2.1.

Sample

Twenty-one right-handed individuals of both sexes (10 males and 11 females) were recruited from the Drexel University community. Subjects claimed no history of neurological or psychological disorders nor use of any medication. They attended an orientation session at least one day prior to the actual experiment when they received detailed information on the experimental protocol and signed the informed consent. During orientation, subjects performed a CPT at 1°C in the real experimental setting to minimize anxiety associated with initial exposure to noxious cold water. Subjects were asked to refrain from smoking and drinking alcohol at least 3 h prior to the experiment.

2.2.

Procedure

Four CPTs at different temperatures of water, i.e., 15°C, 10°C, 5°C, and 1°C, were employed to generate low, moderate, and severe pain levels. Pioneering work on using CPT for pain study by Wolf and Hardy⁴⁰ suggested that pain threshold to cold water is at 18°C. Any decrease in water temperature is reported to be proportional to increase in pain score. The cold water temperatures in this research were chosen accordingly to induce different levels of pain.

For each trial of CPT, a baseline was recorded for 30 s, then subjects were told to immerse their right hand up to the wrist in tepid water ($\sim 23°\mathrm{C}$) for 2 min for adaptation. Then, the experimenter verbally asked subjects to submerge the same hand in a constant temperature bath (15°C, 10°C, 5°C, or 1°C) for as long as they could tolerate the stimulated pain but no more than 5 min. During each experiment, numerical rating scales on a 0 to 10 scale (NRS-11)—where “0” indicates no pain and “10” indicates the worst imaginable pain—were recorded every 15 s in addition to pain threshold (when the first pain was felt) and tolerance (when the pain intensity became intolerable).⁴¹ There was at least half an hour rest period between trials. A block diagram of the protocol is shown in Fig. 1.

Fig. 1

Block diagram of the cold pressor protocol.

The cold and tepid water containers were equipped with commercial aquarium pumps to circulate water and minimize heat buildup around the immersed hand. The cold water container had a separate compartment for ice cubes to prevent any direct contact of the subject’s skin with ice.

2.3.

Instrument

fNIRS is an optical imaging modality that exploits visible light at NIR range (650 to 950 nm), also termed the optical window. A photodetector placed at a certain distance from the light source detects the back-scattered light after it has passed a banana-shape pathway. The depth of light penetration is a function of the distance between the light source and photodetector (S–D): the larger the S–D, the deeper the light penetration. By choosing two appropriate S–Ds, one can sample absorption changes in a short pathway through superficial tissues and in a large pathway scanning deeper tissues within the head. Several theoretical and experimental studies have investigated depth-dependent changes in absorption using different S–D separations.^42–44 We used two S–D separations: 1 cm (“near” channel) and 2.8 cm (“far” channel). Each probe had one “near” channel and two “far” channels. The “near” channel measured the hemodynamic response at the extracerebral layers while the “far” channels measured a superimposition of the hemodynamic changes within the extracerebral layers and the cortical tissue [Fig. 2(b)].

Fig. 2

The fNIRS probes. (a) A demonstration of the placement of fNIRS probes on a subject’s forehead. The probes are shown reversed to illustrate the location of the light source and photodetectors. (b) A schematic of the fNIRS probe configuration. Photons travel from a light source to a photodetector through a banana-shape pathway with a penetration depth of half the source–detector distance. Measures are approximate. Reprinted with permission from Ref. 12.

We employed the continuous wave fNIRS system (730 and 850 nm) developed at Drexel University.^45–47 Our fNIRS device had three main units: (1) two flexible fNIRS probes, each consisting of one multiwavelength light emitting diode and three photodetectors, (2) a control box for operating the hardware, and (3) a computer running the COBI Studio software⁴⁸ for data acquisition and visualization. The two fNIRS probes were placed symmetrically on the subjects’ forehead to measure relative changes in HHb and ${\mathrm{HbO}}_2$ throughout the experiment [Fig. 2(a)].¹² Although the “pain matrix” encompasses multiple regions of the cerebral cortex, we chose to study the frontal region because of the ease of measurements on the forehead.

2.4.

