4.1.1.

Preparation of temporal signals (hemoglobin concentrations)

FNIRS optical time course data are obtained from diverse commercial devices in different types or formats. Some devices only output the raw optical intensity data, whereas others support the calculation of the relative hemoglobin concentration change with in-house conversion functions. Different fNIRS recording systems use different output formats (such as .txt and .csv). These inconsistent data sources pose difficulty for following-up analysis, thus a unified format for hemoglobin concentration is need.

NIRS-KIT provides time course preparation functionality for diverse fNIRS data sources (Fig. 3). If the raw optical intensity data are obtained, such as from Hitachi ETG4000/7000 (.csv), or NIRX (.wl1 and .wl2), they can first be converted into optical density data and then converted to concentration changes of HbO and HbR via the modified Beer–Lambert law.⁴¹ The converted hemoglobin concentration data will be saved in the NIRS-KIT data format. If the converted hemoglobin concentration change data are directly obtained, such as from Hitachi ETG4000/7000 (.csv) or Shimadzu LABNIRS (.txt), they will be reformed and saved in the NIRS-KIT supported format.

Fig. 3

User interface for data preparation module with supported data sources.

If the recording system is not one of the above, NIRS-KIT also provides a manual input function to generate the required data format. Here users just need to reorganize their raw data (optical density data or hemoglobin concentration data) into a specific format (.csv) according to the format in the sample file included in the toolbox.

NIRS-KIT has a good compatibility with two widely used fNIRS data analysis packages. Both raw and preprocessed fNIRS time course data from NIRS-SPM (.mat) or HomER2 (.nirs) can be read and saved in the NIRS-KIT supported data format.

Recently, the Shared Near-Infrared Spectroscopy Format (SNIRF), a new fNIRS data format standard, has been proposed by the fNIRS community recently (see Ref. 26), and adopted by Homer3 and FieldTrip. NIRS-KIT also supports SNIRF as input.

4.1.2.

Preparation of topospatial information (probe setup)

The probe setup, consisting of the configuration of the sources, detectors, and measurement channels, provides spatial information that needs to be included in the data file. The topographical geometry of the channels is very useful for checking time course signals for problems and result visualization, especially when adopting a complex or irregular probe design. NIRS-KIT provides several standard probe settings in the software package’s sample folder (including standard $3\times 3$, $3\times 5$, $3\times 5\times 2$, $3\times 11$, and $4\times 4$ probes). If an appropriate probe setup file does not already exist, users can use the topomaker module (see Fig. 3) to generate, in a simple and flexible way, a customized probe setup file with arbitrary arrangements of sources and detectors.

4.1.3.

Preparation for spatial information in standard brain space

Three-dimensional information in standard brain space, consisting of the spatial locations of optodes and channels, is important for result interpretation, visualization in brain space, and publication so as to enable replication and meta-analysis. NIRS-KIT allows spatial registration of NIRS optodes or channels to standard MNI space with the NFRI toolbox developed by Singh et al. (2005).⁴² Two text files for each participant are necessary. One contains fiducial (landmark) coordinates in real space (usually obtained using a 3D digitizer), in which at least four reference positions of the international 10–20 system⁴³ are given. The reference positions should be spatially separated, and the combination of five standard positions—Iz (inion), Nz (nasion), AL (left preauricular point), AR (right preauricular point), and Cz (central zero)—are commonly used with good effect. The other text file contains the optode and channel coordinates in real space. After loading the above files, a text file containing MNI coordinates of each channel and optode in standard brain space will be outputted. In addition, NIRS-KIT also reports the structural labels of the brain regions closest to the channel locations in three brain atlases (Brodmann, AAL, and LPBA40 atlases).

4.3.1.

Time point trimming

The first few time points of recordings sometimes need to be discarded due to participants’ process of adaptation to the experimental environment. If this option is selected, NIRS-KIT will delete the specified initial and/or last time periods from the data of each participant.

4.3.2.

Detrending

FNIRS signals may suffer from systematic increases or decreases over time due to long-term physiological shifts or instrumental instability. NIRS-KIT uses a polynomial regression model to estimate a linear or non-linear trend and subtracts it from the raw hemoglobin concentration signal [as illustrated in Fig. 5(b)]. The default order of the polynomial model is first order to remove linear detrends from the raw time course.

4.3.3.

