2.1.

Subjects

Sixteen healthy volunteers (right-handed, native Chinese speakers, age $21–23\text{years}$ old, 6 females and 10 males) participated in this experiment. All subjects had normal or corrected-to-normal vision and normal color vision. No subject had a history of neurological, medical, or psychiatric disorders, and none was taking medication during the measurement period. Before the formal experiment, consent was obtained from each volunteer, and a practice exercise was arranged to familiarize subjects with the task.

2.2.

Stimuli and Procedure

An event-related Stroop task was designed for this study. The stimuli consisted of two Chinese characters, and subjects were instructed to decide whether the color of the upper one matched the meaning of the lower one.¹³ If the answer was “Yes,” subjects pressed the F button with the index finger of their left hand, whereas if the answer was “No,” they pressed the J button with the index finger of their right hand.

Three kinds of stimuli were displayed [Fig. 1a ]: congruent, incongruent, and neutral. These all included lower color-meaning characters in black font. In the congruent trial, the upper character was a color-meaning word (红, 綠, 黃, 蓝 meaning “red,” “green,” “yellow,” and “blue”) for which the color and meaning were consistent. In the incongruent trial, the upper character was a color-meaning word with disparate color and meaning. Meanwhile, in the neutral trial, the upper character was a non-color-meaning word (貫, 獎, 球, 涂, meaning “pass through,” “prize,” “ball,” and “print”). The test included 60 judgment trials (20 congruent, 20 incongruent, and 20 neutral stimuli). In half of the trials, the color of the upper character matched that of the lower character, whereas the two characters were unmatched in the remaining half of the trials

Fig. 1

Experimental design. (a) Stimulus design of each trial, including congruent, incongruent, and neutral stimuli. Each subject was asked whether the color of the upper character was consistent with the meaning of lower character. The correct answer was “Yes” for the three pictures in the upper row and “No” for the three pictures in the lower row. 貫, 獎 mean “pass though” and “prize,” respectively. 红, 綠, 黃, 蓝 correspond to “red,” “green,” “yellow,” and “blue.” (b) Test protocol: between $30\text{-}\mathrm{s}$ rests, 60 trials are administered in a pseudorandom sequence. Each stimulus is displayed for only $100\mathrm{ms}$ , and each trial lasts $12\mathrm{s}$ total. (Color online only.)

The experimental procedure was as follows [Fig. 1b]. A white cross appeared in the center of the screen for $30\mathrm{s}$ , and the subjects should rest. A chime sounded right before the first stimulus appeared, commencing the 60 trials. During the trials, each picture remained on the screen for, followed by an empty screen for $\mathrm{11,900}\mathrm{ms}$ . The order of stimuli was pseudorandom. After the final trial ended, the white cross again appeared for $30\mathrm{s}$ , indicating a rest period.

2.3.

NIRS

The CW-NIRS instrument used to measure changes in oxy-, deoxy-, and tot-hemoglobin concentration is a portable brain function imager operating at three wavelengths (735, 805, and $850\mathrm{nm}$ ). It includes a probe, a measuring and controlling module, and a personal computer.³⁶ The probe has two sources and four detectors that define four areas. As shown in Fig. 2, channel A corresponds to the inferior left PFC, B corresponds to the superior left PFC, C corresponds to the inferior right PFC, and D corresponds to the superior right PFC. The sampling frequency is $3.3\mathrm{Hz}$ , and the separation of source-detector pair is $4\mathrm{cm}$ , which makes the light available to penetrate the scalp, skull, and cerebrospinal fluid, and be modulated by cortical activity.³⁷

Fig. 2

Probe locations and defined area. (a) locations of NIRS optodes and ERP electrodes. (b) area defined for four channels. Channel A contains electrode FP1 of ERP and the Source 1-Detector 1 pair of NIRS; channel B contains electrode AF3 of ERP and the Source 1-Detector 2 pair of NIRS; channel C contains electrode FP2 of ERP and the Source 2-Detector 3 pair of NIRS; and channel D contains electrode AF4 of ERP and the Source 2-Detector 4 pair of NIRS.

