| Abstract [eng] |
In clinical practice the most common arrhythmia is atrial fibrillation. The purpose – to research the methods to reduce false atrial fibrillation alarms caused by physiological noise and motion artifacts. The method to reduce false atrial fibrillation alarms is introduced. The proposed method is based on the quality parameters of signals estimated from the electrocardiogram signals. The study of quality parameters of signals was made using simulated and clinical signals, estimated whether noise level assessing parameters could distinguish noise episodes from atrial fibrillation and sinus rhythm. The classificator, based on artificial neural network is implemented to assess the quality of signals. Artificial neural network is trained with signals registered on patients during myocardial infarction, with signals from both Physionet and MIT-BIH databases, dividing these signals into three groups: atrial fibrillation, sinus rhythm and noise signals. A trained neural network was tested with long-term (up to 48 hrs.) electrocardiogram signals and it was clarified, that this method reduces 33,62 % of false atrial fibrillation alarms, but 10.76% of atrial fibrillation was identified as noise. |
