采用加性核SVM的二尖瓣瓣根识别

目的 超声心动图中图像噪声严重、分辨率低以及成像范围有限等缺点，导致二尖瓣（MA）瓣根的识别非常困难，采用加性核函数的支持向量机（SVM）分类器识别超声心动图中的二尖瓣瓣根位置。方法 心脏二尖瓣瓣根位置对于心室的分割、心脏建模以及多模态配准很重要。本文提出将加性核支撑向量机分类算法并结合一个局部的上下文特征用于二尖瓣瓣根的识别。主要创新点有：1）利用图像中的上下文特征提取二尖瓣瓣根部特征；2）应用最小加性核的SVM分类器快速识别二尖瓣瓣根的候选点；3）对于候选点应用加权模板，计算候选点的加权密度；4）在加权密度场中，采用二分查找算法，自适应确定一个阈值，剔除二尖瓣瓣根的错分点，确定二尖瓣瓣根的位置。结果 本文算法在10个儿科病人的超声四腔心动图上测试，和手动选出的二尖瓣瓣根点相比，平均误差控制在1.52±2.25个像素。结论 采用加性核函数的SVM分类器能够快速、准确地识别二尖瓣瓣根点。

Recognition of mitral annulus hinge point using additive SVM classifier

Objective The main difficulties identifying hinge points are due to the inherent noise and the low resolution of echocardiography. In this paper,a local context feature combined with additive support vector machines(SVM) classifier is proposed to identify the hinge points of mitral annulus(MA).Method The position of the hinge point of MA is important for segmentation, modeling, and multi-modalities registration of mitral valve. The innovation is as follows: 1) Extracting the hinge point of MA by local context feature.2) Applying the SVM classifier to identify the candidates of MA.3) Compute the weighted density field of candidates which represents the blocks of candidates.4) Applying the binary search algorithm on the weighted density field to maintain an adaptive threshold. This threshold is used to exclude the error from the SVM classifier. Result This algorithm is tested on echocardiographic four chamber image sequence of 10 pediatric patients. Compared with the manually selected hinge point of MA,the mean error is in 1.52±2.25 pixels. Conclusion Add-itive SVM classifier can fast and accurately identify the MA hinge point.