Some noise is easy to detect – simply plotting the time courses of NIRS signal, you will find the noise so out of range and you will discard the data. However, I find in some instances the noise level seems to be in the normal range but the data should be discarded based on a new criterion.
This criterion is the correlation between oxy and deoxy-Hb. oxy and deoxy-Hb should be negatively correlated most of the time. If you plot a running correlation you normally see fluctuations close to -1. However, when you find the correlation is actually equal to 1 for all the time points, there must be something wrong. In this case, you should discard the data – even though the original signal may seem normal.
This demonstrates the utility of using correlation between oxy and deoxy Hb in NIRS data analysis. We previously had used this correlation to improve signal quality. Check out our paper.
If you find something perfect, be careful.