摘要
乳腺癌病理图像的准确识别需要具有相关专业病理与医学影像知识的医师,诊断过程耗费大量人力与医疗资源。利用深度学习技术,选用DenseNet为基础分类网络,融合协调注意力机制构造出CA-DenseNet模型,提升网络特征提取能力。网络训练实验结果表明,CA-DenseNet比其他模型具有更强的数据学习能力,基础数据集验证准确率为90.23%,BreakHis数据集10倍数据增强后,网络过拟合程度减轻,采用Focal Loss增强难分样本的损失贡献,CA-DenseNet网络二分类、八分类验证准确率高达94.48%,该网络较同等条件下的DenseNet提高验证准确率0.38%。
Accurate identification of pathological images of breast cancer requires physicians with relevant professional knowledge of pathology and medical imaging,and the diagnosis process consumes a lot of manpower and medical resources.This paper uses deep learning technology,selects DenseNet as the basic classification network,and integrates coordinated attention mechanism to construct CA-Densenet model to improve the feature extraction ability of network.The results of network training experiment show that CA-densenet has stronger data learning ability than other models,and the validation accuracy of basic data set is 90.23%.After the 10-fold data enhancement of BreakHis dataset,the overfitting degree of network is reduced.Focal Loss is used to enhance the loss contribution of difficult samples.Two and eight classification verification accuracy of CA-densenet network is 94.48%,which is 0.38%higher than DenseNet enhanced verification accuracy under the same conditions.
作者
喻殿智
张欣
迟杏
Yu Dianzhi;Zhang Xin;Chi Xing(College of Big Data and Information Engineering,Guizhou University,Guiyang 550000,China;Guizhou Equipment Manufacturing Vocational College,Guiyang 550000,China)
出处
《国外电子测量技术》
北大核心
2022年第5期137-143,共7页
Foreign Electronic Measurement Technology
基金
国家自然科学基金(61865002,62065002)
贵州大学“双一流”研究重大项目(GDSYL2018)资助
关键词
注意力机制
乳腺癌
病理识别
深度学习
attention mechanism
breast cancer
pathological identification
deep learning
