Accurate segmentation of respiration waveforms from infants enabling identification and classification of irregular breathing patterns.

Although the precise causes of Sudden Infant Death Syndrome (SIDS) are still unclear there is evidence to suggest that hypoxaemia may be a contributory factor. Transcutaneous oxygen monitors can be used but are unsatisfactory for young babies in the home. As an alternative approach, respiratory patterns can be studied but attempts at classification of individual breaths are often unsuccessful particularly during periods of extraneous noise or movement artefact. We have developed a robust algorithm which provides accurate segmentation and classification of breaths even in the presence of noise or movement. This improves on previous techniques by deferring the decision on an uncertain candidate breath until more information is available; yet the delay incurred is two breaths at most. The use of look-up tables and decision trees means that computational requirements are kept to a minimum and implementation in a simple home monitor is quite possible.