Qualitative and Comparative Cortical Activity Data Analyses from a Functional Near-Infrared Spectroscopy Experiment Applying Block Design

Summary

We describe the analysis of continuous-wave functional near-infrared spectroscopy experiment using a block design with a sensorimotor task. To increase the reliability of the data analysis, we used the qualitative general linear model-based statistical parametric mapping and the comparative hierarchical mixed models for multi-channels.

Abstract

Neuroimaging studies play a pivotal role in the evaluation of pre- vs. post-interventional neurological conditions such as in rehabilitation and surgical treatment. Among the many neuroimaging technologies used to measure brain activity, functional near-infrared spectroscopy (fNIRS) enables the evaluation of dynamic cortical activities by measuring the local hemoglobin levels similar to functional magnetic resonance imaging (fMRI). Also, due to lesser physical restriction in fNIRS, multiple variants of sensorimotor tasks can be evaluated. Many laboratories have developed several methods for fNIRS data analysis; however, despite the fact that the general principles are the same, there is no universally standardized method. Here, we present the qualitative and comparative analytic methods of data obtained from a multi-channel fNIRS experiment using a block design. For qualitative analysis, we used a software for NIRS as a mass-univariate approach based on the generalized linear model. The NIRS-SPM analysis shows qualitative results for each session by visualizing the activated area during the task. In addition, the non-invasive three-dimensional digitizer can be used to estimate the fNIRS channel locations relative to the brain. To corroborate the NIRS-SPM findings, the amplitude of the changes in hemoglobin levels induced by the sensorimotor task can be statistically analyzed by comparing the data obtained from two different sessions (before and after intervention) of the same study subject using a multi-channel hierarchical mixed model. Our methods can be used to measure the pre- vs. post-intervention analysis in a variety of neurological disorders such as movement disorders, cerebrovascular diseases, and neuropsychiatric disorders.

Introduction

Neurorehabilitation plays an important role in the functional recovery following sensorimotor disturbance. To clarify the mechanisms of neuroplasticity-associated functional recovery, various neuroimaging technologies have been used, such as functional magnetic resonance imaging (fMRI), positron emission tomography (PET), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS). Different imaging modalities have different advantages and disadvantages. Although the fMRI is the most typical device, it is affected by magnetic fields, has a high cost, high physical restriction, and limited sensorimotor tasks^1,2,3,4. The fNIRS device stands out as a noninvasive optical neuroimaging and has a relatively lower spatial resolution, but it has a better temporal resolution than fMRI⁴. fNIRS is suitable when verifying treatment effects because it compares the pre- versus post-intervention effects, has dynamic motor tasks, is portable, and functions more in natural environments than fMRI^1,2,4. NIRS has been reported to be more suitable in the fields of cerebrovascular disease, epileptic disorders, severe brain injury, Parkinson's disease, and cognitive impairment^1,5. With regard to sensorimotor tasks, it is widely used in gait and standing balance^6,7,8, upper limb function (hand grasping, finger tapping)^8,9, complex motor skill training^10,11, robotics^12,13,14,15, and brain-computer interface^16,17,18. The fNIRS is based on the principles of optical neuroimaging and neurovascular coupling, which measure cortical metabolic activity, increased blood flow, and consequently cortical activity as secondary signals¹⁹. fNIRS signals have been reported to have strong correlations with signals of blood oxygen level-dependent fMRI²⁰. A continuous-wave fNIRS uses the modified Beer-Lambert law to determine the changes in oxygenated hemoglobin (HbO₂) and deoxygenated hemoglobin (HHb) cortical concentration levels based on measured changes in broadband near-infrared light attenuation^21,22. Because it was not possible to measure the differential path-length factor (DPF) using the continuous-wave NIRS system, we assumed that the DPF was constant and that hemoglobin signal changes were denoted in arbitrary units of millimole-millimeter (mM x mm)^2,18.

The fNIRS experiments need to select the most adequate methods including the probe settings, the experiment designs, and the analysis methods. Regarding the probe setting, the international 10-20 method used in EEG measurement is the setting standard used by many researchers in neuroimaging. In recent years, coordinate settings based on the standard brain on the basis of Montreal Neurological Institute (MNI) coordinates have been used. The experiment uses a block design, generally used for sensorimotor tasks, and an event-related design. This is a method of comparing changes in hemoglobin concentration at rest and during tasks; HbO₂ concentration levels increase and HHb concentration levels decrease with changes in cerebral blood flow associated with task-dependent cortical activity. Although there are various analysis methods, the NIRS-SPM free software enables an analysis similar to the statistical parametric mapping (SPM) of fMRI. The treatment of NIRS data uses a mass-univariate approach based on the general linear model (GLM). When performing task-dependent brain activity analysis, the fNIRS measurements can be affected by evoked or non-evoked neuronal activity and systemic physiological interferences (heart rate, blood pressure, breathing rate, and autonomic nervous system activity) in the cerebral and extracerebral compartment²³. Therefore, pre-analysis processing, filtering, wavelet conversion, and principal component analysis are useful²³. Regarding filtering and artifacts of the data processing using the NIRS-SPM, low-pass filtering⁹ and the wavelet minimum description length (Wavelet-MDL)²⁴ detrending were used to overcome the motion or other sources of noise/artifact. For details of this analytic method, refer to the report of Ye et al.²⁵. Although there are reports using only SPM, it is only a qualitative index by image analysis, and due to the low spatial resolution of NIRS, extreme caution is required for group analysis. Moreover, when the DPF is constant, numerical comparisons between channels and individuals should not be performed, but the difference in the changes in each channel can be verified. Based on the above conditions, in order to supplement the NIRS-SPM group analysis results, we used the original analysis method for multi-channel analysis after improving the accuracy of spatial registration. This multi-channel analysis compared the amplitude of the change in HbO₂ and HHb levels between the rest and on-task periods at each channel before and immediately after treatment using hierarchical mixed models with fixed interventions (before or after), fixed periods (rest or on-task), and random individual effects.

