Methods for the analysis of short-term variability of heart rate and blood pressure in frequency domain

Cardiovascular variability signals provide information about the functioning of the autonomous nervous system and other physiological sub-systems. Because of large interand intra-subject variability, sophisticated data analysis methods are needed to gain this information. An important approach for analysing signals is the analysis in the frequency domain. In this thesis, spectral analysis of cardiovascular variability signals was addressed by two different approaches. The first approach was based on univariate spectral analysis. The novelty of the approach is the quantification of the shift in spectral power within a frequency band. Three different estimators for the spectral shift were compared. The band-wise mean and median frequencies were found to provide better performance than the parameter used in earlier studies, namely central frequency. The band-wise median frequency was successfully applied to real clinical data. In the other approach multivariate closed-loop analysis of the cardiovascular system was studied. A framework based on linear time series modelling and spectral decomposition was presented. The application of multivariate autoregressive (MAR) modelling on real cardiovascular data was addressed in detail, and a method for overcoming the problem of correlating noise sources in MAR modelling was applied successfully. A non-causal model for controlling the effect of respiration on cardiovascular system was proposed. Practical considerations of applying multivariate linear time series modelling to real cardiovascular data were discussed. Methods were demonstrated using real data.