Prediction of Ventilator Breathing Parameters on an Automated Ventilation System using Regression Models

Medical staff face issues when ventilating patients manually using the Bag-Valve-Mask (BVM) for long periods. As for ventilators in hospitals, medical specialists must check on patients frequently and adjust settings manually. This study aims to conduct machine learning (ML) study using data collected from the CHU Sainte-Justine database to provide suitable setting recommendations for patients using linear and Poisson regression. The response variables were the fraction of inspired oxygen (FiO2) setting, positive end-expiratory pressure (PEEP) setting and tidal volume (TV) setting. Linear regression scored an R2 value of 0.936 and Poisson regression scored 0.836 when predicting TV setting. Using linear regression, effects of each predictor variable on the FiO2 setting, PEEP setting and TV setting using different datasets was carried out to study their respective R2 values, where TV setting was deemed the best response variable. To validate the TV setting formula obtained through ML, three experiments were conducted using a ventilator prototype on an artificial test lung. The experiments yielded error results ranging from 53% to 79%, indicating that the TV setting values obtained from the prototype were incomparable to mechanical ventilator data.