A neural network based investigation of high frequency components of the ECG

New information retrieval method is applied to detect low amplitude high frequency components of electrocardiogram (ECG). The special neural network using similarities to prototype features is suggested. Prognosis error is chosen as similarity measure of a signal to a prototype. This measure is preferable in the case of a poor signal to noise ratio. New technique was successfully applied for classification of ECG recordings of myocardial infarction (MI) patients with the complication of ventricular fibrillation (VF) vs. the MI patients who have not had the VF, a problem where standard methods failed to provide satisfactory separation of pattern classesInformatikos fakultetasVytauto Didžiojo universiteta