2.1.

Participants

Two experiments (exp. 1 and exp. 2) were conducted according to the guidelines outlined within the declaration of Helsinki and were approved by the ATR Human Subject Review Committee and the Institutional Ethical Committee of Nagaoka University of Technology. Eighteen healthy, right-handed participants (P1 to 18, fourteen men, mean age: 37.6 years) took part in the experiment. In each experiment, data were collected from eleven participants (P1 to 11, eight men, mean age: 45.6 years for exp. 1; P8 to 18, 10 men, mean age: 24.5 years for exp. 2). Five participants (P7 to 11) took part in both exp. 1 and exp. 2. All participants provided written informed consent prior to participation in the experiments.

2.2.

Experimental Procedures

2.2.1.

Event-related design experiment (exp. 1)

We adopted an event-related design for exp. 1, which utilized a brisk ball-grasping task. Participants were asked to remain relaxed while seated in a chair and hold a plastic ball in either their left or right hand. They were asked to avoid movements not related to the task while gazing at a fixation point during the experiment. About 1 s after a cue beep (1 kHz) was heard from a personal computer, the participants were asked to quickly squeeze and release the plastic ball [Fig. 1(a)]. The cue beep was presented using a random interval of 5 to 15 s in length (mean: 10 s, uniformly distributed), so that the participants were not able to predict the cue timing. A single trial was defined as one grasping movement following the cue beep, with each session consisting of 20 trials. To detect the timing of the grasping event in a trial, we recorded ball pressure levels using a custom-made pressure sensor for five participants (P1 to 5). For the remaining participants (P6 to 11), we recorded muscle activity of the brachioradialis using an electromyography sensor (Trigno, Delsys Inc., Natick, Massachusetts) to detect the event. The event onset was defined as the time when the pressure level or muscle activity exceeded the mean of the value over the course of the session. A 10-s rest period was included at the beginning of each session for baseline correction. Five out of the 11 participants (P1 to 5) completed the experimental sessions over a 2-day period, performing 10 sessions/day for each hand. Thus, a total of 200 trials were performed for each hand. The remaining participants (P6 to 11) completed the experimental sessions with both the left and right hands in a single day, performing a total of five sessions for each hand (10 sessions/day). Thus, a total of 100 trials were performed for each hand. One participant (P7) completed only four sessions for each hand due to scalp pain. We assumed that the data obtained from the left- and right-hand tasks for each participant were independent and analyzed each as a separate sample, as assumed in previous studies.^6,8 Thus, a total of 22 samples (11 participants × 2 hands) were obtained for exp. 1.

Fig. 1

Experimental setting. (a) Task timing in exp. 1. Around 1 s after the beep, the participant was required to squeeze the ball one time. The intertrial interval was randomized from 5 to 15 s. (b) Probe positions. Sixteen sources (red circles) and 16 detectors (blue rectangles) were placed over bilateral sensorimotor areas with double density arrangements.³⁷ (c) Channel positions. Data were measured from 40 long channels and 12 short channels covering the bilateral sensorimotor areas. Due to the double-density arrangement, the distance between two long channels was 15 mm. (d) Channel positions projected onto the standard brain image. White circles indicate each long-channel position. The image was created using MRIcron software.

2.3.

Measurement of Functional Near-Infrared Spectroscopy Signals

We used a multichannel fNIRS system (FOIRE-3000, Shimadzu Corp., Kyoto, Japan; wavelengths: 780, 805, and 830 nm). The sampling period was 130 ms. The fNIRS probes (i.e., source and detector) were placed around Cz of the international 10 to 20 system to cover the sensorimotor cortex of both hemispheres [Fig. 1(b)]. This probe arrangement consisted of 40 standard 30-mm probe distance channels (long channels) and 12 shorter probe distance channels (15 mm, short channels) [Fig. 1(c)]. In order to improve the spatial resolution of the fNIRS measurement, we set the long channels at double the density³⁷ of the standard arrangement [Fig. 1(c)]. Following the fNIRS experiments, probe positions were measured using a stylus marker (FASTRAK, Polhemus, Colchester, Vermont). The measured signals were converted to relative changes in hemoglobin concentration ($\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$) according to a modified Beer–Lambert law.¹

2.4.

Analysis

3.1.

