Application of Statistical Features and Multilayer Neural Network to Automatic Diagnosis of Arrhythmia by ECG Signals

Abnormal electrical activity of heart can produce a cardiac arrhythmia. The electrocardiogram (ECG) is a non-invasive technique which is used as a diagnostic tool for cardiac diseases. Non-stationarity and irregularity of heartbeat signal imposes many difficulties to clinicians (e.g., in the case of myocardial infarction arrhythmia). Fortunately, signal processing algorithms can expose hidden information within ECG signal contaminated by additive noise components. This paper explores a method of de-noising ECG signal by the discrete wavelet transform (DWT) and further detecting arrhythmia by estimated statistical parameters. Parameters of the de-noised ECG signals were used to form an input data vector determining whether the examined patient suffers from a cardiac arrhythmia or not. Input data were transformed using selected linear methods in order to reduce dimension of the input vector. A neural network was used to detect illness. Compared with the results of recent studies, the proposed method provides more accurate diagnosis based on the examined ECG signal data.