A Principal Component Regression Approach for Estimation of Ventricular Repolarization Characteristics

The time interval between Q-wave onset and T-wave offset, i.e., QT interval, in an ECG corresponds to the total ventricular activity, including both depolarization and repolarization times. It has been suggested that abnormal QT variability could be a marker of cardiac diseases such as ventricular arrhythmias, and QT-interval has also been observed to lengthen during hypoglycemia. In this paper, we propose a robust method for estimating ventricular repolarization characteristics such as QT interval and T-wave amplitude. The method is based on principal component regression. In the method, QT epochs are first extracted from ECG in respect of R-waves. Then, correlation matrix of the extracted epochs is formed and its eigenvectors computed. The most significant eigenvectors are then fitted to the data to obtain noise-free estimates of QT epochs. Nonstationarities in QT-epoch characteristics can also be modeled by updating the eigenvectors dynamically. The main benefit of the proposed method is robustness to noise, i.e., it works also when using ECGs that have low SNR, for example, signals measured during normal-life environments. One application of the proposed method could be the detection of the hypoglycemia.