Volumetric CO2 data from patients in anaesthesia delivery systems are sought after by physicians. The CO2 data obtained with the commonly used sidestream sampling technique are not considered adequate for volumetric CO2 estimation due to distortion and desynchrony with patient flow. The purpose of this thesis was to explore the possibility of using signal enhancing methods to the sidestream data to accurately estimate CO2 flow using a Flow-i anaesthesia delivery system.

To evaluate sidestream performance, experimental data was acquired using a mainstream and a sidestream capnograph connected in series to a FRC test lung with known CO2 content, ventilated by a Flow-i anaesthesia machine. The data was then enhanced and analysed using signal processing methods including sigmoid modelling and neural networks.

A Feed Forward Neural Network achieved results closest resembling the mainstream capnogram of the evaluated signal processing methods. The mainstream capnogram, considered the benchmark, produced large internal scattering and approximately 25 % offset from actual CO2 flow while using the inherent patient flow data produced by the Flow-i anaesthesia system. When using patient flow data from a Servo-i ventilator, the resulting CO2 flow estimates were drastically improved, producing estimates within 10 % error.

This thesis concludes that there are several potential processing methods of the sidestream data to approximate the mainstream signal, however the patient flow of the Flow-i system are a suspected source of error in the CO2 flow estimation.