基于反向传播人工神经网络模型预测非肿块型乳腺癌新辅助化疗疗效  

Predicting the efficacy of neoadjuvant chemotherapy of non-mass breast cancer by back propagation artificial neural networks

作　　者：周靖宇[1] 谢婷婷[1] 林冰玲 葛秀云 单慧明[1] 石桥[1] 戚玉龙[1] 成官迅[1] ZHOU Jingyu;XIE Tingting;LIN Bingling;GE Xiuyun;SHAN Huiming;SHI Qiao;QI Yulong;CHENG Guanxun(Department of Radiology,Peking University Shenzhen Hospital,Shenzhen,Guangdong Province 518000,China)

机构地区：[1]北京大学深圳医院医学影像科,广东深圳518000

出　　处：《实用放射学杂志》2021年第8期1286-1289,1297,共5页Journal of Practical Radiology

基　　金：北京大学深圳医院科研项目(LCYJ2017006)。

摘　　要：目的探讨反向传播人工神经网络(BP-ANNs)预测非肿块型乳腺癌新辅助化疗(NAC)疗效的应用。方法纳入非肿块型乳腺癌52例,所有病例完成NAC 8个周期后行根治性切除术,根据Millei-Payne病理评级分为组织学显著反应(MHR)组与非显著反应(NMHR)组。动态对比增强磁共振成像(DCE-MRI)检查在治疗前(E1)、NAC 2个周期后(E2)和NAC 4个周期后(E3)进行,共3次。从DCE-MRI图像中提取纹理特征,在R Project中构建BP-ANNs模型预测非肿块型乳腺癌NAC疗效。结果52例病例中,MHR组22例,NMHR组30例。以下纹理特征存在统计学差异:直方方差(E1)、直方一致性(E1)、惯性矩(E1)、逆差距(E1)、直方一致性(E2)、直方均值(E3)、直方方差(E3)、直方一致性(E3)、熵比值(E2/E1)。采用上述参数构建BP-ANNs模型,在训练组模型预测准确度92.3%,在测试组模型预测准确度69.2%。结论采用治疗前的直方方差、一致性、惯性矩、逆差距,NAC 2个周期后的直方一致性,NAC 4个周期后的直方均值、方差、一致性以及NAC 2个周期与治疗前的熵比值构建BP-ANNs模型可以预测非肿块型乳腺癌NAC疗效。Objective To explore the application of back propagation artificial neural networks(BP-ANNs)to predict the efficacy of neoadjuvant chemotherapy(NAC)for non-mass breast cancer.Methods 52 patients with non-mass breast cancer were enrolled.All of the patients underwent radical resection after 8 cycles of NAC,and were divided into the major histological response(MHR)group and the non-major histological response(NMHR)group according to the Miller-Payne system.Dynamic contrast-enhanced magnetic resonace imaging(DCE-MRI)was performed before treatment,after 2 cycles of NAC and 4 cycles of NAC.Texture features extraction from DCE-MRI and BP-ANNs modeling was built in R Project to predict the efficacy of NAC.Results Among the 52 patients,there were 22 cases of MHR and 30 cases of NMHR.Variance,consistencyfinertia,homogeneity,consistency,mean,variance,consistency and entropy ratio showed significant differences between the two groups.The parameters above were used to build BP-ANNs model.This model achieved an accuracy of 92.3%in the training group and 69.2%in the test group.Conclusion BP-ANNs model built with variance,consistency,inertia f homogeneity before treatment,consistency after 2 cycles of NAC,mean,variance,consistency after 4 cycles of NAC,and entropy ratio of after 2 cycles of NAC to before treatment can predicts the efficacy of NAC for non-mass breast cancer.

关 键 词：乳腺癌 新辅助化疗 纹理特征 反向传播人工神经网络 

分 类 号：R737.9[医药卫生—肿瘤] R818.051[医药卫生—临床医学]