In this way, there are several fNIRS measurement and analysis methods; however, no standard method has been established. In this paper, we introduce our methods, qualitative GLM-based statistical parametric mapping and the comparative multi-level hierarchical mixed model, to analyze data obtained from a multi-channel fNIRS experiment of pre- vs. post-intervention using a block design with sensorimotor tasks.

Access restricted. Please log in or start a trial to view this content.

Protocol

This study was approved by the institutional review board (IRB) of the Fukuoka University, Japan (IRB No. 2017M017). Prior to participation, all patients provided written informed consent.

1. Preparation of the fNIRS experiment

NOTE: A multi-channel continuous-wave laser-based NIRS system for this experiment was used. The wavelengths of the near-infrared light were 780 nm, 805 nm, and 830 nm, and the sampling rate was set at 7.8 Hz. The time and spatial resolution (distances between the light emitter and detector probe) were 0.13 s and 3.0 cm, respectively.

1. Set the fNIRS device in a dark noiseless place. Conduct experiments at room temperature. Start the fNIRS instrument 30 min before the experiment.
   NOTE: fNIRS systems are used under controlled temperature and humidity conditions to ensure stable operation of the equipment²⁶.
2. Use a whole head caps for fNIRS recording and attach the head cap on the subject's head such that the position corresponding to the central (Cz) of international 10-20 system is located at the holder No. 245 of the head cap. (Figure 1).
3. Attach the marking sticker to reference locational points: the nasion (Nz), right external auditory meatus (AR), and left external auditory meatus (AL).
   NOTE: Because the three-dimensional (3D) coordinates are read around the positions of the Nz, AR, AL marking sticker and Cz holder, it is necessary to attach the marking sticker before taking a picture.
4. After the calibration of a high-resolution digital camera for spatial registration, take pictures of the subject's head with the probe location while showing the reference points (Cz, Nz, AR, and LR) from 15 perspectives.
   NOTE: Please take a picture before placing the probe. If a picture is taken after placing the probe, the marker landmark may be hidden by the probe and wiring cord. As recommended by the manufacturer, after taking 12 pictures 30° diagonally forward to the right of the subject, take three or more pictures slightly above so that the Cz (holder No. 245) appears in the picture. This is because it is easy to make it three-dimensional when capturing a total of 15 or more shots.
5. Carefully separate the subject's hair that interferes with the optode using a light-emitting diode (LED)-lit plastic rod to attach the probe. Arrange the probe so that the optodes are attached at a minimal distance from the scalp surface and in contact with the scalp.
   NOTE: Check carefully whether there is any pressure or discomfort for the patient due to the attachment of the optodes, because of increasing strength of the systemic confounders associated with autonomic nervous system activation²³.
6. Arrange the 48-channel system with 32 optodes (16 light sources and 16 detectors; 4 x 4 array for each hemisphere) to a head cap bilaterally over the frontal and parietal areas as regions of interest (Figure 2).
7. Start and use the 3D-digitizer software to determine the spatial registration.
8. After scanning the picture data of an entire head, determine the spatial coordinate of each patient by auto-measuring and save as the Origin and the Others file (*CSV file).
   ​NOTE: If the coordinate points could not be detected from the images using the automatic measurement, input the adjustment manually.

2. Run the experiment

1. Select a block design for the experiment, and the task can be any movement of interest for the study such as hand opening/closing, finger tapping, etc. In our previous study, the task was the robot-assisted elbow movements¹⁵.
   NOTE: Each cycle is made of three blocks (15 s of rest - 15 s of task - 15 s of rest), and each patient completes seven cycles in each session.
2. Make the participant wait in a comfortable position until the starting signal. Instruct the subject to close their eyes during the rest and the task.
3. Give start and stop cues (i.e., "Repeat flexion and extension of the elbow", "Stop and relax").
   NOTE: Do not speak during measurements. Carefully check for artifacts on the monitor screen during measurements.
4. Perform the block design task in the same posture. Upright posture with standing or sitting position is desirable not to distort the headset.
   NOTE: If the patient feels uncomfortable after wearing the head holder for an extended period of time, remove or loosen the probe during robot-assisted exercise.
5. After completing the NIRS measurement, remove the head holder and marking sticker to end the experiment.
   ​NOTE: Carefully check for skin damage to the scalp due to prolonged wearing.

