Evaluation of an algorithm for highly automated measurements of QT interval

Noise, artifact, and labile morphology of ECGs collected from freely moving animals in safety pharmacology studies render accurate measurements of QT interval challenging. Consequently, a high percentage of beats are uninterpretable and results provided by currently available analysis algorithms often require extensive manual review to correct errors. Performance of a novel algorithm, Multi-Domain Signal Processing™ (MDSP), is evaluated as a means of removing noise (denoising) without distorting morphology and for obtaining accurate beat-to-beat QT measurements.

Performance was evaluated using controlled experiments and an observational evaluation as follows: a) a clean ECG strip was intentionally corrupted with varying levels of noise to provide recordings of known signal-to-noise ratio (SNR). SNR and fidelity were compared pre- and post-MDSP denoising, b) beat-to-beat QT of a noisy ECG was measured manually pre- and post-MDSP denoising and automatically by MDSP, c) beat-to-beat QT of a clean ECG was measured manually and automatically using MDSP, and d) beat-to-beat QT was computed for 3 freely moving non-human primates (NHP) pre- and post-torsadogen administration and the impact of averaging on QTSD, QT/RR dynamics and relationship was evaluated.

MDSP reduced noise amplitude by up to 85% while preserving signal morphology. Mean QTs for manual and automatic measurements on a noisy ECG were within 2±15ms. MDSP-denoising prior to manual QT measurements resulted in a 22% decrease in QTSD compared to measurements obtained without denoising. Average QT standard deviation of the mean (QTSD) for automatic MDSP-derived measurements for 3 freely moving subjects was 7ms with 2.5% of beats automatically excluded due to noise.

This work demonstrates that the MDSP algorithm shows promise as a tool for providing accurate automatic beat-to-beat measurements of QT interval from NHP in safety pharmacology studies. A methodology is presented for characterizing the impact of noise on algorithm performance.