The raw data were bandpass filtered using a Hanning Window with a cutoff frequency of $0.01\mathrm{Hz}$ to remove the baseline drifts and $0.7\mathrm{Hz}$ to filter pulsations due to heartbeat and other physiological noise. Three parameters were analyzed: the average amplitude, peak amplitude, and latency of the peak amplitude of each trial. We assessed 36 combinations based on three stimuli (congruent, incongruent, and neutral), three hemodynamic parameters (oxy-, deoxy-, and tot-Hb) and four channels (A–D). The average amplitude of each stimulus condition was averaged across 20 trials per subject. The oxy-Hb and tot-Hb maxima and deoxy-Hb minimum were considered the peak amplitudes of each trial.

2.4.

ERP

The electrophysiological signal was recorded by a 64-channel system using a bandpass filter of $0.05–100\mathrm{Hz}$ and sampling rate of $1000\mathrm{Hz}$ (SynAmps² amplifiers, Neuroscan Ltd.). Only four electrodes (FP1, FP2, AF3, and AF4 of 64 scalp sites standardized by the American Electroencephalographic Society) located on the PFC were used in the study and symmetrically distributed from the NIRS optodes (Fig. 2). The reference electrodes were placed on the bilateral mastoid bone behind the ear. An electrooculogram was also recorded by four electrodes placed around the eyes. The impedance of all electrodes was $<5\mathrm{k}\Omega$ during recording.

The ERP data were processed offline using Scan4.3 software (Neuroscan Ltd.). For each subject, the ocular artifact was first removed. Next, the data were arranged into epochs ( $200\mathrm{ms}$ prestimulus to $1000\mathrm{ms}$ poststimulus), baseline corrected, averaged with other data originating from the same kind of stimuli, filtered by finite impulse response (FIR) ( $0.8–17\mathrm{Hz}$ , $24\mathrm{dB}$ /oct, zero phase shift), and baseline corrected again. All subjects’ data were then group averaged.

2.5.

NIRS-ERP Synchronization and Comparison

Presentation 0.53 software emitted signals to trigger NIRS and ERP sampling at the beginning of the test. The NIRS-ERP synchronization error was much smaller than the NIRS sampling interval. Both correlations and linear regression were utilized to compare the NIRS and ERP signals. No correction for multiple comparisons was performed, and results were considered significant if $p<0.05$ .

3.1.

Behavioral Data

As shown in Fig. 3, the mean reaction time for the incongruent stimulus $(1319.65\pm 67.14\mathrm{ms})$ was longer than that for the congruent $(1236.91\pm 74.08\mathrm{ms})$ or neutral stimulus $(1297.47\pm 70.34\mathrm{ms})$ . Repeated measure analysis of variance (ANOVA) indicated statistically significant differences between the incongruent stimulus and the other two stimuli (congruent versus incongruent $F_{1,15}=7.081$ , $P=0.018<0.05$ ; congruent versus neutral $F_{1,15}=1.795$ , $P=0.200>0.05$ ; incongruent versus neutral $F_{1,15}=5.212$ , $P=0.037<0.05$ ). The mean error rate was higher for the incongruent stimulus $(7.5\pm 2.4\%)$ than for the congruent $(6.9\pm 2.1\%)$ or neutral stimuli $(6.25\pm 1.5\%)$ . Error rate had no significant effect for all three stimuli by repeated-measure ANOVA ( $F_{2,30}=0.211$ , $P>0.05$ ). The relative degrees of error rate for the three stimuli were also similar. However, the between-subjects effect was significant and the variance of each combination was large in magnitude.

Fig. 3

Reaction time (line) and error rate (bar) averaged across all subjects. $\text{Mean}\pm \mathrm{SEM}$ (standard error of measurement).

3.2.

