Automated recognition of corrupted arterial waveforms using neural network techniques.

A data acquisition system that automatically discards corrupted or undesirable signals would save untold hours of drudgery for researchers. Continuous recording of variables to provide detailed behavior patterns generates huge amounts of raw data. Unfortunately waveforms usually require visual inspection for isolating desired behavior or validating signal integrity. This tedious and time-consuming step can potentially be eliminated using a novel computer science technique. We have trained a simulated neural network to recognize corrupted arterial pressure waveforms. Our system can now evaluate the validity of the arterial waveform without human intervention with an average false positive error rate of 2.2% and an average false negative error rate of 12.6%.