Fuzzy C-Means Clustering and New-Structure Particle Swarm Optimization for Modelling of Relative Dead-Space and Carbon Dioxide Production

Abstract

Mechanical Ventilation plays a major role for life support of critically-ill patients in the Intensive Care Unit. Medical practitioners assess patient's oxygenation status by observing the blood gases from arterial blood samples. However, this procedure to sample arterial blood is invasive and must be done cautiously. This paper proposes new fuzzy logic-based models for estimating non-invasively the relative ratio of dead space to the tidal volume, known as relative dead-space and the production of carbon-dioxide of ventilated patients. These parameters are needed for a non-invasive and automatic blood gas estimation system called the SOPAVent system. The fuzzy models are designed using fuzzy c-means clustering and new-structure particle swarm optimization technique which looks at the coefficient of determination and the mean squared error as performance indices. The prediction results are validated with actual ICU patients. The simulation results showed high accuracy in prediction of the relative dead-space parameter and the production of carbon-dioxide parameter.