基于乳腺X线影像组学特征的预测模型在鉴别三阴型与非三阴型乳腺癌中的价值

Triple-negative and non-triple-negative breast cancer prediction by mammographic radiomics features

目的探讨基于乳腺x线影像组学特征的预测模型鉴别三阴型乳腺癌（TNBC）与非三阴型乳腺癌（NTNBC）的价值。方法回顾性分析2015年8月至11月天津医科大学肿瘤医院经手术病理证实为乳腺浸润性导管癌，且具有完整的乳腺X线摄影资料的459例患者，纳入TNBC患者34例，并选取同期的102例NTNBC患者。对病变的乳腺头尾位（CC）和内外斜位（MLO）图像进行手动分割并分别提取43个影像组学特征。分别构建CC位、MLO位单视角及CC和MLO位双视角的分类模型，采用分类准确率、ROC下面积（AUC）、敏感度及特异度进行十次十折交叉验证取其平均值作为最终的分类结果，比较3个模型的分类效能。采用t检验（正态分布）或Kruskal—Walls U检验（偏态分布）进行TNBC与NTNBC间比较分析。结果MLO位预测模型的AUC、正确率和特异度均高于CC位，CC和MLO位双视角模型的AUC、正确率、敏感度及特异度分别为0．791、0．798、0．776和0．806，高于2个单视角模型。TNBC与NTNBC患者间灰度跨度（CC位）、逆差距（CC位）、圆度（MLO位）特征参数的差异有统计学意义（P值分别为0．043、0．010和〈0．01），对应的AUC分别为0．626、0．660、0．753；FD12（CC位）、峰度（CC位）、自相关（CC位）、FD11（MLO位）、FD24（MLO位）、峰度（MLO位）、自相关（MLO位）差异均无统计学意义（P均〉0．05）。结论基于X线影像组学特征的分类模型可有效区分TNBC与NTNBC。

Objective To develop and validate a radiomies predictive model based on mammogram for preoperative predicting triplenegative breast cancer （TNBC） or non- triple- negative breast cancer （NTNBC）. Methods We retrospectively analyzed 459 Chinese women who were diagnosed with invasive breast cancer （confirmed by pathology） during August 2015 to November 2015. Our cohort included 34 TNBC and random selected 102 NTNBC cases. Regions of interest （ROIs） were manually selected from craniocaudal and mediolateral oblique mammograms by radiologists through manual lesion segmentation, and 43 radiomics features were evaluated. Craniocaudal （CC） single-view, mediolateral oblique （MLO） singleview and CC and MLO double- view classification model were constructed respectively. Classification performance was evaluated by the area under receiver operating characteristic curve （AUC）, accuracy, sensitivity and specificity. Kruskal-Walls U test and t test were used to compare the radiomics features between TNBC and UTNBC. Results The model that used the combination of both the CC and MLO view images achieved the overall best performance than using either of the two views alone, yielding an AUC of 0.791,accuracy of 0.798, sensitivity of 0.776 and specificity of 0.806 for TNBC comparing with NTNBC. Three features were selected by the model （gray scale span and inverse different moment for CC, roundness for MLO） showed a statistical significance （P〈0.05） and AUC〉0.6 in the subtype classification. Conclusion This research constructed model based on mammograms classification model can effectively distinguish between TNBC and NTNBC. This model has potential value for breast cancer molecular subtype classification and clinical treatment.