NIRS data

As shown in Figure 4a, the average amplitudes of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ after an incongruent stimulus were significantly larger than those resulting from the other two stimuli ( $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ : congruent versus incongruent $F_{1,63}=6.947$ , $P=0.011<0.05$ ; incongruent versus neutral $F_{1,63}=6.749$ , $P=0.012<0.05$ ; $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ : congruent versus incongruent $F_{1,63}=5.436$ , $P=0.023<0.05$ ; incongruent versus neutral $F_{1,63}=5.216$ , $P=0.026<0.05$ ). The activation of congruent and neutral stimuli were not significantly different from each other ( $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ : $F_{1,63}=0.008$ , $P=0.928>0.05$ ). Compared to the superior PFC (channels B and D), the inferior PFC (channels A and C) had a higher signal amplitude ( $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ : $F_{1,95}=26.201$ , $P=0.000<0.05$ ). In addition, note that the amplitude change of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ was larger than $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ in all channels and the amplitude of $\Delta \left[\mathrm{tot}\text{-}\mathrm{Hb}\right]$ was similar to that of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ .

Fig. 4

Hemodynamic signal measured by NIRS. (a) Average amplitudes of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ , $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ , and $\Delta \left[\mathrm{tot}\text{-}\mathrm{Hb}\right]$ detected by NIRS. Each value is an average of all changes at the same location and for the same trial around all subjects. Four channels are defined as in Fig. 2. $\text{Mean}\pm \mathrm{SEM}$ . (b) Time courses of the averaged $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ in channel A (left inferior area). The time period extends from $2\mathrm{s}$ prestimulus to $12\mathrm{s}$ post-stimulus, and the gray bar represents the $100\text{-}\mathrm{ms}$ stimulus-displaying period.

Figure 4b shows the amplitude of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ during a trial in the left inferior area (channel A). The response to incongruent stimuli occurred more rapidly, and its decrease occurred later, than that of congruent or neutral stimuli. The time courses of both $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ also showed that the incongruent stimulus had the largest hemodynamic response, which returned to baseline after $\sim 11\mathrm{s}$ poststimulus ( $1\mathrm{s}$ before the next stimulus).

The time course of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ showed two small peaks for all stimuli. However, repeated-measure ANOVA showed no significant difference between these two peaks for all three stimuli. The time courses of both $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and $\Delta \left[\mathrm{tot}\text{-}\mathrm{Hb}\right]$ had two peaks only in channel B after congruent and neutral stimuli; in this study, we only chose the second peak for data analysis.

The first peak latency of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ was between 5 and $6\mathrm{s}$ , and the second was between 7 and $8\mathrm{s}$ . The only peak latency of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and $\Delta \left[\mathrm{tot}\text{-}\mathrm{Hb}\right]$ was between 5 and $7\mathrm{s}$ , between the first and second peak latencies of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ , except for that of the incongruent stimulus. The incongruent stimulus, which was the most challenging, had the shortest latency time ( $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ : congruent versus incongruent $F_{1,15}=5.757$ , $P=0.030<0.05$ ; incongruent versus neutral $F_{1,15}=9.001$ , $P=0.009<0.01$ ; $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ peak 1: congruent versus incongruent $F_{1,15}=14.087$ , $P=0.002<0.01$ ; incongruent versus neutral $F_{1,15}=4.879$ , $P=0.043<0.05$ ; $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ peak 2: congruent versus incongruent $F_{1,15}=12.093$ , $P=0.003<0.01$ ; incongruent versus neutral $F_{1,15}=4.825$ , $P=0.044<0.05$ ; see Fig. 5 ).

Fig. 5

Latency time of NIRS and ERP measurements averaged across all subjects and channels. The NIRS signal includes four parameters because $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ has two peaks in each channel ( $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ peak 1 and $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ peak 2). The ERP signal includes P450, N500, and P600 as its three parameters. To more clearly display the average values, the error bar was omitted.

3.3.

