Identification of QRS complex in non-stationary electrocardiogram of sick infants. (1st August 2017)

• Record Type: Journal Article
• Authors: Kota, S.; Swisher, C.B.; Al-Shargabi, T.; Andescavage, N.; du Plessis, A.; Govindan, R.B.
• Is Part Of: Computers in biology and medicine. Volume 87(2017)
• Page Start: 211
• Page End: 216
• Publication Date: 2017-08-01
• Subjects: Electrocardiogram -- Hilbert transform -- Premature birth -- Non-stationary; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2017.05.033
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: Background: Due to the high-frequency of routine interventions in an intensive care setting, electrocardiogram (ECG) recordings from sick infants are highly non-stationary, with recurrent changes in the baseline, alterations in the morphology of the waveform, and attenuations of the signal strength. Current methods lack reliability in identifying QRS complexes (a marker of individual cardiac cycles) in the non-stationary ECG. In the current study we address this problem by proposing a novel approach to QRS complex identification. Method: Our approach employs lowpass filtering, half-wave rectification, and the use of instantaneous Hilbert phase to identify QRS complexes in the ECG. We demonstrate the application of this method using ECG recordings from eight preterm infants undergoing intensive care, as well as from 18 normal adult volunteers available via a public database. We compared our approach to the commonly used approaches including Pan and Tompkins (PT), gqrs, wavedet, and wqrs for identifying QRS complexes and then compared each with manually identified QRS complexes. Results: For preterm infants, a comparison between the QRS complexes identified by our approach and those identified through manual annotations yielded sensitivity and positive predictive values of 99% and 99.91%, respectively. The comparison metrics for each method are as follows: PT (sensitivity: 84.49%, positive predictive value: 99.88%), gqrs (85.25%, 99.49%), wavedet (95.24%, 99.86%), and wqrs (96.99%, 96.55%). Thus, the sensitivity values of the four methods previously described, are lower than the sensitivity of the method we propose; however, the positive predictive values of these other approaches is comparable to those of our method, with the exception of the wqrs approach, which yielded a slightly lower value. For adult ECG, our approach yielded a sensitivity of 99.78%, whereas PT yielded 99.79%. The positive predictive value was 99.42% for both our approach as well as for PT. Conclusions: We propose a novel method for identifying QRS complexes that outperforms common currently available tools for non-stationary ECG data in infants. For stationary ECG our proposed approach and the PT approach perform equally well. The ECG acquired in a clinical environment may be prone to issues related to non-stationarity, especially in critically ill patients. The approach proposed in this report offers superior reliability in these scenarios. Highlights: Introduced a novel approach to identify QRS complexes in a non-stationary ECG. Approach tested on ECG from both preterm infants and normal adults. Performance was assessed using manual annotations as reference. Proposed approach outperformed four different approaches in preterm ECG.