基于卷积神经网络的乳腺癌分子分型预测研究  

Molecular Typing Prediction of Breast Cancer Based on Convolutional Neural Network

作　　者：任湘 张朋 范明 厉力华 REN Xiang;ZHANG Peng;FAN Ming;LI Lihua(School of Life Information Science and Instrument Engineering,Hangzhou Dianzi University,Hangzhou Zhejiang,310018,China)

机构地区：[1]杭州电子科技大学生命信息与仪器工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2018年第5期66-71,共6页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：国家自然科学基金资助项目(61271063);国家重点研发计划课题资助项目(2017YFC0109402);浙江省自然科学重点基金资助项目(LZ15F01001)

摘　　要：探索深度卷积神经网络在乳腺癌分子分型预测中的应用。回顾性分析171例术前、化疗前行免疫组化病理检查及动态增强磁共振DCR-MRI检查的乳腺癌患者。根据病例免疫组化检查结果将乳腺癌病例分为Luminal A,Luminal B,HER-2过表达和Basal-like 4种分子分型。考虑到样本类别数量及其平衡性,对Luminal B型与非Luminal B型(包括其他3种类型)进行研究。首先,根据医生标注的病灶信息从原始DCE-MRI影像中提取包含病灶的目标区域图像。然后,运用深度卷积神经网络对感兴趣区域进行卷积运算,通过训练获得分类模型。最后,对分类模型的预测结果进行分析。结果表明,通过深度卷积神经网络对乳腺癌分子分型预测的受试者的DCE-MRI影像进行分析,其工作特征曲线下面积最高值为0.697,有一定预测效果。In this study,we explored the application of convolution neural network in prediction of molecular subtypes in breast cancer.171 malignant breast cancer patients with immunohistochemical and DCE-MRI examinations before chemotherapy were retrospectively analyzed.The patients were divided into 4 molecular subtypes including Luminal A,Luminal B,HER-2 overexpression and Basallike 4 subtypes according to the immunohistochemistry results.Considering the imbalance of sample number in different subtypes,two molecular subtypes prediction study were conducted：prediction of Luminal B and non-Luminal B（including the other three subtypes）including 93 and 78 cases,respectively.Firstly,we extracted the object region that contains the lesion area from original DCEMRI image based on the lesion information marked by the physician.Then,the convolution neural network was used to establish the molecular subtype prediction model.Finally,the effectiveness of the model was evaluated by（calculating AUC values）.The results show that the deep convolutional neural network has a certain effect on prediction of molecular subtypes in DCE-MRI for breast cancer.In addition,the method provides a new idea for diagnosis of molecular subtypes and has potential clinical values.

关 键 词：乳腺癌 分子分型 卷积神经网络 深度学习 

分 类 号：R318[医药卫生—生物医学工程]