基于空-频域差值与超分辨率反馈网络的乳腺微钙化区域检测技术  

Space-frequency domain difference with super-resolution feedback network for breast microcalcification detection

作　　者：邢素霞[1] 申楠 刘子骄 鞠子涵 何湘萍[2] 潘子妍 王瑜[1] XING Suxia;SHEN Nan;LIU Zijiao;JU Zihan;HE Xiangping;PAN Ziyan;WANG Yu(School of Artificial Intelligence,Beijing Technology and Business University,Beijing 100048,China;Department of Breast,Haidian District Maternal and Child Health Care Hospital,Beijing 100080,China)

机构地区：[1]北京工商大学人工智能学院,北京100048 [2]北京海淀妇幼保健院乳腺科,北京100080

出　　处：《中国医学物理学杂志》2022年第7期840-849,共10页Chinese Journal of Medical Physics

基　　金：国家自然科学基金(61671028);北京市自然科学基金-北京市教育委员会科技计划重点联合项目(KZ202110011015)。

摘　　要：乳腺癌的早期症状在乳腺钼靶图像中主要表现为微钙化点,微钙化区域的真假阳性检测对于乳腺癌早期筛查具有重要意义。首先,对DDSM乳腺数据集中的图像进行预处理,去除噪声及无关组织干扰;其次,基于空-频域差值图像技术实现了疑似微钙化点的分割,取得的敏感性为91.00%,但假阳性率也较高(34.00%),并根据疑似点的质心位置自动截取感兴趣区域;然后,通过超分辨率反馈网络算法进行微钙化区域超分辨率重建;最后,提取感兴趣区域的纹理特征,将GentleAdaBoost算法和单层决策树算法相结合,构建强分类器GAB-DS对区域进行分类,将微钙化区域和正常组织分离开来,GAB-DS分类模型取得了96.25%的准确率、94.38%的敏感性以及98.13%的特异性。实验结果表明,该模型在微钙化区域检测上性能优越,可用于辅助临床乳腺癌检测及诊断,具有一定的临床应用价值。The early symptoms of breast cancer are mainly characterized by microcalcifications in mammogram.The true and false positive detections of microcalcifications are of great significance for the early screening of breast cancer.After preprocessing the images from DDSM breast data for removing noise and irrelevant tissue interference,the suspected microcalcifications are segmented based on the space-frequency domain difference image technique,obtaining a sensitivity of 91.00%and a false positive rate of 34.00%.According to the centroid position of the suspected point,the region of interest is automatically intercepted.Then,super-resolution feedback network algorithm is used to realize the super-resolution reconstruction of the microcalcifications.Finally,the texture features of the region of interest are extracted,and the Gentle AdaBoost algorithm is combined with the single layer decision tree algorithm to construct a strong classifier GAB-DS for classifying the regions,separating the microcalcifications from the normal tissues.The accuracy,sensitivity and specificity achieved by GAB-DS model were 96.25%,94.38%and 98.13%,respectively.The experimental results show that the model has superior performance in detecting microcalcifications,and can be used to assist in clinical detection and diagnosis of breast cancer,with high application value in clinic.

关 键 词：乳腺癌 微钙化区域 差值图像技术 超分辨率重建 GAB-DS 

分 类 号：R318[医药卫生—生物医学工程] TP391.4[医药卫生—基础医学]