Maximizing the Reliability of a Full-Automatic ECG-Waveforms Delineating Algorithm Using Extensive ECG Databank

Antoun Khawaja
Khawaja Medical Technology GmbH


There is in fact a large interest in accurate ECG delineation and measurements not only because they play a huge role in further ECG analysis, but also because they are considered themselves important biomarkers for making decisions. Some ECG measurements, like RR intervals, QT/QTc interval prolongation, are even considered as the standard surrogate biomarker for cardiac drug safety as stated by ICH clinical evaluation guideline E14. The objective of this work is to validate the precision of the delineation program published in [1] using big number of manually annotated ECG signals. This databank contains more than seventy thousand 12- lead resting ECG records from Thorough QT (TQT) trials and several late-phase clinical studies. For each ECG signal, a full corresponding set of measurement including QT interval and RR interval is annotated manually by various cardiologists and certified medical experts. Those annotations are considered as golden reference for this work. The measurements obtained by the delineation algorithm [1] using the databank under study were compared with golden reference including RR intervals and QT intervals. The mean, median and standard deviation of the differences between the full-automated program and the golden reference for all relevant interval and segment measurements are very low. For QT interval measurements, the mean, median and standard deviation of the differences are less than 2, 3 and 10 [ms], respectively. The results obtained in this work express that the algorithm described in [1] is a useful tool in all kind of ECG applications including resting ECG, Holter analysis, stress ECG, ECG monitoring and for the evaluation of cardiac safety in clinical trials. We interpret these findings as indicator of the reliability of our algorithm in the full-automatic measurements of ECG waveforms.
[1] ISSN: 2325-887X DOI: 10.22489/CinC.2018.209