基于快速可拒识-双层支持向量分类器的微钙化点的检测算法

Micro-calcification detection algorithm based on fast double-layer support vector classifier with reject performance

为克服微钙化点检测中假阳性高的缺点,本文构造了一种带拒识能力的双层支持向量模型分类器用于钙化点检测。对于输入模式,首先利用基于最大间隔超平面的支持向量分类器(SVM)进行分类判决;然后对真实的钙化点样本特征空间求取最小包含球形的边界,得到钙化点样本的球形支持向量域表示(SVDD),对于输入模式即可利用钙化点的支持向量域表示进行拒识或接受处理;最后利用SVM与SVDD两个分类器的结果进行综合判决。无论是第一层的求取最优分类超平面,还是第二层的边界优化训练,都根据各个训练数据的类间最近邻距离进行排序操作,选择合适的训练样本子空间进行SVM和SVDD训练。仿真实验结果表明,本文提出的算法在不影响微钙化点检出率的情况下,可以部分解决钙化点检测中假阳性高的问题。

To solve the problem of false positive in micro-calcification detection, a double-layer support vector classifier model with reject feature is proposed. Firsdy the first layer of support vector machine classifier （SVM） with maximum margin between two classes is used for the classification of the input pattern; then the sphere support vectors of true micro-calcification points, which describe the distribution of the sample, are obtained by searching for the smallest sphere boundary containing the samples of this class, then the input pattern of non-object classes is rejected by the second support vector domain description （SVDD）. In addition, nearest distance of inter-classes is used to select sub- space for reducing the training time in SVM of the first layer and SVDD of second layer. Lastly the results of SVM and SVDD classifiers are integrated to obtain the right results. Simulation experiment results demonstrate that the method partially solves the problem of false positive without decreasing the detection rate of micro-calcification.