Towards modelling of patient-ventilator interactions using model based methods

Mechanical ventilation is an important life-saving intervention on the ICU. Lung-protective ventilation techniques such as pressure support ventilation (PSV) are used frequently in the ICU. However, asynchronies, poor patient-ventilator interactions during PSV, are shown to be harmful and are linked with increased lung injury and mortality. There is a need for automatic detection and classification of asynchronies for clinical studies, algorithm development and for real time clinical decision support for smart ventilation technologies. So far, reasonable results of detection of asynchronies have been obtained, but classification is still a challenge. In this work, we generate training and classification waveforms for our machine learning study using a patient-ventilator simulation model. From these models the flow, pressure and volume waveforms can be created for different types of parameter settings. Note that the type of asynchrony and timing of the patient effort are known.