基于YOLOv3的乳腺X线图像肿块检测方法  

YOLOV3-BASED BREAST MASS DETECTION METHOD IN MAMMOGRAPHY

作　　者：潘以轩 陈智丽[1] 高皓 张辉[1] 夏兴华[1] Pan Yixuan;Chen Zhili;Gao Hao;Zhang Hui;Xia Xinghua(School of Information and Control Engineering,Shenyang Jianzhu University,Shenyang 110168,Liaoning,China)

机构地区：[1]沈阳建筑大学信息与控制工程学院,辽宁沈阳110168

出　　处：《计算机应用与软件》2024年第7期136-144,共9页Computer Applications and Software

基　　金：国家自然科学基金项目(61602322);辽宁省自然科学基金项目(20180550059);辽宁省教育厅重点攻关项目(lnzd201904)。

摘　　要：乳腺X线摄影术是目前国际上公认的有效的乳腺癌早期筛查手段。提出一种基于YOLOv3网络的乳腺X线图像肿块检测方法。该方法能够在保证精度的同时,以较快的速度一次完成对整幅图像中肿块的检测。应用迁移学习技术,将由数字化乳腺X线图像学习到的肿块病变检测知识迁移到全域数字图像,有效解决了目前全域数字图像数据集缺乏的问题。使用五折交叉验证方法,在DDSM和INbreast数据集上进行实验验证,最终得到的五折间肿块检测平均准确率为81.34%。Mammography is internationally recognized as an effective screening tool for early breast cancer.This paper proposes a mammographic mass detection method based on YOLOv3 network.The method could complete mass detection of the whole image at a faster speed while ensuring accuracy.By applying transfer learning technology,the mass lesion detection knowledge learned from the digitized mammograms were transferred to the full-field digital mammograms,which effectively solved the current lack of full-field digital mammography datasets.The five-fold cross-validation method was used for evaluation based on DDSM and INbreast datasets.Through extensive experiments,the obtained average accuracy of the mass detection over the five folds is 81.34%.

关 键 词：深度学习 YOLOv3 乳腺X线图像 肿块检测 迁移学习 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]