Abstract: | Arterial blood pressure is monitored during cardiac operations and subsequent intensive care. During this time medical procedures such as taking a blood sample, can cause surges in the pressure and trigger threshold alarms. Such alarm signals are often considered inappropriate. The wave forms created by the medical procedures, known as artefacts, are characteristic and recognisers have been built to suppress inappropriate alarms. This project looks at applying feedforward neural networks and back propagation to the classification of three artefacts and the normal pulsatile wave. Problems were encountered with local minima and some investigation of error functions and a simple adaptive parameter technique were undertaken to address this problem. Networks were trained that could recognise the leading edge of the different artefacts and it was felt that these could contribute to a full artefact recognition system.
|