Estimated Superficial-Tissue Hemodynamic Artifacts in the Event-Related Motor Task

Using data obtained from short channels, we estimated global superficial-tissue hemodynamics for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ [Fig. 2(a)] in exp. 1. Increases in both hemoglobin signals were observed from the onset of movement, which peaked at $\sim 4$ to 5 s (average: 4.38 s; Table 1). When we conducted the same analysis for the block design experiment (exp. 2), the global superficial-tissue hemodynamics similarly increased from the onset of movement, peaking at around 5 s [Fig. 2(b)]. We then compared the superficial-tissue hemodynamic data obtained from exp. 1 (event-related design) with that obtained in exp. 2 (block design) and with the cerebral hemodynamic model (i.e., cerebral activity) in order to examine the characteristics of the estimated global superficial-tissue hemodynamics. Despite different experimental designs, the global superficial-tissue hemodynamics for both event-related and block design tasks exhibited similar task-related increases. Furthermore, both resembled the cerebral hemodynamics of the event-related design [Fig. 2(c)]. However, peak cerebral hemodynamic values in the block experiment were observed 11 s after the onset, while peak latencies for superficial-tissue hemodynamics in the event-related and block-design were $\sim 5\mathrm{s}$ after onset of movement (Table 1). Averaged global superficial-tissue hemodynamics was highly correlated with averaged cerebral hemodynamics in the event-related tasks, though low correlation was observed for the block design task ($r_{\text{cereb}-\text{superficial}}$; Table 1).

Fig. 2

Superficial-tissue hemodynamic signals extracted from short channels. (a) Average superficial-tissue hemodynamics across samples for the event-related experiment. The bold red line represents $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$, while the thin blue line represents $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$. Shaded areas represent the standard deviation for each sample. The data for 8 s after movement onset are shown due to the minimal rest period included at the end of each session (i.e., duration after the final motor task in the session) was 8 s. Amplitude (vertical axis) is arbitrary unit (a.u.) because the superficial-tissue hemodynamic signals were extracted after applying PCA to normalized ($z$-scored) short-channel data. (b) Average superficial-tissue hemodynamics across samples for the block design experiment. Bold magenta and thin cyan lines represent $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$, respectively. Shaded areas represent the standard deviation for each dataset. (c) Comparison of the cerebral- and superficial-tissue hemodynamics between the experiments. The bold and thin black lines represent the cerebral-hemodynamic models for the event-related and block design experiments, respectively. Note that each amplitude was normalized so that the maximum amplitude of the superficial-tissue hemodynamics for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ in each task was 1.

Table 1

Characteristics of the extracted superficial-tissue hemodynamic artifacts.

3.2.

Brain Activity with Superficial-Tissue Hemodynamic Artifacts

3.2.1.

Individual analysis (single-sample level)

Figure 3 shows an example of the event-triggered averages for measured estimated superficial-tissue and cerebral hemodynamics in a representative participant. Superficial-tissue hemodynamics [red line in Fig. 3(a)] was observed across measured channels while increases in cerebral hemodynamics of $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ were found in the contralateral (left) sensorimotor channels [black line, Fig. 3(a)]. Superficial-tissue hemodynamics of $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ was subtle [Fig. 3(b)]. Next, we evaluated cerebral activity for each sample using GLM analysis. Figure 4(a) shows data from a representative participant. Broad activation was detected without removal of superficial-tissue hemodynamic artifacts (RAW), especially for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ [Fig. 4(a), upper left]. Because strong activation is expected in the contralateral areas to the moving hand or fingers, our results suggest that false-positive activation occurred in the ipsilateral hemisphere. Using the GLM analysis that incorporated the removal of superficial-tissue hemodynamic artifacts (ShortPCA), activation in the ipsilateral hemisphere was suppressed, and the LI was increased for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ (e.g., LI was 0.44 for ShortPCA and 0.10 for RAW; Fig. 4(a), lower left). The LI for ShortPCA was significantly increased relative to that for RAW data [0.33 for ShortPCA and 0.27 for RAW; $t[21]=2.31$, $p<0.05$; Fig. 4(b)]. In contrast, results for $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ were nearly identical, and no significant LI difference was observed [Fig. 4(a), right]. The mean LIs obtained using ShortPCA for block design data (exp. 2) were 0.24 for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and $-0.20$ for $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$.

Fig. 3

Event-triggered averages. (a) $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and (b) $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ for a representative participant (P1, right-hand task) are shown. Gray thin lines represent measured (raw) data. Black bold lines show cerebral hemodynamics estimated by GLM analysis. Red bold and blue dashed bold lines indicate estimated superficial-tissue hemodynamics for $\mathrm{\Delta }\mathrm{Oxy}$- and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$, respectively. Dashed vertical line represents task onset. Note that the scale of the vertical axis is different between (a) and (b).

Fig. 4

Cerebral activation maps (t-maps) and laterality indices. (a) T-maps of RAW (top) and ShortPCA (bottom) analyses for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ (left) and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ (right) during the right-hand task of the representative participant (P1). Asterisks represent the channels for which statistically significant differences were observed ($p<0.05$). Values above the map denote the LI. (b) Average LI across samples for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ (red, top) and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ (blue, bottom). The error bar represents the standard deviation. The asterisk represents statistical significance ($p<0.05$). (c) Group-level t-maps for the right-hand task. (d) Group-level t-maps for the left-hand task.

