摘要
提出一种基于相关向量机(RVM,relevance vector machines)的乳腺X线图像结构扭曲(AD)检测方法。首先利用小波变换对感兴趣区域(ROI)进行特征提取;然后通过交叉验证方法确定最优RVM核函数类型及参数;最后利用RVM对测试样本进行AD的识别分类,得到最终的AD检测结果。在Mini-MIAS(mammographic image analysis society)乳腺图像库和北京大学人民医院乳腺中心乳腺图像集上进行验证的实验结果表明,相比常用的基于支持向量机(SVM)的检测方法,本文方法在获得相同检测性能的情况下,极大提高了检测速度,检测时间缩短90%以上;而且对不同结构特性的东西方女性乳腺图像具有更强的适用性,更适合临床应用。
Detection of architectural distortion （AD） in mammograms is one of important approaches in breast cancer diagnosis. Using support vector machine （SVM） to detect AD can achieve high accuracy rate,but it is also with slow speed, making it not suitable for clinical application. To solve the above problems,a method to detect AD in mammograms based on relevance vector machine （RVM） is proposed. Firstly,the discrete wavelet transform is applied to extract features in region of interest （ROI）. Then the cross validation （CV） method is used to determine the optimum type and parameters of RVM kernel function. Lastly, RVM is applied to classify the test samples to obtain the detection results of AD. The proposed method is evaluated on mammograms from the mammographic image analysis society （Mini-MIAS） and those from the breast cancer of Peking University Peoplers Hospital. The results show that compared with SVM method, the proposed method achieves essentially the same sensitivity with much higher speed of detection,which can shorten the detection time of AD more than 90 %. The proposed method is more applicable for mammograms with different characteristics of both oriental and occidental women.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2013年第4期826-832,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(61271305
61201363)
高等学校博士学科点专项科研基金(20110009110001)
中央高校基本科研业务费专项资金(2011JBM003)资助项目
