Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Analysis of the so called Heart Rate Variability (HRV) is a method for non-invasive assessment of autonomic function disturbances. HRV is based on the analysis of the beat-to-beat fluctuations in heart rate which are considered as a reflection of cardiac autonomic modulation. Reduced HRV is commonly found in patients with the disease Familial
Amyloidotic Polyneuropathy (FAP), where autonomic deregulation is a common problem.
In this study the focus was on the autonomic function test Deep Breathing, where the study group consisted of FAP patients with both regular and irregular HRV patterns, as well as healthy control subjects. In addition to traditional univariate HRV analysis, the influence of respiration was included by coherence and transfer function analysis. The system modeling was performed by two different methods: Fourier-Based (FB) and Parametric Modeling (PM). In the PM method coherence was estimated based on bivariate autoregressive modeling, whereas transfer function analysis was performed based on the State Space (SS) model of the relation between respiration (input) and HRV (output).
To evaluate the performance of the different methods that were used in this study, the analysis was first performed using synthesized signals. The synthesized signals were generated based on the assumption that the HRV and respiration signals often are nearly
sinusoida1. The results of this analysis, where the true gain of the transfer function was known, could be compared with the FB method and PM method and it showed that in most sets of synthesized data, the most accurate results were obtained by the FB method.
However, for some sets of data, a better result was obtained in the estimated transfer function by using the SS model, i.e., the PM method.
The main problem with the PM method is the selection of an appropriate model order. The results for both synthesized data and real data showed that increasing the number of order is not necessarily a good choice to have a more accurate result in PM method. Although the automatically determined "best" model order often appeared to give acceptable results, a fixed model order in the range 5-8 also might be a good compromise.