多级自组织映射用于心电信号QRS波群聚类

提出了一种利用多级自组织映射(MSOM)网络进行心电QRS波群聚类的算法。此方法将归一化的两导联心电数据作为第一层自组织映射网络的输入,其输出作为第二层自组织映射的输入,最后得到聚类结果。网络迭代学习过程采用了特殊的设计,能根据不同类别自适应调整学习参数,从而提高了自组织映射的聚类能力。使用MIT-BIH数据库数据的聚类结果表明,这种方法非常适合心电QRS波群的聚类,对室性早搏(PVC)真阳性检出率达到99.1%,且聚类效率比ART-2网络方法、匹配方法有明显优势。

Multilayer self-organizing maps method for cluster electrocardiogram QRS wave clustering

This paper describies the applications of multilayer self-organizing maps to cluster electrocardiogram QRS waves. Two normalized lead electrocardiogram data were fit to the first layer self-organizing map, with the outputs used as the input to the second layer to achieve clustering results. An intertive adaptive parameter learning iteration process enhanced the self-organizing maps net performance. The clustering capability was evaluated with MIT-BIH data. The results indicate the efficient clustering ability with a 99.1% true positive rate in detecting premature ventricular contraction (PVC) class abnormal electrocardiogram beats, which is better than the ART-2 net and match method.