A comparison study between fainter and non-fainter subjects during head-up tilt test using reconstructed phase space
Abstract
The analysis of cardiac dynamics based on time series extracted from cardiovascular signals (e.g. electrocardiogram, blood pressure) is relevant for differentiating between normal and pathological cases with feasible functions in the diagnosis and risk estimation. In this study, the dynamic behavior of cardiovascular time series is analyzed using reconstructed phase space to identify differences between subjects who developed syncope during head-up tilt test (fainters) and others who did not (non-fainters). Electrocardiogram and arterial blood pressure were recorded from 29 non-fainter and 28 fainter subjects. RR-interval, Amplitude of Systolic blood pressure (AmpS), peak amplitude of the first derivative of blood pressure (dPdt-max) and Pulse Transit Time (PTT) were extracted. Different features, such as the phase space density and indices derived from the recurrence quantification analysis, were computed from the phase space area of the above cited time series. In order to identify fainter and non-fainter groups, we selected the most pertinent parameters using Relief method to be used for further classification by K-nearest neighbor. The results show that the performance of the classification is approximately the same in all these time series with sensitivity (Se) near to 66.5% and specificity (Sp) around 62% during the first 5min of supine position. These values increase in the first 15min of tilted position to Se= 67% and Sp= 73%. Using an optimal fusion node, we demonstrate that the joint analysis of RR and dPdt-max provides a sensibility around 95% and a specificity of 87%. This analysis suggests that a bivavariate analysis enhances the classification performance, and help predict the outcome of the HUTT. © 2015 CCAL.
Keywords
Blood pressure
Cardiology
Electrocardiography
Living systems studies
Nearest neighbor search
Phase space methods
Risk assessment
Risk perception
Time series
Arterial blood pressure
Cardiovascular signals
Classification performance
K-nearest neighbors
Phase space densities
Reconstructed phase space
Recurrence quantification analysis
Systolic blood pressure
Time series analysis