Analysis of ECG to predict atrial fibrillation in post-operative cardiac surgical patients

Atrial fibrillation (AFIB) is characterized by irregular rhythm caused by chaotic electrical impulses that start in the upper chambers of the heart (atria). This often results in the lower chambers of the heart (ventricles) to beat rapidly and irregularly. In the United States alone, every year more than 2 million people are diagnosed with AFIB, with a significant population being post-operative cardiac surgical patients. Presence of AFIB in post-operative cardiac patients results in complications such as stroke, ventricular arrhythmias, and the requirement for a permanent post-surgery pacemaker. The effect of AFIB in these cases not only deteriorates the health condition of the patient but also results in increased hospitalization cost and stay. If AFIB can be predicted ahead of time, the physician could initiate medical treatment to minimize the risk of AFIB or even prevent it.; The overall goal of the proposed research is to develop methods based on electrocardiographic (ECG) analysis that could potentially predict AFIB. Clinical observation and previous studies by others suggest that occurrence of Premature Atrial Contractions (PACs), changes in Heart Rate Variability (HRV) and P-wave morphology could be predictors of AFIB. In this research project, we analyzed these ECG features to identify patterns that can predict AFIB. The research project comprised two main phases. In the first phase, we developed novel techniques to detect PACs, measure HRV and P-wave morphology automatically. The developed techniques used signal processing and wavelet analysis methods to filter noise and detect P-waves. Additionally, the PAC detection technique incorporated steps to eliminate Premature Ventricular Contractions (PVCs) and artifacts. Moreover, as a confirmation of detected ECG features a multi-channel comparison was also performed. These new signal processing steps allowed our algorithms to detect ECG features with greater accuracy and specificity when compared with previously developed techniques. Our techniques were able to detect PACs with 98% accuracy and R-waves with 99% accuracy. Moreover, compared with previous methods that were either manual or semi-manual our methods were fully automatic. As the second phase, the developed techniques were used to identify potential patterns in PAC activity, HRV parameters and P-wave morphology indicative of AFIB. The techniques were applied on ECG collected from 58 AFIB and 57 non-AFIB post cardiac surgery patients. Visual inspection of the trends in the above ECG features was conducted to identify any distinct and consistent patterns that could differentiate AFIB and non AFIB patients. Additionally, within the AFIB group, any progressive trends in ECG parameters leading to AFIB were also explored. Quantitative analysis of the ECG parameters was performed in an attempt to differentiate the AFIB and non-AFIB patient groups. Analysis results indicate that average PAC activity is almost six times in AFIB patients when compared with non-AFIB patients. Additionally, average PAC activity increased progressively towards AFIB. High frequency power of HRV was also noticed to be higher in AFIB patients with an increasing trend towards AFIB. Moreover, it was also noticed that P-wave energy ratio was lower for AFIB patients. Unlike previous studies that focused on ECG data just prior to AFIB, our study analyzed ECG features over a long period of time (4-28 hours) prior to AFIB. This would enable development of methods that can predict AFIB well in advance such that effective prophylactic therapy can be applied.