Evaluating Mathematical Models for Morphological Classification of the QRS Complex

João Paulo do Vale Madeiro1, Délcio Barreto2, João Alexandre Marques3, João Salinet4
1University for the International Integration of the Afro-Brazilian Lusophony, 2Federal University of Ceará, 3University of Saint Joseph, 4CECS, Federal ABC University


The use of the electrocardiogram (ECG) signal is globally accepted as a gold standard in the noninvasive diagnosis of arrhythmias and conduction disorders. Considering that the automatic extraction of ECG parameters is initiated with the segmentation of its characteristic waves, the present work aims to present a comparative study of the performance of different kernels for mathematical modeling and morphological classification of the QRS complex of the ECG signal. Some of identified morphologies may be associated with adverse events. Initially, we use a computer simulator to generate synthetic signals from dynamic models and to analyze variations of a set of parameters (heart rate, sampling frequency, signal duration, Gaussian noise amplitude, mean and standard deviation of heart rate, relationship between LF (Low Frequency) and HF (High Frequency) components, duration, amplitude and morphology of the QRS complex). From the generation of twenty different types of QRS morphology, computational tests for mathematical modeling of the beat waveform (Q, R and S waves) were performed. For that, the following mathematical functions were employed: Gaussian function, Mexican Hat function and Rayleigh probability density function. Subsequently, 10 real signal records were used, with a reference duration of 30 minutes, from the MIT-BIH (Massachucetts Institute Technology - Beth Israel Hospital) Arrhythmia database. Through computing simulations, the proposed functions were tested for a set of morphologies available in the referred database. The preliminary results demonstrate the proposed mathematical functions with adjustable parameters can be applied together for modeling and automatic classification of types of QRS morphologies commonly present in real signals, with efficiency and precision. The computing of normalized RMS error allows the identification of the model which is more appropriate to a given morphology, which can change over the same patient record.