Eigenvector-based spatial ECG filtering improves QT delineation in stress test recordings

The electrocardiogram signal (ECG) represents the electrical activity of the heart measured at the body surface. Characteristic waves and their delineation marks are studied to define cardiac markers without using invasive procedures. Among them, slower adaptation of the QT interval, the time needed for ventricular depolarization plus repolarization, to sudden abrupt changes in heart rate (HR) has been identified as a marker of arrhythmic risk. Such abrupt HR changes are difficult to induce, leading here to explore estimation of this delay from the ramp-like HR variations observed in exercise stress test. However, stress test ECG signals are often corrupted by muscular activity and electrode motion, limiting the robustness of the information that can be extracted from them, as the identification of the T-wave end. The aim of this study is to find proper methods to emphasize the T-wave in order to improve delineation accuracy. Stress test ECG recordings from 447 subjects were analyzed. The first spatially-transformed lead based on two different lead-space reduction (LSR) techniques, and different learning versions, was used to delineate and obtain the QT series. Assuming that QT delineation errors will lead to a high variability in the QT interval series, the power of the high-pass filtered QT interval series was used as performance marker (the lower the power, the most stable the delineation). Periodic component analysis technique showed the lowest power, with no significant differences between its different learning versions.