3. Qualitative GLM analysis using NIRS-SPM software

1. Start the NIRS-SPM on the MATLAB software. Convert the data file related to the change in HbO₂ and HHb concentration acquired from the NIRS device to the file format for NIRS-SPM analysis.
2. Choose the using NIRS system option from the pop-up menu. Select the load button and choose the convert HbO₂ and HHb concentration change options.
3. Detect the spatial registration of NIRS channel location. Select the Stand-alone checkbox and then select the With 3D Digitizer checkbox.
4. Within From Real Coordinates to MNI Space, use the dialog box to choose _origin.CSV referring to the coordinate reference point file, and _others. CSV referring to the coordinate probes/channels file.
5. Select the Registration button. Choose the points to proceed to spatial estimation, and click on the OK button. Click on the Project MNI coordinate to Rendered Brain button.
   NOTE: The spatial position of the NIRS channel locations is estimated on the basis of Montreal Neurological Institute (MNI) brain template.
6. Select the Dorsal View option, and click on the Save button.
7. In the Specify the 1^st Level section, select the NIRS data filename and SPM directory. Select the hemoglobin checkbox; HbO₂ or HHb. Highlight the Specify design' option and select the Sec option. Highlight the Number of Condition/Trials options and enter the number 7.
8. Highlight the Vector of Onset and Duration[s] options and enter a vector of onset multiplied by the duration of the experimental conditions as follows.
   NOTE: In this case, the vector of onset times should be specified as [15:45:285] or [15 60 105 150 195 240 285]. The vector of duration should be specified as [15* ones(7,1)] or [15 15 15 15 15 15 15].
9. For detrending, select the Wavelet-MDL button. Use the precoloring method: low-pass filter and select the hrf button, and correct for serial correlation, then select the none button.
10. In estimating the temporal correlations, check the Individual Analysis.
    NOTE: Save the spatial localization of fNIRS channel positions in the individual MNI coordinate systems as a text file. Similarly, save the map based on the individual Brodmann area (BA) as a text file.
11. In estimating the temporal correlations, check the Group Analysis. The NIRS-SPM aligned the mean optode positions of the number of participants according to the MNI standardized brain coordinate system.
12. Compute the activation map based on the changes in hemoglobin level for the standardized brain. HbO₂ and HHb levels were considered significant at an uncorrected threshold of p < 0.01.
    ​NOTE: Left/right information was flipped in the right-affected sides for group analysis.

4. Multi-channel comparative analysis based on hierarchical mixed model

1. Start the SAS software. Convert the text document (.TXT) of concentration changes of HbO₂ and HHb in the NIRS data file processed with a low pass filter (cut-off frequency was set at 0.1 Hz) to the spreadsheet software comma separated values file (.CSV).
2. Create the Import SAS data (.sas7bdat) using the program.
3. Output the Import file with the following command, libname out "Import file"
4. Output the pre- vs. post-intervention file for each subject, run the following commands in the Analysis SAS. (Figure 3) During the creation of the import file, give a name such that can identify the subject information and pre- vs. post-intervention (e.g., id1 pre, id1 post…)
5. Run the pre- and post-intervention data command for each channel (ch1-48; HbO₂ and HHb) as follows (Figure 4).
6. Based on the data obtained from the output results, input the pre- vs. post-intervention differences in change (difference at on-task and rest), rest, and on-task values (estimated values, upper limits, and lower limits) of each channel in the spreadsheet workbook file (.xlsx).
7. Similarly, input the numerator and denominator degrees of freedom, F value, and P value of the interact item of the fixed-effect Type 3 test to the spreadsheet workbook file (.xlsx).
8. To control the false discovery rate (FDR) in multi-channel testing, use the Benjamin and Hochberg methods²⁷ and control the FDR at a p-value < 0.01.

Access restricted. Please log in or start a trial to view this content.

Results

Herein, we introduce the robot-assisted rehabilitation that our group is currently working on: the biofeedback effects on upper limb motor deficit in patients with acute stroke. We included 10 consenting stroke patients (mean age: 66.8 ± 12.0 years; two women and eight men) who were admitted to our hospital. At the subacute stroke stage, more than 2 weeks after the onset, we evaluated the motor-related cortical activity of these patients using an fNIRS system before and immediately a...

Access restricted. Please log in or start a trial to view this content.

Discussion

In our group analytic methods for fNIRS, in addition to performing an imaging analytic method by qualitative t-statistic mappings, we compared pre- vs. post-intervention (robot-assisted exercise) using the comparative multichannel analysis. For qualitative analysis, we used the NIRS-SPM software as a mass-univariate approach based on the generalized linear model. The NIRS-SPM analysis shows qualitative results of each session by visualizing the activated area during the task. Moreover, the information of the non...

Access restricted. Please log in or start a trial to view this content.

Materials

Name	Company	Catalog Number	Comments
3D-digitizer software	TOPCON	-	NS-1000 software ver.1.50
NIRS system	Shimadzu	-	FOIRE-3000
Robot	CYBERDYNE	-	Single-joint type Hybrid Assitive Limb (HAL-SJ)