摘要
多数现存的心电信号(ECG)分割方法是针对一个心电周期内重要的特征波段而言的,这样的分割方法不能全面反映疾病的综合特征和全貌,特征提取和分类因此受到了影响。为此,提出基于多心电周期融合特征提取研究。文中用不同的ECG分割方法和样本定义得到5个以ARMA系数为特征的向量集,对MIT-BIH数据库中的正常窦性心律(NSR)和心室早期收缩(PVC)分别进行基于Fisher准则和二次判别函数的分类测试。结果表明,基于多心电周期的特征提取能明显地改进分类效果。
Most of existing electrocardiogram (ECG) segmentation methods are based on certain important components that only account for local information in a cardiac cycle. Such segmentation methods are unable to reflect the morphological information, so the feature extraction and classification will be affected and limited. The study of ECG features extracted from multiple cardiac cycles was performed in the research. Five different feature sets were generated using the different ECG segmentation methods and sample definitions, which ARMA coefficients were used as features. The proposed technique was applied to the premature ventricular contraction (PVC) and normal sinus rhythm (NSR) obtained from MIT-BIH database. Two different classifiers were employed in current research, namely Fisher criterion and quadratic discrimination function (QDF) based classifiers. The experimental results show the features extracted from multiple cardiac cycles classify better than that of single cardiac cycle.
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2006年第6期645-649,共5页
Chinese Journal of Biomedical Engineering
基金
浙江省自然科学基金资助项目(Y104284)
浙江省教育厅科研计划项目(0050606)。
关键词
ECG分割
多心电周期
特征提取
分类
ECG segmentation
multiple cardiac cycles
feature extraction
classification