4.1.

Characteristics of Superficial-Tissue Hemodynamics

Analyses of signals from short channels (Fig. 2) revealed several characteristics of superficial-tissue hemodynamics. We observed that superficial-tissue hemodynamics changed in a task-related manner, peaking at $\sim 5\mathrm{s}$ after the onset of movement (Table 1). This result aligns with those of a previous report, which found that an event-related motor task evoked transient increases in fNIRS signals in the visual cortex.²³ Our results indicate that such task-evoked increases are similarly observed in the sensorimotor areas.

Surprisingly, the temporal profiles of the superficial-tissue hemodynamics were closely matched in both the event-related and block design tasks. In the block design paradigm, cerebral activity is gradually increased and peaks at the end of the task duration [Fig. 2(c)] because the cerebral activity is assumed to be a linear function of task parameters.³⁹ Similar temporal changes are expected for superficial-tissue hemodynamics and cerebral activity in the block design task when superficial-tissue hemodynamics is also assumed to be dependent on task parameters (duration). Contrary to this expectation, in this study, the profile of superficial-tissue hemodynamics in the block design task was completely different from the cerebral hemodynamic profile and exhibited no relationship with task duration (i.e., no linearity). To examine whether this common transient increase in the systemic interferences is observed during other task designs, we collected fNIRS short-channel data during a motor task with different task durations (5-, 15- and 30-s grasping) and a cognitive task [verbal fluency task (VFT)⁴⁴]. We then analyzed superficial-tissue hemodynamics (Appendix A.1, data were partly presented previously⁶). Transient increases in superficial-tissue hemodynamics were observed across the analyzed datasets and peaked at $\sim 6\mathrm{s}$, even when the task duration or the type of task differed (Appendix A.1.3 and Fig. 5). Similar superficial-tissue hemodynamic profiles have been consistently reported in previous studies.^6,31 In these reports, superficial-tissue hemodynamics first peaked at 5 to 6 s, increasing again after several seconds, although this rebound was relatively small for the block task in this study [Fig. 2(b)]. Thus, comparisons between event-related and block designs suggest that, at least for motor tasks, superficial-tissue hemodynamics consists of two components: a transient increase following onset, irrespective of task design, and a subsequent gradual increase dependent on task design (duration). The first component may represent the general characteristics of the superficial-tissue hemodynamics, and this “cerebral-hemodynamic-like” change may produce broad, false-positive activation and obscure the actual cerebral hemodynamic profile in event-related tasks.

Fig. 5

Temporal profiles of superficial-tissue hemodynamics for different task designs. The data for (a) $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and (b) $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ are shown. Each line represents data from a different design. The data for exp. 1 (event-related design) and exp. 2 (block design) in Fig. 2 are also shown using black bold and red thin lines, respectively. For illustration purposes, the amplitudes were normalized so that the peak value of $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ from 0 to 10 s after task onset was 1.

However, the physiological origin of this transient superficial-tissue hemodynamic profile remains unclear. A number of authors have suggested that systemic changes in heart rate, respiration, and blood pressure are reflected in superficial-tissue hemodynamics.^5,22,25 Low-frequency oscillations (LFO) or Mayer waves are primary components of these signals, possibly reflecting changes in mean arterial blood pressures (MAP).²² Changes in global superficial-tissue hemodynamics observed in this study (Fig. 2) occurred in cycles of $\sim 10$ to 15 s (i.e., 0.06 to 0.1 Hz), which may reflect these LFO components. Thus, performance of motor tasks likely produce increases in MAP due to task-evoked vasodilation, following which such changes increase the magnitude or reset phases of LFOs, which eventually emerge in the superficial-tissue hemodynamics measured by short channels. Our results suggest that these changes in LFOs are independent of motor task parameters (e.g., duration, strength, and movement types). Such a notion is supported by the results of previous studies,^45,46 in which transient changes in arterial blood pressure (ABP) were observed during motor tasks, with first peaks occurring around 5 to 6 s after movement onset. Another study also revealed that cerebral blood volumes estimated from fNIRS hemoglobin signals ($\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ and $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$), which are strongly influenced by ABP, exhibited task-locked temporal profiles.⁴⁷ Further studies are required in order to more fully examine such changes in a variety of experimental settings.

4.2.

