A Comparative Study of Deep Learning Models for Patient-Ventilator Asynchrony Classification using Phase Space Reconstruction

Mechanical ventilation is a crucial treatment for patients with respiratory failure, but patient-ventilator asynchrony (PVA) can lead to serious complications. Accurately identifying different types of PVA events is challenging due to inter-patient variability. In this study, we propose a novel approach for PVA classification using phase space reconstruction (PSR) and transformer models. Experiments on two cross validation tasks show the proposed model to be superior in quality while being more stable. In conclusion, PSR can effectively capture long-term features of the dynamical system and improve the performance of transformer models in distinguishing between various PVA events.