Comparative investigations of algorithms for the detection of breaths in newborns with disturbed respiratory signals

The correct detection of the beginning of inspiration and expiration in the respiratory signals is an essential prerequisite for accurate lung functio n testing in newborns. Five algorithms for breath detection using pneumotac hographically measured flow and volume signals were investigated with regar d to the error rate. To compare and to evaluate the reliability of these al gorithms 12 minimally and 12 severely disturbed flow and volume signals fro m spontaneously breathing newborns were used. With the exception of an algo rithm based on Walsh-transformed signals, all algorithms work reliably terr or rate <1.1%) if disturbances are minimal. In severely disturbed signals t here is a great difference between the algorithms. The most robust algorith m tested (trigger of the how signal with an additional plausibility check o f the recognized breath) resulted in an error rate of <3.4%. Not all algori thms tested are suitable for real-time applications because they differ con siderably in delay time for breath detection. (C) 1998 Academic Press.