MRI影像组学特征预测乳腺癌分子分型的价值

Value of MRI Imaging Features in Predicting Molecular Typing of Breast Cancer

目的探讨基于脂肪抑制T2WI、DCE-T1WI和二者联合序列的影像组学特征预测乳腺癌分子分型的价值。方法回顾性分析经术后病理证实的122例乳腺癌患者,术前均行常规MRI和动态增强扫描。用ITK-SNAP软件勾画感兴趣区,利用AK软件分别提取脂肪抑制T2WI、DCE-T1WI和二者联合序列三维病灶的影像组学特征。采用卡方检验及方差分析比较不同分子分型乳腺癌患者间年龄、绝经状态、淋巴结转移情况特征的差异;使用单因素方法、相关性分析、多因素逻辑回归及套索算法(LASSO)进行特征筛选并降维,采用Logistic回归算法建立模型。利用受试者工作特征曲线评估模型的预测效能。结果 Luminal A型33例,Luminal B型54例,HER-2过表达型17例,三阴(TN)型18例,不同分子分型乳腺癌患者间年龄、绝经状态、淋巴结转移情况的差异均不具有统计学意义(P>0.05)。预测Luminal A型、Luminal B型、HER-2过表达型、TN型乳腺癌最佳效能模型是基于脂肪抑制T2WI和DCE-T1WI联合序列的影像组学特征建立的模型,曲线下面积分别为0.820(0.742,0.888)、0.808(0.745,0.869)、0.900(0.833,0.954)、0.837(0.758,0.905)。结论基于MRI影像组学特征构建的模型可有效无创预测乳腺癌分子分型。

Objective To explore the value of predicting the molecular classification of breast cancer based on the imaging omics features of fat suppression T2WI,DCE-T1WI and the combined sequence of the two. Methods A retrospective analysis of 122 breast cancer patients confirmed by surgery and pathology was performed.Routine MRI and dynamic enhancement scan were performed before surgery.Use ITK-SNAP software to delineate the region of interest(ROI),and use AK software to extract the three-dimensional imaging features of fat-suppressed T2WI,DCE-T1WI and the combined sequence of the two.Chi-square test and analysis of variance were used to compare the differences in age, menopausal status, and lymph node metastasis characteristics of breast cancer patients with different molecular types;single-factor methods, correlation analysis, multi-factor Logistic regression and least absolute contraction operator(LASSO) were used Regression method is used to screen features and reduce dimensionality, and the model is built using Logistic regression algorithm.The receiver operating characteristic(ROC) curve is used to evaluate the predictive performance of the model. Results There were 33 cases of Luminal A type, 54 cases of Luminal B,17 cases of HER-2 overexpression type, and 18 cases of triple negative(TN) type.There were differences in age, menopausal status, and lymph node metastasis among breast cancer patients of different molecular types.Not statistically significant(P>0.05).The best performance model for predicting Luminal A,Luminal B,HER-2 overexpression, and TN breast cancers is based on the imaging omics characteristics of the fat suppression T2WI and DCE-T1WI combined sequence, and the area under the curve(AUC) They are 0.820(0.742,0.888),0.808(0.745,0.869),0.900(0.833,0.954),0.837(0.758,0.905). Conclusion The model constructed based on the features of MRI imaging can effectively predict the molecular classification of breast cancer noninvasively.