Non-Invasive Inference of Minute Ventilation Using Wearable ECG and Gaussian Process Regression

Abstract

Continuous assessment of air pollutant exposure is vital for patients with chronic pulmonary diseases such as asthma, bronchitis, and emphysema. The effective dose of such exposure is directly proportional to the minute ventilation, aka respiratory minute volume (V E ). Spirometry – still the clinical standard for measuring V E – is highly invasive and not suitable for continuous use in most applications. This paper presents a novel non-invasive method toward continuous assessment of VE using a chest-mount wearable ECG sensor. Data are collected from 25 healthy subjects while performing ambulatory and sedentary activities and physical exercises. The ECG signal is processed to overcome challenges associated with baseline drifting, noisy skin contact, and motion artifacts, and to extract robust and explanatory ECG features. These features are used to train and evaluate multiple regression models, among which, the Gaussian process regression models achieve the lowest error in both learning and inferring V E from the wearable ECG signal. The impacts of inter- and intrapersonal variations on the model performance are shown to reveal the potential of the proposed method for continuous monitoring of pollutant exposure risk in respiratory health applications.