基于迭代顺序滤波子空间约束的可拒识-支持向量机微钙化点检测

Micro-calcifications Detection Based on Iterative Rank-order Filter Subspace and SVM with Rejection Feature

为提高钙化点检测速度,克服微钙化点检测中假阳性高的缺点,本文构造了一种迭代顺序滤波子空间约束的可拒识-支持向量机分类器用于钙化点检测.训练时利用迭代顺序滤波检测作为钙化点的粗检测算子,然后在其约束的子空间内收集非钙化点训练样本.对于输入模式,首先利用基于最大软间隔超平面的支持向量分类器(SVC)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的包含球形边界,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.仿真实验结果表明,本文提出的算法在不影响微钙化点的检出率的情况下,大大提高了检测速度,部分解决了假阳性高的问题.

To improve the speed and the problem of high false positive in micro-calcification detection, a novel micro-calcification detection method based on support vector classifier model with rejection feature and Iterative Rankorder Filter Subspace is proposed. In the training step,iterative rank-order filter is used as coarse detector,and the nonmicro-calcification training samples are selected from the constrained subspace. So it overcomes the effect of number difference between the negative and positive samples, and at the same time it reduces the computing load in training stage. In the detection step, coarse detection is applied at first to find suspect micro-calcification region, then Firs the first layer of support vector classifier （SVC） with maximum margin between two classes will be used for classification the input pattern ;and the sphere support vectors of true micro-calcification points describing the distribution of the sample are obtained by searching all the sphere boundaries containing the samples of this class. So the input pattern of no-object classes can be rejected by the second support vector domain description （SVDD）. Lastly the results of SVC and SVDD classifier are integrated to obtain the accurate results. Experimental results demonstrate that the new calcification detection method performs better in achieving lower false positive rate （FPR） and fast speed.