2.4 Data Preprocessing and Analysis
Hemodynamic data was preprocessed with a combination of nirsLAB (an analysis software that comes with the NIRSport system), Homer2 (Huppert et al., 2009) and custom scripts in MATLAB R2017B. The raw data was converted into optical densities. The channels that fell outside the 80-140 dB optical density range were excluded from further analysis. Motion artefacts were corrected via the wavelet method (Cooper et al., 2012). Subsequently, data was band-pass filtered with a third-order Butterworth filter with cut-off frequencies of 0.05 and 0.2 Hz. Since the system did not feature short-channel separation, a PCA filter was also used to remove non-cortical signals that remained within the band-pass filter frequency range. Lastly, preprocessed optical densities were converted into oxy-hemoglobin (HBO) and deoxy-hemoglobin (HBR) concentrations. Only HBO data was used in further analysis as it was claimed that cortical activation is better reflected by HBO (Dravida et al., 2017; Watanabe et al., 2002).
Accuracy and reaction times were computed for each condition and subject. For trials in which subjects did not give an answer within 3-second long response window, reaction time was set as 3 seconds. A paired sample t-test was used to analyze the accuracy and reaction time data.
To analyze hemodynamic activity, fourteen seconds long segments were generated by using the two seconds pre-stimulus baseline window and twelve seconds post-stimulus activity windows for each trial. Then, the segments were detrended and classified as either SRC or ORC. Subsequently, the average across trials was computed to generate single SRC and single ORC timeseries in each fNIRS channel of each subject. Lastly, the hemodynamic activity induced by conditions was measured by computing the hemodynamic activity strength (HAS) (Mutlu et al., 2020) with the following equation:
\begin{equation} HAS=\frac{\text{mean}_{\text{act}}-\text{mean}_{\text{base}}}{\text{std}_{\text{base}}}\nonumber \\ \end{equation}
where meanact denotes the average HBO concentration of a three seconds long window created around the peak HBO value found within the post-stimulus 2 and 8 seconds window;meanbase and stdbasedenote the average and standard deviation of two seconds long pre-stimulus baseline HBO concentration, respectively. As a result, each subject had a 22 x 2 (channel x condition ) matrix consisting of respective HAS values.
Two levels of analysis were performed on HAS parameter: within-condition and between-conditions. For within-condition analysis, a channel-based one-sample t-test was used to identify the brain regions showing significant hemodynamic activity for each condition, while a channel-based paired sample t-test was used to compare the brain activity between conditions. Channel-based analysis was preferred, as it was claimed that different brain regions have different optical properties which can lead to systematic bias and may cause non-comparable signal quality and/or intensity across different channels (Kujach et al., 2018; Yanagisawa et al., 2010). All statistical analyses were conducted in MATLAB R2017B.