Parameter Identification of Respiratory Mechanics Based on Convolutional Neural Networks

Mechanical ventilation using a ventilator is a common method for the treatment of acute and critical illnesses, and ventilator-associated lung injury is one of the main complications during mechanical ventilation, usually caused by poor matching between ventilation parameters and patients' lung function, and currently, respiratory mechanics parameters such as airway resistance and lung compliance are mostly used to characterize the lung function, therefore, the study of respiratory mechanics parameter identification has important clinical significance and application value for guaranteeing the therapeutic efficacy of mechanical ventilation. and application value. In this paper, a first-order single-chamber respiratory mechanics model of the mechanical ventilation process was established, and a respiratory mechanics parameter identification algorithm based on the least squares method and convolutional neural network was designed. The data of main airway flow, pressure and tidal volume changes under different values of mechanical parameters were simulated, and the deep learning model was trained. A mechanical ventilation simulation experimental platform was constructed using an active simulated lung and a ventilator to simulate the mechanical ventilation process, and an airflow analyzer was used to record pressure, flow, and other simulated ventilation data. The simulation data were used to experimentally validate the respiratory mechanics recognition algorithms based on the least squares method and deep learning, respectively, and the results showed that the two were close to each other in terms of airway resistance and lung compliance recognition errors.