Blood Gas Predictions for Patients Under Artificial Ventilation Using Fuzzy Logic Models

Abstract

This paper proposes a new modelling framework for accurate predictions of arterial blood gases (ABG) of the previously developed SOPAVent model (Simulation of Patients under Artificial Ventilation). Three ABG parameters which were elicited from the SOPAVent model are the partial arterial pressure of oxygen (PaO₂), the partial arterial pressure of carbon-dioxide (PaCO₂) and the acid-base (pH). SOPAVent generate predictions of initial ABG and predictions of ABG after ventilator settings were modified. SOPAVent’s sub-models, the relative dead space (Kd) and the carbon-dioxide production (VCO₂) were designed using interval type-2 fuzzy logic system (IT2FLS). Further explorations of the models were carried out using fuzzy c-means clustering (FCM) and tuning of fuzzy parameters using ‘new structure’ particle swarm optimization algorithm (nPSO). The new models were integrated into the SOPAVent system for blood gas predictions. SOPAVent was validated using real intensive care unit (ICU) patient data, obtained from the Royal Hallamshire Hospital and Northern General Hospital, Sheffield (UK). The prediction accuracy of SOPAVent was compared with the pre-existing SOPAVent model where the Kd and VCO₂ sub-models were developed using machine learning algorithms. Significant improvements in accuracy and correlation were achieved under this frameworks for PaCO₂ and pH for both the initial ABG predictions and the post ventilator settings adjustments.

Appendix