A fast noise-tolerant ECG feature recognition algorithm based on probabilistic analysis of gradient discontinuity

• Published in: Journal of electrocardiology Vol. 50; no. 4; pp. 491 - 503
• Main Author: Watterson, Peter A.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.07.2017; Elsevier Science Ltd
• Subjects: Algorithms; Cardiology; Cardiovascular; Databases, Factual; Electrocardiography; Electrodes; Feature extraction; Feature recognition; Fiducial points; Heart rate; Humans; Image processing systems; QT interval estimation; Reproducibility of Results; Signal Processing, Computer-Assisted; Electrocardiography; Fiducial points; QT interval estimation; Feature recognition; feature recognition; fiducial points; electrocardiography
• ISSN: 0022-0736; 1532-8430; 1532-8430
• DOI: 10.1016/j.jelectrocard.2017.03.013

Improvement in real-time electrocardiogram (ECG) interpretation is still needed, especially for QT estimation. This paper proposes a fast algorithm for ECG feature recognition, based on locating turning points in the waveform gradient. The algorithm places the fiducial point at the maximal value of a probabilistic decision function, assessing line intervals of best fit before and after the point and the point location relative to R-wave peaks already found. Fiducial points were successfully located for the 30 heartbeats annotated by a cardiologist of all 10 normal sinus rhythm records from the PhysioNet QT Database. For a given subject, the algorithm's QT estimation had superior repeatability, with intrasubject QT standard deviation just 5.42ms, 60% lower than the cardiologist's 13.57ms. Initial tests suggest immunity to noise of standard deviation up to about 9% of the signal, depending on noise type. The proposed algorithm is fast to calculate and noise-tolerant, and has shown improved repeatability in its QT estimation compared to a cardiologist. •A fast algorithm for ECG interpretation is proposed, based on gradient discontinuity.•The line interval fitting used provides high frequency filtering and noise tolerance.•Normal sinus rhythm records from the PhysioNet QT Database are successfully analysed.•The algorithm achieved lower intrasubject QT variability than a cardiologist.