Spectral estimation of HRV in signals with gaps. (July 2019)

• Record Type: Journal Article
• Authors: Rodríguez-Liñares, L.; Simpson, D.M.
• Is Part Of: Biomedical signal processing and control. Volume 52(2019)
• Page Start: 187
• Page End: 197
• Publication Date: 2019-07
• Subjects: Heart rate variability -- Biomedical signal processing -- Electrocardiography -- Spectral analysis -- Data loss; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.04.006
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A major problem in the analysis of HRV are the often unavoidable artefacts, which can render the data useless. A great benefit of using surrogate data in spectral analysis is that the 'true' spectrum is known in advance, and all estimates can be compared to this. Even with many gaps (up to 50%), HRV recordings need not to be discarded and spectral parameter estimates can still be used. Algorithms such as Burg for segments can provide an accurate estimate from the remaining high quality segments. Abstract: Heart rate variability is commonly quantified following spectral estimation. However, it is often difficult to obtain continuous recordings of beat-to-beat intervals without interruptions due to artefacts, noise or sporadic arrhythmias. Such data loss may be seen as gaps in the recordings, and often results in such signals being discarded. While a number of methods has been proposed for spectral estimation in heart rate records with gaps, there are no comprehensive comparisons between them. This paper tries to fill this void, comparing methods and identifying the most versatile and reliable one. The mean (bias error) and standard deviation (random error) of estimates of power in the low frequency band (LF), from 0.04 to 0.15 Hz; in the high frequency band (HF), from 0.15 to 0.4 Hz; and their ratio (LF/HF), were calculated in RR-interval time-series with up to 50% of samples missing through large or small gaps introduced into recordings. 'Correlogram (bridging)' and 'Burg for segments' methods proved to be the most robust methods for dealing with gaps, but Burg for segments was found to be more robust, especially in the HF band. Our results clearly show that even large gaps (covering a total of 50% of the recording time) can still yield robust spectral estimates of HRV, provided appropriate methods are used.