ERP Data

On the basis of ERP data, three feature-based components, P450 (positive component from $400–450\mathrm{ms}$ time window), N500 (negative component from $450–550\mathrm{ms}$ time window), and P600 (positive component from $600–700\mathrm{ms}$ time window), showed significant differences for different stimuli. As shown in Fig. 6a, all stimulation responses appeared later than $350\mathrm{ms}$ . There were 36 combinations resulting from the combination of the three stimuli, three feature components and four channels.

Fig. 6

(a) Electrophysiological signal in FP1 (channel A). The first vertical line labels the start of the stimulus, and the second vertical line labels the $350\mathrm{ms}$ after stimulation. The three components of P450, N500, and P600 marked by arrows show significant differences between different stimuli after $350\mathrm{ms}$ . (b) Average amplitude of three components from ERP measurement. Each value is an average of the related component at the same location and for the same trial around all subjects. $\text{Mean}\pm \mathrm{SEM}$ .

Figure 6b shows the components’ amplitude in the channels. The activation level displayed by P600 corresponded to the difficulty level of the stimulus ( $\text{incongruent}>\text{neutral}>\text{congruent}$ ; congruent versus incongruent $F_{1,63}=9.176$ , $P=0.004<0.01$ ; incongruent versus neutral $F_{1,63}=6.607$ , $P=0.013<0.05$ ; congruent versus neutral $F_{1,63}=5.752$ , $P=0.019<0.05$ ). However, N500’s amplitude for the congruent stimulus was the largest (congruent versus incongruent $F_{1,63}=9.176$ , $P=0.004<0.01$ ; incongruent versus neutral $F_{1,63}=7.965$ , $P=0.006<0.01$ ). P600 had the strongest response at the left inferior PFC (channel A), whereas N500’s strongest response was in the right superior area (channel D). For P450, the incongruent stimulus elicited a larger-amplitude response in the superior PFC but a smaller response in the inferior area. The peak latency of ERP also showed that more difficult stimuli resulted in shorter latency time.

3.4.

NIRS-ERP Correlation

To evaluate the correlation between the NIRS and ERP signals, we compared peak latency and peak amplitude data. We first used Pearson correlation analysis to find stronger correlation parameter pairs and then used a linear regression model to characterize the magnitude of the correlation for those pairs. On the basis of the peak latency data, we calculated significant correlations between $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and N500 ( $r=0.718$ , $P<0.01$ ) and between $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ and P600 ( $r=0.744$ , $P<0.01$ ). Meanwhile, P450 had no significant correlation with any hemodynamic parameter. In addition, the single latency peak of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ did not correlate with electrophysiological signals. Analysis using a linear regression model demonstrated good linearity for latencies of N500 and P600 and $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ ( $R^2=0.636$ , $P=0.011$ ). The model also revealed very good linearity for latencies of N500 and P600 and both peak latencies of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ ( $R^2=0.868$ , $P=0.012$ ). When the P450 latency was added as another independent variable, better linearity was achieved ( $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ : $R^2=0.651$ , $P=0.031$ ; $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ : $R^2=0.932$ , $P=0.013$ ).

The peak amplitude of P600 significantly correlated with the amplitudes of many hemodynamic parameters, especially in the inferior left area (Table 1 ). A 0.05-level correlation between P600 and the second peak of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ was observed, but not between P600 and the first peak. In contrast to P600, no significant correlation was observed between N500 or P450 and any hemodynamic parameter. A linear regression model was established using the peak values of N500, P600, and the first peak value of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ as independent variables and the second peak value of $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ as a dependent variable ( $R^2=0.791$ , $P=0.018$ ). A better linear regression model was also established with P450 as an additional independent variable ( $R^2=0.868$ , $P=0.003$ ). Thus, a model with more component information had better linearity, just as in the case of the peak latencies. Meanwhile, the peak value of $\Delta \left[\text{oxy-}\mathrm{Hb}\right]$ had no significant linear relationship with electrophysiological signals.

Table 1

Correlation of NIRS and ERP signal amplitudes (Pearson correlation, N=12 ).

4.1.

