近年来,乳腺癌的发病率逐年增长,严重影响了人们的生活.许多计算机辅助诊断技术被提出用于乳腺摄影图像的自动分析来辅助医生做出诊断.然而因图像差异较微妙、数据库小等原因,许多传统的方法在诊断准确率方面受到限制且缺乏可扩展性.针对以上问题,本文提出了一种基于哈希的大规模图像检索方法来实现乳腺癌的早期辅助诊断.该方法提取待判定图像与已确诊图像的局部特征,并用迭代量化(ITQ)的哈希学习方法将原始特征空间中的特征向量转化为保存了原始特征之间相似性的二进制码,然后比较汉明距离找出与待判定图像最相似的一系列图像,并根据返回图像做出诊断.实验表明该方法可用于大型数据库且具有可扩展性,有效地提高了诊断准确率,可以帮助医生做出正确的诊断.%In recent years,the incidence of breast cancer increased year by year,which brought bad influence to people's life.Many computer aided diagnostic techniques have been successfully developed to automatically analysis breast mammographic image to assist the doctor making a diagnosis.However,due to difference between images is subtle,and reason of small database,most traditional methods is limited in diagnosis accuracy and fall short of scalability.To solve above problems,we proposed a hashing-based large-scale mammographic image retrieval method for the image-guided diagnosis of breast cancer.In this method,local features are extracted from each query image and diagnosed image.Iterative Quantization (ITQ) hashing method is used to compress the mammogram features into compact binary codes which preserved the similarity in original space,and then perform searching in the Hamming space.Finally,a diagnosis result can be obtained according to the returned images.Extensive experiments demonstrate that our system can be used in large-scale database and achieved excellent performance.It is capable of aiding doctors to make reliable clinical decision.
展开▼
