基于坐标匹配和子图切分定位乳腺钼靶图像的感兴趣区域  

Locating the region of interest within molybdenum target images based on coordinates matching and sub-image segmentation

作　　者：章鸣嬛[1] 顾雅佳 肖勤 刘文坚 张璇 陈瑛 ZHANG Minghuan;GU Yajia;XIAO Qin;LIU Wenjian;ZHANG Xuan;CHEN Ying(Research center of Big Data Analyses and Process,Shanghai Sanda University,Shanghai 201209,China;Department of Radiology,Fudan University Shanghai Cancer Center,Shanghai 200032;Humanities and Social Sciences,City University of Macao,Macao 999078,China)

机构地区：[1]上海杉达学院大数据分析与处理研究中心,上海201209 [2]复旦大学附属肿瘤医院放射诊断科,上海200032 [3]澳门城市大学人文社会科学学院,澳门999078

出　　处：《生物医学工程研究》2020年第1期18-22,共5页Journal Of Biomedical Engineering Research

基　　金：国家重点研发计划项目(2016YFC1303003)。

摘　　要：乳腺X线摄影技术是早期发现和诊断乳腺肿瘤的首选方法。提取乳腺钼靶图像的感兴趣区域(region of interest,ROI)并利用人工智能算法对其进行模式识别,可有效提高乳腺肿瘤筛查工作的效率。试验图像均来自DDSM乳腺X线钼靶图像公开数据库,以其中BI-RADS分类为第4类(BI-RADS4)的簇状分布多形性钙化钼靶图像为研究对象,探求在设计乳腺钼靶图像分类器过程中提取ROI的新方法。结果显示,设计出优化的分类器后,可高效地识别试验对象,其测试集上的分类准确率最高可达99.3%。因此,本研究可为医生的临床研判提供辅助信息,并为细分BI-RADS4、进一步精准诊断奠定技术基础。Mammography has been the preferred method for early detection and diagnosis of breast cancer.Extraction of the region of interest(ROI)of mammary molybdenum target image and then pattern recognition with artificial intelligence algorithm have been proved to ameliorate significantly the efficiency of breast cancer screening.The experimental images of this study are from an open database-digital database for screening mammography(DDSM).Taking the Clustered Distribution Pleomorphic Calcification(BI-RADS4)mammography as the research object,we explored a new method of extracting ROI designed for mammographic classifier.The results showed that the classifier optimized with the new method could identify the test object in a more efficient way,and the classification accuracy could be as high as 99.3%.This method can provide additional and useful information for clinical diagnosis,lay a technical foundation for subdividing BI-RADS4 and furthering accurate diagnosis.

关 键 词：乳腺X线钼靶图像 DDSM 感兴趣区域 坐标匹配 模式识别 

分 类 号：R318[医药卫生—生物医学工程] Q334[医药卫生—基础医学]