Motion correction

Two parameter-free motion correction methods are available in NIRS-KIT. One is correlation-based signal improvement (CBSI),⁴⁴ which mainly relies on the assumption that the neural components of HbO and HbR signals are strongly negatively correlated, whereas motion artifacts are positively correlated. The other new method is temporal derivative distribution repair (TDDR),⁴⁵ which is based on robust regression and can effectively remove both spike artifacts and baseline shifts [Fig. 5(c)].

4.3.4.

Filtering

The signal of interest usually occupies a frequency range different from interference, such as machinery and physiological noise (e.g., respiration, heart rate, Mayer wave, and very low-frequency oscillations). NIRS-KIT provides three filtering models (high-pass, low-pass, and bandpass filtering) to remove irrelevant low-frequency and/or high-frequency components [as illustrated in Fig. 5(d)]. Three commonly used digital filter types are available: infinite impulse response (IIR) filter, finite impulse response (FIR) filter, and fast Fourier transform-based ideal filter (FFT-based filter). The IIR filter used in NIRS-KIT is a Butterworth filter (default third order). The FIR filter is a Hamming window filter (default 34th order). For the FFT-based filter, time series of each channel are transformed into frequency domain, the frequency-domain signals are filtered and then transformed back to time domain. The default is bandpass filtering (0.01 to 0.08 Hz) using a third-order IIR filter. The filtering model, high- and/or low-frequency thresholds and filter type can be specified by the user according to study objectives and noise characteristics. For resting-state, low-frequency fluctuations (0.01 to 0.08 Hz) have been suggested to be of physiological importance and may reflect spontaneous neural activity.^46,47 Note, when performing fALFF analysis, the bandpass filtering should be applied to reserve the entire frequency band of the neural signals,⁴⁸ and frequency range selection can refer to fALFF analysis in fMRI studies (usually 0 to 0.25 Hz).

4.3.5.

Noise regression

Undesirable systemic artifacts (especially scalp blood flow) usually contaminate fNIRS functional signals, and how to minimize these sources of interference are still a challenge for fNIRS signal processing. A prominent approach is to use short-distance reference channels (that are sensitive only to signals in the superficial layers of the tissue outside the brain) to record superficial noise,^49–53 then use regression to remove that noise from neural recordings. When short-distance channels are used, NIRS-KIT provides a noise regression functionality, in which the signals of short-distance channels can be used as the regressor and be removed. In some cases, however, short-distance channels are not available. In these cases, another promising signal separation method proposed by Yamada et al. (2012)⁵⁴ can be used to separate functional and systemic signals based on their hemodynamic differences. This method has been incorporated into the preprocessing module to remove the unwanted systemic noise (also see Sec. 4.3.7).

4.3.6.

Resampling

For large scale, multi-site, or hyper-scanning studies, it is sometimes necessary to resample the fNIRS time series data recorded with different recording systems (with different sampling rates) into one sampling rate prior to subsequent processing. Also raw signals sometimes need to be downsampled for computational and memory efficiency. For these reasons, users can select the resample option, set the desired sampling rate, and NIRS-KIT will resample the time series data.

4.3.7.

Customized processing methods

In fNIRS signal preprocessing, users might want to use their own processing methods which are adapted to specific needs. Also fNIRS signal processing methodology is continuously being improved by researchers around the world so that novel and modified processing methods are periodically published. If these methods are not included or have not been updated in NIRS-KIT, users can write customized processing methods as a MATLAB functions, and easily incorporate them into NIRS-KIT via the customized interface [see Fig. 5(a)]. Users can refer to the sample customized processing method file in the package [NK_SignalSperation_Yamada.m, a systemized noise remove method developed by Yamada et al. (2012)⁵⁴ was implemented as an example] to make their own customized function or script.

4.4.1.

Functional connectivity

Resting-state FC measures temporal correlations in spontaneous fluctuations among spatially distributed brain regions^55,56 and has been increasingly used for fNIRS.^17,57,58 NIRS-KIT supports fast ROI2ROI, ROI2Whole-brain, and whole brain (channel-wise) FC analyses.

4.4.2.

Network metrics

Network metrics capture complex topological properties of brain networks, such as local and global efficiency^59,60 and modularity.⁶¹ NIRS-KIT allows definition of the network of interest and then calculates the FC matrix of the desired network and saves the result in the file format of Gretna,⁶² a widely used MATLAB-based toolbox for calculating graph-theoretic metrics. Prior to network topological analysis, a thresholding procedure is typically applied to exclude the confounding effects of spurious relationships in interregional connectivity matrices. Two commonly applied thresholding techniques are provided in Gretna: the absolute connectivity strength threshold and relative sparsity threshold (or proportional threshold).^9,62–64 For the former, users need to set a threshold value and the connections with weight greater than the given threshold are retained. For the latter, the threshold value is defined as the ratio of the number of actual edges divided by the maximum possible number of edges in a network. Note that the absolute threshold method may lead to different numbers of network edges across different datasets, which may confound between-group comparisons.⁶⁵

4.4.3.

