摘要
传统乳腺癌图像分类方法需要从医学图像中人工提取特征,不仅需要具备专业医学知识,而且存在耗时费力、提取高质量特征困难等问题.因此,提出了一种基于特征融合的卷积神经网络乳腺癌图像分类方法.首先预训练了两个不同结构的卷积神经网络,然后利用卷积神经网络自动提取特征的特性,将两个结构提取到的特征进行融合,最后利用分类器对融合的特征进行分类;同时,为避免卷积神经网络模型受小样本量限制出现过拟合现象,通过乳腺病变区域提取、区域细化和数据增强等方法对图像进行适当预处理,并通过过采样方法解决了正负样本不平衡的问题.实验结果显示,该方法在乳腺癌图像数据集BCDR-F03上分类AUC达到89%,对乳腺癌图像的分类精度较传统方法有明显提高.
Traditional breast cancer image classification methods need to extract features from medical images,which not only requires professional medical knowledge,but also is time-consuming,laborious and difficult to extract highquality features.Therefore,a convolution neural network based on feature fusion is proposed for breast cancer image classification.The first pre-training of two different structure of the convolutional neural network,and then use the convolution neural network characteristics of automatic feature extraction,feature extraction to the two structural integration,finally using classifier to classify the feature of fusion;at the same time,in order to avoid the convolutional neural network model by the small sample size limit over fitting this phenomenon,through regional extraction,refinement and enhancement of breast lesions of regional data and other methods of proper image pretreatment,and the sampling method to solve the problem of the imbalance of positive and negative samples.The experimental results show that this method can classify AUC into 89%on the data set BCDR-F03,and the classification accuracy of breast cancer images is significantly improved compared with the traditional methods.
作者
董永峰
刘霞
王利琴
石陆魁
DONG Yongfeng;LIU Xia;WANG Liqin;SHI Lukui(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China)
出处
《河北工业大学学报》
CAS
2018年第6期70-74,共5页
Journal of Hebei University of Technology
基金
河北省科技计划项目(15210345)
天津市科技计划项目(14ZCDGSF00124)
关键词
卷积神经网络
特征融合
支持向量机
数据增强
过采样
乳腺癌病理图像
convolutional neural network
feature fusion
support vector machines
data augmentation
oversampling
breast cancer histological image
