Detection of QT prolongation through approximation of the T wave on Gaussian mixture modeling

To establish a simple and accurate method for the automated identification of the end of a T wave, we approximated electrocardiograph (ECG) traces using a Gaussian mixture model in conjunction with a split-and-merge expectation-maximization algorithm.

A total of 286 ECG traces of heart beats of 50 healthy men were used as control data and ECGs from 15 subjects recorded before and after 400mg oral moxifloxacin as positive controls. An experienced cardiologist determined the reference points by visual inspection of the original ECGs. The primary estimated point for the end of the T wave was selected as the point 2 ms before the point at which the gradient of the approximated wave was not steeper than the common threshold value. This point was then adjusted by applying modification rules proposed by an experienced cardiologist. The absolute value of the average interval between the resulting final estimated point and the manually selected reference point was 1.8±7.7 ms for the control data. After treatment with moxifloxacin, the average QT interval, corrected by Bazett's formula, showed a 17.2±27.1 ms prolongation with a lower bound of the 95% confidence interval of 4.9 ms.

When the modification rules were applied, the accuracy of QT measurement was improved, and the present system was capable of detecting QT prolongation correctly.