ECG based system for arrhythmia detection and patient identification

In this paper a system to detect arrhythmia by automatically classifying normal and two types of abnormal ECG signals is presented. ECG signals are first pre-processed to reduce the baseline drift, noise and other unwanted components that might be present in the signal. The autoregressive modelling of the signals is then applied to extract small set of signal features - coefficients of autoregressive (AR) signal model. Groups of extracted AR parameters for three different ECG types are well separated in feature space which provides for perfect signal classification and heart condition detection for every ECG signal from the test set. In order to assess the accuracy of developed technique for individual patient identification, feature sets are extended with additional parameter - power of AR modelling error. A new ECG based biometric system is proposed and initial patient recognition results presented in the conclusion of the paper.