Methods of Superficial-Tissue Hemodynamic Artifact Removal for Event-Related Motor Tasks

The feasibility of superficial-tissue hemodynamic artifact removal has been confirmed for fNIRS data obtained during event-related motor tasks.¹⁸ This previous method combines the estimation of the hemodynamic response and the removal of superficial-tissue hemodynamic artifacts using 5-mm probe distance (reference) channels. Here, we used another simple method of artifact removal using short-channel signals (15 mm) and GLM analysis (ShortPCA⁶). The use of the 15-mm short channels, however, represents one limitation of this method, as the signals from these channels may contain a small amount of cerebral hemodynamic information.²⁶ However, the influence of the cerebral hemodynamics can be subtle in ShortPCA, as it extracts and removes superficial-tissue hemodynamics globally distributed over the brain using PCA, while the task-related cerebral hemodynamics may be observed locally.⁶ In addition, implementation and measurements of 15-mm short channels are easy to undertake (see Sec. 1). ShortPCA can thus be used for fNIRS measurements in broad areas, as previously indicated in studies utilizing block design tasks.⁶ We found that ShortPCA can be used to identify event-related cerebral activity associated with $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ during motor tasks, as indicated by an improvement in LI values [Fig. 4(b)]. Thus, consistent with the results obtained by Scarpa et al.,¹⁸ our findings support the feasibility of superficial-tissue hemodynamic artifact removal methods for fNIRS data obtained during event-related tasks. In addition, we evaluated group-level spatial maps of cerebral activity [Figs. 4(c) and 4(d)] and found that removal of superficial-tissue hemodynamic artifacts is effective, even for group-level analyses. However, the influence of superficial-tissue hemodynamics was reduced to some extent by averaging across samples even when no method of superficial-tissue hemodynamic artifact removal was applied [Figs. 4(c) and 4(d)]. This suggests that the individual differences in superficial-tissue hemodynamics may be large. It is possible that such differences are eliminated by group-level analyses using large numbers of samples.^13,16

4.3.

Limitations of this Study

While our results support the feasibility of ShortPCA for detecting $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$, this was not the case for $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$. It is possible that ShortPCA fails to extract superficial-tissue hemodynamics in $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ using short channels because the amplitudes of $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ are relatively small, with low SNRs. ShortPCA assumes that the superficial-tissue hemodynamics are predominantly contained within short-channel signals and thus can be extracted using PCA, which identifies major components in the signals. Therefore, if SNR is low, the predominance of superficial-tissue hemodynamics is decreased, and PCA may extract local signals or other noise components. Indeed, when we evaluated SNRs for each of the hemoglobin signal,³⁴ we confirmed that SNR for $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ was smaller than that for $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ (${\mathrm{SNR}}_{\mathrm{\Delta }\mathrm{Deoxy}}/{\mathrm{SNR}}_{\mathrm{\Delta }\mathrm{Oxy}}$ was 0.52, see Appendix A.2 for details), possibly because of individual differences in the sensitivity of $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$. As a result, the contribution ratio of the first component extracted by PCA in $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ was relatively small compared to that in $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$ (Table 1). Thus, ShortPCA may not be applicable when the SNR of the fNIRS signals is small.⁶ It is also possible that $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ does not contain superficial-tissue hemodynamic artifacts. A number of previous studies have reported that the effects of the artifact removal method are not observed for $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$,^{23,24,29,33,48,49} most likely due to the different origins (compartments) of $\mathrm{\Delta }\mathrm{Deoxy}\text{-}\mathrm{Hb}$ and $\mathrm{\Delta }\mathrm{Oxy}\text{-}\mathrm{Hb}$.

Among the superficial-tissue hemodynamic artifact removal methods proposed previously, we only tested ShortPCA for the removal of superficial-tissue hemodynamic artifacts, as this method is easily applicable to measurements of broad brain regions.⁶ However, this method has several limitations. First, as discussed above, usage of 15-mm short channels may have a small risk of contamination by cerebral activity when one estimates superficial-tissue hemodynamics. Second, the method cannot reduce the effects of local systemic components of superficial-tissue hemodynamics observed at some specific channels,^24,49–51 while it estimates and removes the global components. In addition, this method requires short-channel measurements and cannot be applied for data measured previously without short channels. Considering the cost of the short-channel measurements and their applicability, it may be better to use a method using long channels only. However, one should take care when applying the method using long channels, as it may have a risk of removing actual cerebral activity due to the similar temporal profiles between the cerebral and systemic components (Fig. 2).^23,24 Owing to its similarity, this could be problematic even when using short channels. As proposed by Zhang et al.,³³ taking the spatial natures of the systemic interferences into account may overcome this limitation. Our method also uses spatial information of the superficial-tissue hemodynamics by using multiple short channels located across measurement areas; this may reduce the risk of removing cerebral components to some extent.

The use of the long channels and the previously proposed systemic artifact removal methods using short channels may also be applicable to event-related designs, especially when using real-time adaptive algorithms.^30,35,36 Although comparisons of the existing artifact removal methods were beyond the scope of our study, it is of interest to conduct further research aimed at comparing the different methods, especially using simultaneous recordings of fNIRS and fMRI.⁴⁸ This may reveal the most effective and accurate way to investigate changes in cerebral activity during event-related motor tasks.