Reich Y, Thomas C W, Pao Y H, Liebman J, Rudy Y
Rafael, Armament Development Authority, Haifa, Israel.
IEEE Trans Biomed Eng. 1990 Oct;37(10):945-55. doi: 10.1109/10.102807.
A statistical classification method is suggested for body surface potential maps (BSPM). The initial data reduction utilizes the Fourier expansion and time integration, resulting in physiological-oriented features. Based on Fischer's criterion, optimal discriminant vectors are used to map the features to an optimal subdomain. Experimental criteria determine the dimensionality of the subdomain and the number of features to be mapped into it. Classification is performed in two steps. In the first, a k-nearest neighbor (k-NN) rule is used for every two-category problem, the results of which are fed into a voting rule for final classification. The method is tested with 123 patients divided into four categories: normal (NR), ischemia (IS), myocardial infarction (MI), and left bundle branch block (LB) patients. The success is between 88% (for IS) and 100% (for LB) for QRS segment integration. Departure maps were used to explain the misclassified patterns.
本文提出了一种用于体表电位图（BSPM）的统计分类方法。初始数据约简利用傅里叶展开和时间积分，得到面向生理的特征。基于费舍尔准则，使用最优判别向量将特征映射到最优子域。实验准则确定子域的维度以及要映射到其中的特征数量。分类分两步进行。第一步，对每一个二类问题使用k近邻（k-NN）规则，其结果输入到投票规则中进行最终分类。该方法对123例患者进行了测试，这些患者分为四类：正常（NR）、缺血（IS）、心肌梗死（MI）和左束支传导阻滞（LB）患者。对于QRS段积分，成功率在88%（IS患者）到100%（LB患者）之间。偏差图用于解释误分类模式。
