ECG identification based on neural networks

Abstract

Electrocardiogram (ECG) can be used in clinical diagnosis for cardiac function. Also, because individuals have different ECG traces, therefore, they can be acquired as promising biometric features for human identification. Data for experiment in this paper were chosen from MIT-BIH Arrhythmia Database. Lead I ECG traces of 33 normal individuals were used. QRS complexes were extracted from filtered ECG data as features for identification. After dimension reduction by principal component analysis, Back Propagation Neural Networks was used as classifier. Finally, identification results were determined by voting mechanism. The results showed that, accuracy of classification can reach up to 99.6% using the method proposed in this paper. Besides, this method surpasses other researches in a comprehensive way by considering aspects such as the number of leads, data set, complexity and accuracy.