基于迁移学习的乳腺结构紊乱异常识别
发布时间：2018-08-19 16:49
【摘要】：针对乳腺X图像中结构紊乱识别困难、样本数量较少的问题,提出基于迁移学习的结构紊乱识别方法,把基于Gabor的毛刺模式特征、GLCM特征以及熵特征等新特征运用其中。基于恶性肿块与结构紊乱的相似性,把恶性肿块作为源域中正样本,负样本由结构紊乱检测算法中的伪正样本构成,对正负样本区域提取多种特征,把结构紊乱作为目标域的训练和测试集分别进行特征提取,使用自适应支持向量机(A-SVM)进行分类。实验在乳腺钼靶摄影数字化数据库(DDSM)上进行,实验结果表明,该方法克服了结构紊乱样本数量少的问题,提高了结构紊乱的识别率。
[Abstract]:In view of the difficulty of structural disorder recognition and the small number of samples in mammogram, a new method of structural disorder recognition based on transfer learning is proposed, in which the new features such as Gabor burr pattern feature and entropy feature are used. Based on the similarity between malignant mass and structural disorder, the malignant mass is regarded as a positive sample in the source domain. The negative sample is composed of pseudo positive samples in the structural disorder detection algorithm, and a variety of features are extracted from the positive and negative sample regions. The training and test sets of structure disorder as the target domain are used for feature extraction, and adaptive support vector machine (A-SVM) is used for classification. The experiment was carried out on the digital mammography database (DDSM). The experimental results show that the method overcomes the problem of small number of structural disorder samples and improves the recognition rate of structural disorder.
【作者单位】： 武汉科技大学计算机科学与技术学院;武汉科技大学智能信息处理与实时工业系统湖北省重点实验室;
【基金】：国家自然科学基金项目(61403287、61472293、31201121) 中国博士后科学基金项目(2014M552039) 湖北省自然科学基金项目(2014CFB288)
【分类号】：R737.9;TP391.41
，
本文编号：2192244
[Abstract]:In view of the difficulty of structural disorder recognition and the small number of samples in mammogram, a new method of structural disorder recognition based on transfer learning is proposed, in which the new features such as Gabor burr pattern feature and entropy feature are used. Based on the similarity between malignant mass and structural disorder, the malignant mass is regarded as a positive sample in the source domain. The negative sample is composed of pseudo positive samples in the structural disorder detection algorithm, and a variety of features are extracted from the positive and negative sample regions. The training and test sets of structure disorder as the target domain are used for feature extraction, and adaptive support vector machine (A-SVM) is used for classification. The experiment was carried out on the digital mammography database (DDSM). The experimental results show that the method overcomes the problem of small number of structural disorder samples and improves the recognition rate of structural disorder.
【作者单位】： 武汉科技大学计算机科学与技术学院;武汉科技大学智能信息处理与实时工业系统湖北省重点实验室;
【基金】：国家自然科学基金项目(61403287、61472293、31201121) 中国博士后科学基金项目(2014M552039) 湖北省自然科学基金项目(2014CFB288)
【分类号】：R737.9;TP391.41
，
本文编号：2192244
本文链接：https://www.wllwen.com/yixuelunwen/zlx/2192244.html