Data Analysis

To eliminate high-frequency noise, respiration, and heart pulsation artifacts, raw intensity measurements were first filtered by a finite-impulse response low-pass filter with a cut-off frequency set to 0.14 Hz, which was chosen based on the literature.⁴⁹ Changes in the concentrations of HHb and ${\mathrm{HbO}}_2$ were calculated relative to the mean value of the optical intensity during the first 15 s of the prestimulus baseline recording. The HHb and ${\mathrm{HbO}}_2$ data were then smoothed using a spline basis expansion by imposing a penalty on the roughness of the second derivative of the data with a smoothing parameter ($\lambda$) of 300, which was chosen by visual judgment and prior knowledge of the process.⁵⁰ An objective, data-driven method is also developed using the generalized cross-validation (GCV) measure⁵¹ and an optimum $\lambda$ minimizes $\mathrm{GCV}(\lambda )$ function plotted against ${\log }_{10}\lambda$. GCV is the most popular procedure for selecting an optimal value for the smoothing parameter. However, like other cross-validation methods, GCV tends to under-smooth the data. For this study, $\lambda$ was chosen empirically because the GCV plot resembled a sigmoid function which was not informative.⁵²

Based on our previous study,^12,52 we defined the following outcome measures: (1) the maximum change in HHb and ${\mathrm{HbO}}_2$ in response to a cold water stimulus, which may occur during or a few seconds after the stimulus, relative to the prestimulus value—hereafter referred to as $\mathrm{\Delta }\mathrm{HHb}$ and $\mathrm{\Delta }{\mathrm{HbO}}_2$; (2) maximum rate of change of ${\mathrm{HbO}}_2$ with respect to time, which occurs within a few seconds after hand immersion in cold water, calculated as the maximum value of the first derivative function—hereafter referred to as $\dot{m}({\mathrm{HbO}}_2)$. This variable was not calculated for HHb due to the high variability of the response. We speculated that $\dot{m}({\mathrm{HbO}}_2)$ variable is best associated with pain threshold as both variables are measured immediately after the hand immersion in cold water. HHb and ${\mathrm{HbO}}_2$ outcome measures of the two “far” channels for each probe were averaged.

We used linear mixed effects models for the statistical analysis of HHb and ${\mathrm{HbO}}_2$.^53,54 Linear mixed models provide a flexible and powerful approach for the analysis of repeated measures. Repeated measures data involve multiple observations on the same subject across a repeated factor, which is temperature here. These repeated measurements are likely to be correlated, therefore, any model fit requires parameter estimation of the covariance structure. Unlike classical repeated measures analysis of variance (ANOVA), which assumes the sphericity of the covariance matrix, linear mixed models allow a wide selection on the form of the variance-covariance matrix, which provides efficiency and flexibility. In a linear mixed model approach, the dependent variable is modeled as a combination of population characteristics, called fixed effects, and subject specific effects that are unique to a particular individual, called random effects. In our model, we included depth (“far” and “near”), side of the measurements (left and right), and gender as the fixed main effects and channel by side, channel by gender, and channel by side by gender as the interactions. We allowed different intercepts and slopes for individuals’ regression lines across temperatures.

For subjective measures of pain, we analyzed the pain threshold, pain tolerance, and maximum pain score reported during the stimulus.

All signal processing calculations were performed in MATLAB (R2011a, MathWorks, Natwick, Massachusetts), and the smoothing was performed using the functional data analysis package for MATLAB.⁵⁰ Statistical analyses were conducted using the IBM SPSS Statistics 19. The significance criterion was $\alpha <0.05$ for all analyses. Post hoc pairwise comparisons were adjusted by Sidak criterion.

3.1.

Subjective Reports of Pain

Friedman’s ANOVA by ranks revealed a significant difference in the median value of the pain threshold [Fig. 3(a)], tolerance [Fig. 3(b)], and maximum pain score [Fig. 3(c)] with respect to temperature [${\chi }^2(3)=42.43$, $p\text{-}\text{value}<0.001$; ${\chi }^2(3)=23.73$, $p\text{-}\text{value}<0.001$; and ${\chi }^2(3)=31.97$, $p\text{-}\text{value}<0.001$, respectively]. Post hoc Wilcoxon signed ranks tests adjusted by Bonferroni criterion yielded that:

Fig. 3

The intensity of the cold stimulus adjusted by water temperature was proportional to the subjective measures of pain sensation: (a) pain threshold, (b) pain tolerance, (c) the maximum pain rating score reported during a CPT on a scale from 0 to 10 ($n=21$; 11 females, 10 males).

We did not find any gender difference in the pain tolerance, pain threshold, or maximum pain score.

3.2.

