Development of a combined time-frequency technique for accurate extraction of pNN50 metric from noisy heart rate measurements

Abstract

In the design of human-in-the-loop robotic systems , it is of importance to evaluate human mental workload changes as this could inform the robots the degree to which a human is overwhelmed and error-prone in decision making. With an accurate assessment of mental workload, one can modulate the robots accordingly with the aim to assist the human in a task toward reduced mental workload. One key metric reliably utilized in the literature to assess human mental workload changes is called the pNN50—a metric associated with variations in the instantaneous frequency of human heart rate (HR). pNN50 can be extracted, for instance, from photoplethysmography (PPG) sensor measurements by implementing a well-known time-domain technique that is based on inter-beat intervals (IBI) of HR. When PPG measurements are contaminated with noise, the traditional time-domain approach may lead to inaccurate estimations of the pNN50 metric, as it heavily depends on precise calculation of the IBI from time series. In this manuscript, we present a combined time-frequency technique to remedy this problem. This new approach does not rely on time-domain based IBI data; and, hence, it can effectively avoid problems associated with noisy signals, rendering reliable computation of the pNN50. Examples using noisy synthetic signals and experimentally measured PPG data with and without noise contamination confirm the benefits of the proposed technique over the traditional time-domain based approach.