Atrial fibrillation (AF) is one of the most common heart diseases. Billions of people have suffered from it in the world. Although it can lead to terrible complications such as stroke and heart failure, the underlying mechanisms of it are still under-explored. Besides, there is no so-called optimal therapy for the patients. As the disease is progressive, it is important to detect it in an early stage. To develop methods for understanding and detecting AF, the interpretable parametric model can be an option. This model can provide physiological information at the signal level. In this case, the electrocardiogram, as the most commonly used invasive measurement of cardiac conditions, can be the data to model the heart structure and cardiac activities.
This thesis proposes an interpretable parametric model based on P-waves extracted from the ECG signals. Specifically, the autoregressive (AR) model is implemented, which is also known as linear predicting coding (LPC). The goal is to model the atrium and understand the function of the atrium, which can reflect on the varying parameters in the SR and AF cases. In this context, The formant of P-waves is modeled and estimated, which is a representation of the atrium activities. In addition, the parameters of the model are mapped into 2-dimension by the zero-pole plots in order to interpret the differences between SR and AF situations. Based on the differences between parameters and formants, a parametric classifier of high interpretability is developed to detect AF. An alternating searching algorithm is proposed to determine the parameters of the classifier.