摘要
乳腺病理图像的分类在临床医学领域具有重要应用价值。针对分类过程中人为提取图像特征专业知识要求高、耗时多、准确率低的问题,提出一种基于深度卷积神经网络的病理图像分类方法。本方法结合深度卷积神经网络的特征提取能力,在已有的Inception-V3网络模型上进行改进,实现乳腺病理图像的自动分类,并利用数据增强和迁移学习方法改善因数据集较小导致的模型过拟合问题。对乳腺病理图像进行二分类和多分类实验验证算法效果,二分类实验将病理图像分为良性和恶性两种,二分类识别精度达到97%;多分类实验将病理图像分为8种不同类型的乳腺肿瘤,分类识别精度达到89%,具有较好的泛化性能。
The classification research of breast pathological images is of vital importance in clinical medicine.The method based on artificial feature extraction requires professional knowledge,and it is time consuming.In the proposed algorithm.deep Convolutional Neural Network(CNN)was utilized to extract features.To avoid the over-fitting problem,data augmentation method was adopted to augment the database and fine-tune was applied to train the models,In this paper,two kinds of experiments were conducted.The binary classification classified the pathological images into benign and malignant tumors with accuracy 97%,the research of multi-class classification classified the images into 8 different tumor species with 89%accuracy.The results prove the algorithm’s advantage over accuracy.
作者
詹翔
张婷
林聪
冯玮延
赵杏
ZHAN Xiang;ZHANG Ting;LIN Cong;FENG Weiyan;ZHAO Xing(School of Automation,Beijing Institute of Technology,Beijing 100081,China)
出处
《计算机应用》
CSCD
北大核心
2019年第S02期118-121,共4页
journal of Computer Applications