ALFF and fALFF

ALFF and fALFF, two common resting-state indices characterizing regional spontaneous brain activity,^28,29 were not available in any existing fNIRS analysis software. We implemented this functionality in NIRS-KIT [Fig. 6(b)]. The time series signal can be approximated by a finite sum of weighted cosine and sine functions whose frequencies ($f_k$) depend on the sampling frequency ($f_s$) and the number of collected time points ($N$):

Fig. 7

The diagram for computing individual ALFF and fALFF.

4.6.1.

Channel-wise visualization

In ALFF/fALFF and ROI2WholeBrain FC, each channel has a single scalar representing the group-level statistical value or individual-level index. Hence we use channel-wise visualization to display these results. For channel-wise 2D visualization, the probe setup (see Sec. 4.1) is required. A blank canvas with channel locations is generated according to the provided probe geometry. Then color values are mapped with interpolation [Fig. 9(a)] or without [Fig. 9(b)]. For ROI2Whole-brain FC, non-interpolated topographical visualization is recommended, and the toolbox will automatically identify the ROI channels and mark them with red circles [Fig. 9(b)]. For channel-wise 3D visualization, the MNI coordinates of each channel are needed for 3D visualization in standard brain space. In NIRS-KIT, three non-interpolated and one interpolated channel-wise 3D visualizations are provided. The first non-interpolated 3D visualization is mapping the channel-wise statistical values on to the standard brain space using the nfri_mni_plot function in the NFRI toolbox.^42,69,70 The second is to project the analysis result of each measurement channel directly on to a standard surface template (e.g., ICBM152), set the display parameters (such as, the sphere size and colorbar limits), and then output the resulting figure. The last, more flexible approach, is to output a brain image in the widely used NIFTI format. Then users can load and plot this image with several other brain visualization toolboxes, such as MRIcroGL⁷¹ and Surfice.⁷² For interpolated channel-wise 3D visualization, the EasyTopo toolbox provided by Tian et al. (2013)⁷³ is used. Here the stereotaxic coordinates of the brain surface and fNIRS channels are converted into spherical coordinates, where 2D angular interpolation of the channel-wise data is implemented to obtain a topographic image of brain activation in the latitude–longitude space. Then the interpolated image is projected back onto the original brain surface. The first non-interpolated 3D visualization using the NFRI toolbox was illustrated in Fig. 9(c). The other three visualization types will be shown in Sec. 5.6 for displaying the task fNIRS analysis results.

Fig. 9

Examples of visualization for resting-state analysis results: (a) group-level zfALFF map (interpolation mode); (b) ROI2Whole FC map (non-interpolation mode, and channel 12 is the seed channel); (c) non-interpolated channel-wise 3D visualization for group-level zfALFF result using the NFRI toolbox; (d) 2D whole brain FC matrix map; and (e) 3D visualization of whole brain connectivity matrix by BrainNet Viewer using the node and edge files generated by NIRS-KIT. Note: All above color bars represent group-level ($n=9$) one-sample $t$-test statistical values, and the color axis limits are set for illustration only. In A and B panels, only the left hemisphere probe (channels 1 to 22) is displayed to reduce clutter.

4.6.2.

Connectivity matrix visualization

The result of whole brain FC is saved as an $n\times n$ matrix ($n$ is the number of channels), where each value in the matrix represents the FC between a paired channels or group-level statistic’s value of each FC. For 2D visualization of whole brain FC matrix, users can choose to only display the significant edges by setting a threshold $p$-value, define the displayed subnetworks, and arbitrarily reorder the channels [Fig. 9(d)]. For 3D visualization of whole brain FC matrix, an Excel (Microsoft Corp., Redmond, WA, USA) file containing MNI coordinates of each channel (see sample files in the package) is need to be loaded, then NIRS-KIT will output a node file and an edge file that can be loaded and visualized by the most existing brain connectome visualization toolkits [Fig. 9(e)], such as BrainNet Viewer.⁷⁴

Several visualization options are available, such as setting display modes, color bar limits, and $p$-value thresholds.