阴性乳腺癌中的价值

The Value of FFDM-Based Radiomics in Predicting Triple-Negative Breast Cancer

目的探究基于全屏数字化乳腺X线摄影(FFDM)的影像组学特征在预测三阴性乳腺癌(TNBC)中的价值。方法回顾性分析200例经病理证实为乳腺癌患者的FFDM图像及临床资料,根据病理结果分为TNBC 50例,非三阴性乳腺癌(NTNBC)150例。用3D Slicer勾画感兴趣区并提取影像组学特征,利用LASSO算法进行特征筛选;采用秩和检验、单因素分析筛选临床特征及X线特征,用Logistic回归构建模型,模型的预测效能通过曲线下面积(AUC)来评估。结果经LASSO算法筛选得到4个影像组学特征后计算影像组学评分,TNBC与NTNBC组间初潮年龄、分娩次数、肿块伴随征象、肿块边缘的差异有统计学意义(P<0.05),分别建立临床+FFDM模型、影像组学模型、临床+FFDM+影像组学联合模型,三种模型在训练集中的AUC值分别为0.788、0.813、0.836,在测试集中的AUC值分别为0.773、0.804、0.815。结论基于FFDM的影像组学特征结合临床与X线征象所构建的联合模型能够在一定程度上鉴别TNBC与NTNBC,可作为一种非侵入性的预测方式来支持临床决策。

Objective The purpose of this paper was to investigate the value of FFDM-based radiomics features in predicting triple-negative breast cancer.Methods The FFDM images and clinical data of 200 patients with pathologically confirmed breast cancer were retrospectively analyzed.50 cases of triple-negative breast cancer(TNBC)and 150 cases of non-triple-negative breast cancer(NTNBC)were classified according to the pathological findings.Outlined the region of interest and extract radiomics features with 3 D Slicer,and used LASSO algorithm for feature screening;Clinical characteristics and X-ray features were screened by rank sum test,univariate,Logistic regression was used to construct the model,and the predictive efficacy of the model was evaluated by AUC.Results The radiomics score was calculated after obtaining four radiomics features by LASSO algorithm.The differences in age of menarche,number of deliveries,accompanying signs of the mass,mass margins between TNBC and NTNBC groups were statistically significant(P<0.05),establish clinical+FFDM,radiomics,and combined clinical+FFDM+radiomics models,respectively,with AUC values of the three models were 0.788,0.813 and 0.836 in the training set,and 0.773,0.804 and 0.815 in the test set,respectively.Conclusion The combined model based on the radiomics features of FFDM combined with clinical and X-ray signs can discriminate TNBC from NTNBC to a certain extent and can be used as a non-invasive predictive modality to support clinical decision making.