Denoising of Heart Rate Variability signals during tilt test using independent component analysis and multidimensional recordings

Abstract

Vasovagal Syncope (VVS) represents the most frequent cause of loss of consciousness. Additionally to its clinical usefulness, the tilt test is a good quality physiological gold standard for the spectral analysis of Heart Rate Variability (HRV). Noise removal in HRV signals is problematic, due to the presence of ectopic beats and non-stationary short-term trends. Given current Tilt Test systems simultaneously record several physiological signals, we hypothesize that independent component analysis (ICA) may separate physiological from mostly-noise components, and denoising can be properly done. Four-dimensional recordings (HR, systolic/diastolic blood pressure, and ejection volume) were obtained during 50 Tilt Test. After ICA decomposition, a 5th order median filter was applied to the noisiest component, prior to signal reconstruction. In order to check the denoising performance, a gold-standard was made by manually removing ectopic beats and artifacts from the original signals by an expert. For comparison purposes, a 5th order median filter was also applied separately to the HR signal. The spectrum analysis showed that denoising of multidimensional recordings with ICA during Tilt Test yields HRV signals with lower distortion at HF band.