Physiological Noise: Definition, Estimation, and Characterization in Complex Biomedical Signals.

Abstract

Nonlinear physiological systems exhibit complex dynamics driven by intrinsic dynamical noise. In cases where there is no specific knowledge or assumption about system dynamics, such as in physiological systems, it is not possible to formally estimate noise. We introduce a formal method to estimate the power of dynamical noise, referred to as physiological noise, in a closed form, without specific knowledge of the system dynamics. Assuming that noise can be modeled as a sequence of independent, identically distributed (IID) random variables on a probability space, we demonstrate that physiological noise can be estimated through a nonlinear entropy profile. We estimated noise from synthetic maps that included autoregressive, logistic, and Pomeau-Manneville systems under various conditions. Noise estimation is performed on 70 heart rate variability series from healthy and pathological subjects, and 32 electroencephalographic (EEG) healthy series. Our results showed that the proposed model-free method can discern different noise levels without any prior knowledge of the system dynamics. Physiological noise accounts for around 11% of the overall power observed in EEG signals and approximately 32% to 65% of the power related to heartbeat dynamics. Cardiovascular noise increases in pathological conditions compared to healthy dynamics, and cortical brain noise increases during mental arithmetic computations over the prefrontal and occipital regions. Brain noise is differently distributed across cortical regions. Physiological noise is very part neurobiological dynamics and can be measured using the proposed framework in any biomedical series.