HHb and ${\mathit{HbO}}_2$

Descriptive statistics of the outcome variables $\mathrm{\Delta }\mathrm{HHb}$, $\mathrm{\Delta }{\mathrm{HbO}}_2$, and $\dot{m}({\mathrm{HbO}}_2)$ are provided in the Appendix. For ${\mathrm{HbO}}_2$, we found a three-way interaction between “gender, side, and depth (channel)” and a two-way interaction between “side and depth” (Table 1 and Fig. 4). A two-way interaction between “gender and depth” and significant main effects of temperature, side, and depth were found for HHb and ${\mathrm{HbO}}_2$, whereas a significant main effect of gender was found only for ${\mathrm{HbO}}_2$ (Table 1).

Table 1

Statistics for the interactions and main effects for the outcome variables ΔHHb, ΔHbO2, and m˙(HbO2). For the definition of the outcome variables, refer to Sec. 2.4.

Fig. 4

Gender difference and asymmetrical hemodynamic activation was found at the “far” channels ($\mathrm{S}-\mathrm{D}=2.8\mathrm{cm}$) but not the “near” channels ($\mathrm{S}-\mathrm{D}=1\mathrm{cm}$) for (a) $\mathrm{\Delta }\mathrm{HHb}$, (b) $\mathrm{\Delta }{\mathrm{HbO}}_2$, and (c) $\dot{m}({\mathrm{HbO}}_2)$ ($n=21$; 11 females, 10 males). S–D stands for source–detector separation. See Sec. 2.4 for the definition of $\mathrm{\Delta }\mathrm{HHb}$, $\mathrm{\Delta }{\mathrm{HbO}}_2$, and $\dot{m}({\mathrm{HbO}}_2)$.

3.3.

Correlation Analysis

Correlation analysis revealed a statistically significant within-subjects correlation between the pain threshold and $\dot{m}({\mathrm{HbO}}_2)$ measured on the right side (right “near” channel: $R=0.44$, $p\text{-}\text{value}<0.05$, right “far” channel: $R=0.46$, $p\text{-}\text{value}<0.05$). When calculated for males and females separately, the correlation was significant for males (right “near” channel: $R=0.80$, $p\text{-}\text{value}<0.01$, right “far” channel: $R=0.77$, $p\text{-}\text{value}<0.01$) but not for females (right “near” channel: $R=0.40$, $t\text{-}\text{value}=1.30$¹¹, right “far” channel: $R=0.48$, $t\text{-}\text{value}=1.65$¹¹).

4.1.

Caveats

First, we understand that the cortical response to CPT is largely confounded by the extracerebral activation given the substantial systemic changes in the skin blood flow. The evoked hemodynamic response measured at the “near” channels is associated with the systemic blood flow changes in the extracerebral layers (e.g., skin and meninges), mainly due to task-induced changes in heart rate and blood pressure; whereas the response measured by the “far” channels is a superimposition of the skin and cortical hemodynamic activation. The origin of the evoked hemodynamic response at the cortical level is unknown to us since the nociceptive component and the autonomic aspect of the response are convoluted. We speculate that it originates from either the evoked local neurovascular coupling secondary to functional brain activity or the evoked systemic cerebral blood flow changes due to cardiovascular reactions such as heart rate and blood pressure variations.

Second, we used water stimulus at 15°C, which is close to cold pain threshold, to elicit less autonomic response. However, our results showed significantly different responses at the skin and the cortex regardless of the water temperature. It seems that dissociation of the nociceptive and autonomic elements of response requires careful control of the experiment, e.g., by including innocuous challenges for autonomic arousal and then, isolating its response from the cortical nociceptive processing. In an effort to evoke a pure autonomic response, Harper et al.⁶⁵ used the Valsalva maneuver and found significant activation in a number of brain regions including the medial and orbital prefrontal cortex that are activated in response to CPT as well. Their observation suggests that the evoked hemodynamic response to the CPTs in the prefrontal cortex may be due to autonomic nervous system activation. Thus, because cold noxious stimulus evokes substantial cardiovascular reaction, careful adjustment of the experiment’s parameters is warranted to differentiate nociceptive processing and autonomic reaction to the CPT challenge.

