Shape-only identification of breathing system failure

Breathing system failure accounts for approximately 7% of all critical incidents during anaesthesia. Current smart alarm solutions to this problem tend to have been developed for a specific manufacturers equipment and use relatively expensive sensors. What is needed is an intelligent alarm capable of working in all systems from easily available signals. This paper reports the results of research into a shape-only alarm system aimed at providing such a system. It uses pressure, flow and capnograph waveforms gathered at the patient connector. Single breath segments from these waveforms were extracted and roughly synchronised and normalised to the same dynamic range and time base, i.e. the waves were plotted on a 1×1 x/y graph regardless of the original amplitude or breathing rate, in a simple similarity transformation. A neural network classifier was then trained to recognise the failure modes from the shape of these segmented waveforms after further pre-processing including a genetic algorithm search for relevant features within the waveforms. The system has been tested using an Enclosed Afferent Reservoir (EAR) and a Bain breathing system during simulated spontaneous and controlled ventilation. Correct classification rates for failures of 97.6% and 94.3% were obtained for the EAR and the Bain studies respectively in the face of over 100 unseen simulated failures for each breathing system. In both studies only one false positive alarm, i.e. an indication of a failure when no failure was present, was indicated by the alarm, and only one false negative instances were observed