Date of Award

Fall 1997

Degree Type

Thesis - Restricted

Degree Name

Master of Science (MS)


Biomedical Engineering

First Advisor

Ropella, Kristina M.

Second Advisor

Jeutter, Dean C.

Third Advisor

Prieto, Thomas E.


Automated detection of atrial arrhythmias from surface electrocardiograms often relies on contextual techniques. These contextual techniques require explicit detection of atrial and ventricular depolarization and repolarization events, and the relative timing of these events. These computerized interpretation schemes exhibit poor recognition of atrial fibrillation which results in false alarms and necessitates the presence of a cardiologist to verify each diagnosis. Atrial fibrillation constitutes one of the largest classes of arrhythmias seen in the standard 12 lead electrocardiogram. Hence, timely and accurate diagnosis of atrial fibrillation is an important requirement for any computerized arrhythmia detection system. Magnitude-squared coherence (MSC), as applied to intracardiac data, has proven to be a reliable discriminator of fibrillatory and non-fibrillatory cardiac rhythms. In previous studies on surface ECG, techniques using percent power, RR variability and regularity index measures, have been compared to MSC in their ability to differentiate between fibrillatory and non-fibrillatory atrial rhythms. In this research, MSC is compared to contextual analysis techniques in the ability to discriminate between fibrillatory and non-fibrillatory atrial rhythms recorded from the surface ECG. Surface leads II and VI have been used in this study since they are electrically "orthogonal" and typically provide adequate registration of atrial excitation. MSC is a frequency domain, bivariate measure which examines the phase relationship between two simultaneously recorded signals. During fibrillation, the isoelectric regions of ECG recordings are highly irregular in morphology and timing due to the lack of discrete P waves or an organized excitation wavefront. We have shown that the multiple wavelet nature of atrial fibrillation causes MSC between two surface ECG leads to be low. During sinus rhythm or atrial flutter, the ECG recordings in both the leads are more organized due to the presence of a single, organized propagating wavefront. Consequently, MSC values between two orthogonal surface leads tend to be moderate to high for these rhythms. Contextual analysis schemes rely on the specific detection of atrial and ventricular activity, and their relative timing, in the surface ECG. Atrial depolarization waves, such as P waves, are examined in context to ventricular depolarization complexes, such as the QRS complexes. Measurements of P wave duration, PR intervals and PP intervals are made and decisions about the rhythm type are based on these morphological characteristics and their relative timing with respect to other waveforms. Some forms of atrial arrhythmia analysis require the identification and removal of ventricular activity to generate an atrial remainder ECG. In previous studies, atrial remainder ECGs were obtained by creating templates of ventricular activity with manually selected limits for window size. The QRS-T templates were then subtracted out from the original ECGs. In this study, atrial remainder generation was fully automated by setting empirical values for QRS-T template size, making analysis quicker and repeatable. MSC spectra were obtained between the atrial remainders of lead II and Vl in the training and test data sets. Mean MSC in two different frequency bands were calculated using two different segmentation schemes for MSC estimation. In addition, RR variability indices were determined, and I contextual analysis techniques were applied to both the training and test data sets...



Restricted Access Item

Having trouble?