PFC Activation in Conflict Solving Processing

As mentioned earlier, responses to the Stroop task usually include two-step stimulus evaluation processing and response production processing (also known as conflict processing). Although the hemodynamic signals measured by NIRS provided good PFC location information, they could not resolve or distinguish whether the step occurred in the PFC due to the Stroop effect. Meanwhile, ERPs directly reflect neural activity and the component latency is consistent with the sequence of cognitive activity.

Compared to previous studies of the Stroop effect, stimulated using either English words¹² or Chinese characters,³⁸ our ERP data showed similar time courses that also suggested that conflict processing happens in the PFC. On the basis of these data, the three feature-based components P450, P600, and N500 showed significant correlation with the Stroop effect. We also found significant spatial and temporal correlations between P600 and hemodynamic signals, whereas N500 and P450 did not show significant spatial correlation. Considering that ERP data have higher temporal resolution but lower spatial resolution than NIRS data, the generators of N500 and P450 are located farther from the PFC than that of P600. P600 shows the most significant correlation with the left inferior area, suggesting that its generator is proximal to this area. The larger amplitude of hemodynamic and electrophysiological signals in the left inferior area also supports this inference.

P600 is the last component of ERP, representing the final course of cognitive neutral activity (Fig. 6). In the Stroop task, the final course of cognition is solving conflict and receiving judgment. The results of our study support the hypothesis that conflict solving is one function of the PFC and that the PFC is the last activation area in the Stroop task. N500 is slightly earlier, possibly reflecting conflict monitoring processing. Some researchers have reported that the ACC is activated after the PFC,^{10, 38} whereas others have shown that the ACC plays a role in conflict monitoring,^{39, 40, 41, 42} implying that it is activated earlier than the PFC. Our study supports the former finding.

4.2.

Physiological Signals in Conflict Solving Processing

The results of our linear regression model suggest that hemodynamic signals reflect the cumulative effect of electrophysiological signals. However, the contribution of each electrophysiological component to the peak hemodynamic signal is different. For example, our data show that the contribution of P450 is not as substantial as that of P600, suggesting that hemodynamic change mainly reflects conflict solving processing. The linear regression models of peak amplitude imply the cumulative effect directly. The models of latency time similarly imply a cumulative effect, although the peak amplitude of the hemodynamic signal appears later than that of electrophysiological components.

As shown in Sec. 3.2, the measured time course of the hemodynamic signals included two peaks. However, a previous study with a similarly event-related design reported only one peak.¹³ This discrepancy may be because the previous study’s NIRS signal was fit to a Gaussian function. Additionally, the study’s behavioral data differed from those of the current report, in which the reaction time was $>1\mathrm{s}$ for all subjects. The difference in reaction time is related to the response bias of the subject and experimental instructions. Longer reaction time may cause divided peaks.

Our results also demonstrate that $\Delta \left[\text{deoxy-}\mathrm{Hb}\right]$ is a more sensitive parameter that strongly correlates with electrophysiological signals. $\Delta \left[\text{Deoxy-}\mathrm{Hb}\right]$ reflects oxygen consumption due to active cognition, so its two peaks imply the occurrence of two cognitive processes. It is suggested that the hemodynamic parameters, in addition to electrophysiological signal, can provide comprehensive information for dissociating cognitive processes.

In conclusion, we investigated the correlation between hemodynamic and electrophysiological signals in the Chinese-character Stroop paradigm using a dual-modality (NIRS and ERP measurements) approach. The positive component from $600\text{to}700\mathrm{ms}$ showed strong correlation with the hemodynamic parameters in the PFC. On the basis of correlation analysis of NIRS and ERP signals, the PFC executes conflict solving. The hemodynamic signal reflects the cumulative effect of the neural response. $\Delta \left[\text{Deoxy-}\mathrm{Hb}\right]$ had two peaks during the trial and was more sensitive than other hemodynamic parameters. Supplementing NIRS with ERP’s temporal data allows NIRS to identify the specialization function of an area and reveal, cognitive connections between areas. Thus, this dual-modality method is a promising approach to investigating neurovascular coupling in physiology and cognition.