Third, in our study, we could not find any gender difference in either the pain tolerance or the pain threshold; although gender difference in perception and expression of pain has been widely addressed in literature.⁶⁶ Reports on gender differences in cold pain tolerance are more prevalent than cold pain threshold. Several studies have shown that women are more sensitive to cold pain than men are. A comprehensive survey of literature on gender differences in pain perception during 1998 to 2008 found that 80% of CPT studies showed that females tolerate significantly less pain than males.⁶⁷ Another review noted that only one study (which included 15 male and 19 female participants) out of 23 did not show a gender difference in cold tolerance.⁶⁶ It is suggested that considering the moderate effect size of threshold and tolerance, a sample of 41 subjects for each gender group is needed to reach adequate power.⁶⁸ We acknowledge that the sample size of our study was small (10 males, 11 females) thus, our study may lack enough power to catch the gender differences in cold pain tolerance which is reported in the literature. Nonetheless, since males and females’ subjective reports of pain are comparable in our sample, it is plausible to compare their fNIRS signal under the condition of equivalent pain perception.

The observed gender difference in hemodynamic activation could be due to differences in anatomy, cortical processing, or autonomic regulation. There are some evidences for gender differences in the cerebral blood volume changes during a cognitive task³⁹ and in the regulation of sympathetic nervous system.⁶⁹ However, because we did not do any structural imaging or hormonal test, we cannot make any deterministic statement.

Finally, we did not have access to an automatic blood pressure recorder to monitor blood pressure during the CPTs. We ran similar analyses on the total hemoglobin (${\mathrm{HbO}}_2+\mathrm{HHb}$) and found the same results as ${\mathrm{HbO}}_2$, mainly because the amplitude of ${\mathrm{HbO}}_2$ response is considerably larger than HHb thus, the total hemoglobin response is vastly dominated by ${\mathrm{HbO}}_2$ response.

4.2.

Potential Applications and Future Direction

fNIRS application for the assessment of pain is very recent but the literature shows a fast growing interest in such a novel solution.³² Under the assumption of reliable, self-reporting in healthy subjects, we can study the deviations of the discovered hemodynamic markers from normal values in case of an ailment or under influence of drugs such as morphine (e.g., morphine is shown to aggravate pain sensitization). At least four studies in adults^20–22,70 and five studies in infants^23–27 have used fNIRS in a clinical setting during a painful clinical intervention. It is crucial to report any pharmacologic treatments during experimental or clinical pain studies. For example, it has been reported that administration of morphine in critically ill infants was associated with significantly less changes in ${\mathrm{HbO}}_2$ compared with those who did not receive any analgesic medication.²⁷ Gelinas et al.²⁰ also reported that changes in regional oxygen saturation values during the insertion of intravenous line and arterial line were lower in patients who received morphine compared with those who did not. More studies with specific target populations such as elderly and cognitively impaired are needed to test the validity and reproducibility of fNIRS for pain assessment.

Some suggested applications of the technique proposed here include: (1) assessment of the preoperative sensitivity to CPT and its association with postoperative pain,⁷¹ (2) study of spontaneous pain under natural conditions like walking and talking, (3) examination of young children and elderly who would not stay still in a magnet or a PET scanner, (4) assessment of patients that are under heavy sedation or muscular blocking agents since the sedating agents significantly reduce pain behaviors, and (5) study of factors modulating the pain tolerance, such as long-term opioid use, drug abuse, smoking, and alcohol drinking. For instance, CPT has long been used to assess the opioid-induced hyperalgesia (OIH).⁷² The affordability and ease of use of fNIRS can provide a scheme for longitudinal monitoring of OIH development.

We suggest the use of fNIRS together with a complementary technique such as EEG for improving spatial and temporal resolutions, in addition to behavioral and subjective measures, for a comprehensive, multidimensional pain assessment. Although current fNIRS measurements are subjected to a number of limitations including low spatial resolution ($\sim 1\mathrm{cm}$), limited depth of penetration (0.5 to 2 cm), and signal contamination by physiological noise, technical advancement in hardware and software may improve these factors in future. In particular, isolating the cortical response may help in developing neurofeedback frameworks that aim to design personalized treatments using an individual’s own data for self-regulation. Techniques for separation of superficial hemodynamic response from the cortical activity are based on the assumption that responses measured by the near and far optodes are independent of each other; i.e., the near channel measures a global physiological inference which is independent of the task-evoked response measured by the far channel. This assumption is violated in the CPT as the task evokes a large systemic response in addition to a cortical nociceptive response. Further research is needed to address this important signal processing challenge.

Finally, the bilateral activation at the “far” and “near” channels observed in our study showed very similar trends of change through the time course of the experiment. Time-series analysis of fNIRS parameters was not conducted here. Given the fact that the autonomic response fairly quickly adapts to the stimulus while the pain perception may last for the entire duration of immersion, analysis of the dynamics of the hemodynamic response may unravel some interesting mechanisms.⁵²