Arrhythmia modelling via ECG characteristic frequencies and artificial neural network

ECG refers to non-invasive bioelectrical recording of the heart. Under the clinical settings, the ECG is interpreted by cardiologists via conventional inspection techniques. The methods however are exposed to visual error which leads to inaccurate diagnosis of the heart condition. Hence, as an attempt towards an automated diagnostic system, the paper elaborates on arrhythmia modelling based on ECG characteristic frequency features and artificial neural network. Initially, ECG is acquired from the PTB Diagnostic ECG Database for healthy, bundle branch block, cardiomyopathy and dysrhythmia conditions. A total of 264 segments of 5 seconds ECG have been obtained and converted into power spectral density. The characteristic frequencies; identified through the dominant overshoots in the power distribution were extracted. The relationship between characteristic frequency features and arrhythmias has been successfully modelled via the artificial neural network with 100% training, validation and testing accuracies. The model has also fulfilled the requirements of correlation tests.