Multimodal vs Unimodal Estimation of Sympathetic-Driven Arousal States

Abstract

Estimation of sympathetic-driven arousal state (SDAS) traditionally consists of computing frequency-based measures of heart rate variability. However, in the presence of confounds such as breathing frequency, these measures can incorrectly estimate the underlying SDAS. In this work, we present an example of such a case during a three-stage paced breathing task. Using a state space framework, we demonstrate that a unimodal model that relies solely on these frequency-based heart rate variability measures overestimates SDAS during the slowest breathing stage and underestimates it in subsequent stages. On the other hand, a multimodal model with both time and frequency domain heart rate variability observations as well as electrodermal activity information provides a more realistic estimate of SDAS throughout the task. This suggests that multimodal estimation of SDAS is more accurate and robust than unimodal estimation.