Real-time classification of patient-ventilator interaction using surface electromyography

Abstract

Patient-ventilator asynchronies (PVA) are common during mechanical ventilation and are associated with discomfort, prolonged stay in the intensive care unit (ICU), and increased mortality. Respiratory surface electromyography (sEMG) enables continuous, non-invasive monitoring of patient effort. For automatic assessment of PVA in the ICU, robust, real-time segmentation and classification algorithms are required. In this article, we adapt two previously published offline algorithms for inspiratory onset and offset detection in respiratory sEMG into real-time capable implementations and evaluate their performance for detecting different forms of PVA. Both proposed real-time algorithms achieve high detection performance with median sensitivity and positive predictive value (PPV) above 0.9. For PVA classification, offline and real-time algorithms yield similar results with median PPV above 0.8 for most types of PVA. Our results demonstrate the potential of respiratory sEMG in the ICU for non-invasive, automatic and real-time monitoring of PVA.