基于DenseNet201的乳腺癌病理图像的预测研究  

Predictive Study of Breast Cancer Pathological Images Based on DenseNet201

作　　者：韩杨 苗壮 孙悦 郭金兴 陈广新[1] 高铭泽 HAN Yang;MIAO Zhuang;SUN Yue;GUO Jin-xing;CHEN Guang-xin;GAO Ming-ze(Medical Imaging College of Mudanjiang Medical University,Mudanjiang,Heilongjiang 157011,China;Basic Medical College of Mudanjiang Medical University,Mudanjiang,Heilongjiang 157011,China;Hongqi Hospital Affiliated with Mudanjiang Medical University,Mudanjiang,Heilongjiang 157011,China)

机构地区：[1]牡丹江医学院医学影像学院,黑龙江牡丹江157011 [2]牡丹江医学院基础医学院,黑龙江牡丹江157011 [3]牡丹江医学院附属红旗医院,黑龙江牡丹江157011

出　　处：《新一代信息技术》2024年第1期6-11,共6页New Generation of Information Technology

基　　金：黑龙江省属高校科研基本业务费科研项目(No.2023-KYYWF-0940)

摘　　要：本文研究利用DenseNet201网络构建乳腺病理图像分类预测模型。本研究的数据集包括162个乳腺癌标本,并使用DenseNet121进行实验对比。对比结果显示DenseNet201相较于DenseNet121在乳腺癌检测上的表现更为出色,通过20个训练epochs,整体准确性达82%,未患有浸润性导管癌类别精确度90%、召回率84%,F_(1)分数为0.87。患有浸润性导管癌类别精确度66%、召回率77%,F_(1)分数为0.71。相比其他DenseNet网络,DenseNet201在准确率上提高了5%左右。研究表明DenseNet201在处理大规模乳腺癌图像数据集时,具备更强大的特征提取能力,能更好地适应复杂的数据模式和关系,从而提高了乳腺癌检测的准确性和效率。This study takes breast cancer as the object.Through image processing and deep learning technology,DenseNet201 network is used to analyze and classify breast pathological images.The dataset included 162 breast cancer specimens,of which DenseNet121 was used for experimental comparison.Experimental results showed that DenseNet201 performed better in breast cancer detection than DenseNet121.Through 20 training epochs,the overall accuracy reached 82%,the accuracy of category without invasive ductal carcinoma was 90%,the recall rate was 84%,and the F_(1) score was 0.87.With invasive ductal carcinoma category accuracy of 66%,recall rate of 77%,F_(1) score of 0.71.Compared to other DenseNet networks,DenseNet201 improves accuracy by about 5%.When processing large-scale breast cancer image data sets,DenseNet201 has more powerful feature extraction capabilities and can better adapt to complex data patterns and relationships,thus improving the accuracy and efficiency of breast cancer detection.

关 键 词：乳腺癌 DenseNet 病理图像 深度学习 

分 类 号：R195[医药卫生—卫生统计学]