Approximation of ECG T Wave by Using Gaussian Mixtures and Automatic Measurement of QT Interval

Abstract

The purpose of this study is to establish the accurate automated QT measurement system for detecting drug-induces QT interval prolongation. One hundred and thirty five electrocardiograms (ECGs) were recorded from fifty healthy men. We applied Gaussian mixture model to approximate the ECG trace with split and merge expectation maximization (SMEM) algorithm for optimal model parameters. Lead II ECG trace from the beginning of a T wave to the end of subsequent beat's P wave was approximated by six Gaussian functions with SMEM algorithm. The end of T wave from original ECG was registered by an experienced cardiologist and corresponding point on the approximated ECG trace was defined as a reference point. A primary estimated point for the end of T wave was selected by using a threshold that was the averaged height differences between the reference point and the 2 ms preceding point. The averaged interval in absolute value between the estimated point and the reference point was 5.4 ± 4.3 ms. Furthermore, it was diminished to 5.0 ± 3.8 ms by multiple regression model analysis (Dependent variable: the reference point as the end of T wave. Independent variables: the primary estimated point as the end of T wave and the each height associated with the fourth and the sixth Gaussian function). The final estimated QT interval and the reference QT interval were 399.2 ± 27.2 ms and 399.2 ± 28.0 ms, respectively. We conclude that our system is able to measure QT interval correctly.