Heart Rate Variability in the Frequency Domain

Jon Bergenheim

Master Thesis, Report UPTEC F00012

Abstract:

The technique of estimating the autonomic activity from an E.C.G. signal recording is explained from the point of data recording to the calculation of the autonomic indicators. Three different spectral estimation techniques are discussed: General Spectral Analysis (G.S.A.), Coarse Graining Spectral Analysis (C.G.S.A.) and recursive auto-regressive modelling (AR). All methods estimate a spectrum of the heart rate variability and calculate the sympathetic (S.N.S.) and parasympathetic (P.N.S.) indicators from the low-high frequency balance.

The Coarse Graining Spectral Analysis extracts the fractal components of the heart rate variability by taking the Fourier transform of the cross-correlation of the original time series and a rescaled version of it. This causes the spectrum of the harmonic components to approach zero while the fractal components, being self-similar, maintain their spectral properties. The harmonic spectrum can then be calculated by subtracting the fractal spectrum from an original spectrum that includes both harmonic and fractal components. With recursive autoregressive (AR) modelling, it is possible to track time varying parameters providing a much higher esolution in time.

The methods are first tested on data simulated by the I.P.F.M. model and then on data recorded in the Head Up Tilt experiment. All methods showed an increase of the S.N.S. indicator and a decrease of the N.S. indicator with increased tilt level. The low frequency peak was enhanced by the C.G.S.A. method but should be slightly modified for more accurate results. If long recordings of stationary data are available, either the modified version of the C.G.S.A. or AR modelling should be used instead of the G.S.A.. For shorter time periods than 3-5 minutes the AR modelling method is recommended. If transient conditions are to be analysed the recursive A.R. modelling is very promising.

Organization:

Centre for Musculoskeletal Research,
National Institute for Working Life, Umeå

Thesis Advisors:

Milos Ljubisavljevic, National Institute for Working Life, Umeå,
Eugene Lyskov National Institute for Working Life, Umeå,
Mikael Sternad, Signal and Systems Group, Uppsala University