2.10. NIRS data processing

Data analysis was performed using MATLAB 2017a (MathWorks, Natick, MA). In this study, we focused on the pre‐ and postwalking standing rest periods where data were not contaminated by motion. This distinction between motion and rest periods was made using accelerometer data (Figure 3a,b). Saturated channels were removed from the analysis and the raw NIRS data of the remaining channels were normalized to their means (Fabiani et al., 2014). Intensity data were filtered with a bandpass filter with cutoff frequencies at 0.5–5 Hz (Tan et al., 2017) to preserve brain physiology data while eliminating unwanted high‐ and low‐frequency noises. Motion artifacts were identified using accelerometer data and movement artifact functions in the HOMER2 toolbox (Huppert, Diamond, Franceschini, & Boas, 2009). In this study, data were inspected visually, channel by channel, to ensure movement artifacts were removed from the analysis. Next, we followed the approach in Pollonini et al. (2014) as a quality check for heartbeat epochs in one channel. In short, a good source detector coupling presents a prominent synchronized cardiac oscillation in both wavelengths. Hence, after preserving the cardiac component in both wavelengths, a cross‐correlation at time lag zero showed how well two wavelengths were coupled. The resulting number is called scalp coupling index (SCI) and served as a quality check for each channel. In this study, only channels with an SCI ≥0.8 were considered for further analysis.

(a) Example of accelerometer data averaged over three axes including standing rest before short‐duration walking, walkig,and standing rest after short‐duration walking. (b) Example of the NIRS intensity data for wavelength 850 nm. A higher pulse amplitude in the standing rest before short‐duration walking in comparison with after short‐duration walking is clearly identifiable. NIRS, near‐infrared spectroscopy

Heartbeat pulse epochs were analyzed for each resting period, BW and AW, separately. Local maxima (peaks) and local minima (nadirs) were determined with a semi‐automatic approach using the “FindPeak” function in MATLAB 2018a (MathWorks, Natick, MA). Each heartbeat epoch was tagged based on defining a local extremum. However, if a peak or nadir was flat, the individual heartbeat epoch was discarded from further analysis. Nevertheless, in some channels, we observed heartbeat waveform variability. Therefore, manual and visual inspection were performed on all the channels for each participant to ensure that any misidentifications or detection errors of the peaks and nadirs were fixed manually. Manual fixes were effected by adjusting the “FindPeak” algorithm parameters or discarding the heartbeat epoch as an artifact from further analysis. Following identification of peaks and nadirs, each individual heartbeat was separated. The baseline shift of each heartbeat epoch was removed by piecewise cubic spline interpolation.

The waveforms with cerebral pulse amplitude with standard deviation twice greater than the mean were considered motion artifacts and were removed from the analysis. The average waveform of the remaining heartbeats was determined and a correlation between each individual and the average heartbeat epoch was calculated. The heartbeat epochs with a correlation ≥0.8 were kept. Channel data were divided into four quantiles and the second and third quantiles were averaged (median). In each heartbeat epoch obtained from the NIRS signal, cerebral pulse amplitude and PRF were calculated separately. Next, the average of the pulsatility indices was used to represent the pulsatility index for that channel. In addition, the average across all the channels was used as the global pulsatility index in each participant. Figure 4 provides a diagram of NIRS data processing to extract pulsatility indices. For each participant, we averaged the cerebral pulse amplitude across all channels as an indicator of the global cerebral pulse amplitude for the participants. The same procedure was followed for global PRF.

Schematic diagram of NIRS data processing to extract pulsatility indices and analysis. ANOVA, analysis of variance; HB, heartbeat epoch; M (SD), mean ±standard deviation; NIRS, near‐infrared spectroscopy; PRF, pulse relaxation function; SCI, scalp coupling index

A one‐way analysis of variance (ANOVA) using group as between subject factor (LCVRF, HCVRF, and CAD) was performed on the data. The aim of the analysis was to evaluate whether at standing rest, cardiovascular degradation from LCVRF to HCVRF and CAD could alter cerebral pulse amplitude and PRF. In this model, the dependent variables were cerebral pulse amplitude and PRF. Finally, to explore the impact of walking on alterations of global cerebral pulse amplitude in each group individually, we employed a paired student t‐test within each group. All statistical analysis was performed with SPSS (IBM statistics for Macintosh, Version 25.0).

In addition to the above analysis assessing the global pulsatility index combining all NIRS channels for a single participant, we conducted channel‐wise analysis on pulsatility parameters to explore their spatial properties BW and AW. A two‐sample t‐test was then conducted on the pulsatility scores across the groups, testing the null hypothesis that there was no statistically significant difference in the measured cerebral pulse amplitude or PRF across the three groups for this channel. This yielded a total of 38 NIRS channels to be included in the channel‐wise analysis (Figure 5). The t‐values of each comparison were then corrected for multiple comparison with a false discovery rate (FDR) approach (Benjamini & Hochberg, 1995). Along with the raw pulsatility scores, the t‐values that survived the FDR correction were then projected on the cortical layer of the Colin27 template based on the anatomical locations of the corresponding channels. This projection was performed using the open‐source toolbox Atlas Viewer (Aasted et al., 2015). These individual values were then spatially interpolated to generate cortical contrast maps. Channel‐wise analysis was performed with Atlas Viewer, HOMER2 (Huppert et al., 2009) and some in‐house scripts.

Spatial distribution of NIRS channels included in the channel‐wise analysis of pulsatility parameters and the corresponding cortical sensitivity matrix, frontal view. Red numbers indicate near‐infrared light sources and blue numbers indicate near‐infrared light detectors. Yellow lines connecting sources to detectors represent NIRS channels. This figure shows that the 38 channels included in the channels‐wise analysis had reasonable sampling sensitivity in most of the prefrontal areas and in bilateral supplementary motor cortices. NIRS, near‐infrared spectroscopy