Scalp coupling index
The scalp coupling index is a quality check to assess the correlation between the two wavelength channels in the cardiac band. It is originally described in [1]. It is applied separately to each pair and comprises the following steps:
- the signal is bandpass-filtered to keep only the cardiac band. In [1], they use a rather wide band of [0.5 - 2.5] Hz, but this can be left to the user to define.
- amplitude normalization
- SCI computation = absolute cross-correlation value at 0-time lag
For in-phase and counter-phase identical waveforms SCI=1, whereas SCI=0 for totally uncorrelated signals, eg where there is only noise in both channels. A threshold of SCI > 0.75 is recommended by the authors in [1].
Input: fNIRS time-series
Parameters: bandpass window to extract the cardiac band (Hz)
Ouput: SCI map
The following SCI map could be obtained from the sample NIRS finger tapping data set:
Here are the signals for the pair S1D5, associated with an SCI of 0.7:
The signal wavelength 685nm has a rather low SNR with not clear heart beats. This means that the SCI will be low even if the other wavelength (680nm) has high SNR and clear heart beats. This is the main concern in this data set: most of the information is carried by the 830nm.
Here are the signals for the pair S1D6, associated with an SCI of 0.1:
SNR is comparable across wavelengths, but heart beats are not very clear. This is why SCI is very low.
Lastly, not that a lot of channels have low SCI (<0.1) but some of these channels do contain hemodynamics activity with a clear HRF. Scalp coupling index is hence not an absolute quality check and should be supplemented with other measures and manual inspections.
Scalp coupling index computation is integrated in the process that tags bad channels, see this page