Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/spie/comparison-of-oxyhemoglobin-and-deoxyhemoglobin-signal-reliability-GEm0w5JLAG
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- Publisher
- SPIE
- ISSN
- 2329-423X
- eISSN
- 2329-4248
- DOI
- 10.1117/1.NPh.5.1.011006
- Publisher site
- See Article on Publisher Site
Abstract
Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability with and without global mean removal for digit manipulation motor tasks Swethasri Dravida Jack Adam Noah Xian Zhang Joy Hirsch Swethasri Dravida, Jack Adam Noah, Xian Zhang, Joy Hirsch, “Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability with and without global mean removal for digit manipulation motor tasks,” Neurophoton. 5(1), 011006 (2017), doi: 10.1117/1.NPh.5.1.011006. Neurophotonics 5(1), 011006 (Jan–Mar 2018) Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability with and without global mean removal for digit manipulation motor tasks a b b b,c,d,e, Swethasri Dravida, Jack Adam Noah, Xian Zhang, and Joy Hirsch * Yale School of Medicine, Interdepartmental Neuroscience Program, New Haven, Connecticut, United States Yale School of Medicine, Department of Psychiatry, New Haven, Connecticut, United States Yale School of Medicine, Department of Neuroscience, New Haven, Connecticut, United States Yale School of Medicine, Department of Comparative Medicine, New Haven, Connecticut, United States University College London, Department of Medical Physics and Biomedical Engineering, London, United Kingdom Abstract. Functional near-infrared spectroscopy (fNIRS) could be well suited for clinical use, such as measuring neural activity before and after treatment; however, reliability and specificity of fNIRS signals must be ensured so that differences can be attributed to the intervention. This study compared the test–retest and longitudinal reli- ability of oxyhemoglobin and deoxyhemoglobin signals before and after spatial filtering. In the test–retest experi- ment, 14 participants were scanned on 2 days while performing four right-handed digit-manipulation tasks. Group results revealed greater test–retest reliability for oxyhemoglobin than deoxyhemoglobin signals and greater spatial specificity for the deoxyhemoglobin signals. To further characterize reliability, a longitudinal experiment was conducted in which two participants repeated the same motor tasks for 10 days. Beta values from the two tasks with the lowest and highest test–retest reliability, respectively, in the spatially filtered deoxyhemoglobin signal are reported as representative findings. Both test–retest and longitudinal methods confirmed that task and signal type influence reliability. Oxyhemoglobin signals were more reliable overall than deoxyhemoglobin, and removal of the global mean reduced reliability of both signals. Findings are con- sistent with the suggestion that systemic components most prevalent in the oxyhemoglobin signal may inflate reliability relative to the deoxyhemoglobin signal, which is less influenced by systemic factors. © The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI. [DOI: 10.1117/1.NPh.5.1.011006] Keywords: reliability; functional near-infrared spectroscopy; finger-tapping; global mean. Paper 17036SSRR received Mar. 8, 2017; accepted for publication Aug. 21, 2017; published online Sep. 14, 2017. with one particular method or tool. A number of reports 1 Introduction have previously investigated test–retest reliability using Functional near-infrared spectroscopy (fNIRS) is a neuroimag- fNIRS. Some of these reports compared reliability in blood oxy- ing technique that records changes in blood oxygen levels, gen saturation across devices. In one study, Yoshitani et al. which are used as a proxy for localized neural activity. showed that the type of NIRS machine and the methodology Recent advances in fNIRS hardware allow for whole-brain im- used to obtain signals can influence the measures of blood oxy- aging in ecologically valid contexts and have prompted a dra- gen saturation (SO ) differentially in the presence of changing 1 2 matic increase in fNIRS research. Of particular interest are CO concentration in the blood. A number of methodological longitudinal studies, including studies that focus on changes issues were reported for discrepancies, including how blood sat- 2–4 in brain activity underlying learning and training. fNIRS is uration was measured by each machine, the influence of extrac- a relatively inexpensive, radiation-free method of obtaining con- ranial blood flow on both measures, and the source of light used tinuous or repeated measurements that can be used to evaluate for measurements (laser diode versus LED). Reliability in tissue learning, training, intervention, or neurofeedback. For these saturation in infants has also been investigated. This study applications, it is necessary to understand the reliability of reported test–retest and inter-rater reliability for changes in the NIRS signal with respect to changes in oxyhemoglobin hemoglobin measures as well as oxygen saturation using (OxyHb), deoxyhemoglobin (deOxyHb), and systemic effects, NIRS on infants during resting state. The results showed that particularly when signal changes may reflect either the outcome SO measures were reliable both day to day as well as between of a treatment or signal attenuation. raters, but significant differences in hemodynamic measures Reliability is defined as the reproducibility of a were found between baseline measures and across raters. measurement. Variation in the reliability of neural recordings Errors in placement of the single measurement channel were can come from many sources, including equipment, signal-to- one potential source of error suggested by the authors. noise ratio, and participant variability. Test–retest reliability is Functional NIRS has also been evaluated for reliability with an indicator of consistency in repeated measurements made respect to its use in cognitive screening. This study showed mixed results for the reliability of OxyHb and deOxyHb in multiple areas of the frontal lobe but suggested OxyHb may *Address all correspondence to: Joy Hirsch, E-mail: joyhirsch@yahoo.com be more reliable for some types of cognitive tasks. Visual Neurophotonics 011006-1 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . 10 11 and auditory stimulation, as well as motor output tasks, have Connecticut. Each person was compensated for participation also been tested for reliability. In the 2006 paper, the authors in the study. showed that OxyHb signals during a passive visual viewing task were more reliable than deOxyHb. In their 2007 study, they found that deOxyHb showed more localized responses 2.2 Paradigm in a finger-thumb tapping task, but reliability was low for In the test–retest experiments, participants completed four tasks both oxy and deOxyHb signals. The authors suggest that that required right-handed digit manipulation. For the first task differences in probe placement may have contributed to some (“ball squeeze”), participants squeezed an elastic stress ball in of the variability. While there is general agreement about response to cues presented on a computer screen. During the basic mechanisms of action relating to tissue saturation and second task (“double finger tap”), participants tapped each fin- changes in hemoglobin concentration during functional activity, ger sequentially against the thumb twice per cue. For the third the relative reliability of OxyHb and deOxyHb remains an active run (“finger tap”), participants tapped each finger sequentially area of research. The goal of this experiment was to build on against the thumb once per cue. During the fourth run (“follow what has been previously reported and evaluate the effects of the number”), a number from 1 through 4 appeared randomly on global mean removal on the reliability of OxyHb and the screen. Participants were instructed to tap the first finger deOxyHb signals. against the thumb in response to “1,” the middle finger in Functional NIRS records changes in oxyhemoglobin and response to “2,” the ring finger in response to “3,” and the deoxyhemoglobin concentrations. The specific changes in pinky finger in response to “4.” Each run consisted of six blocks. hemodynamic signals recorded with fNIRS reflect underlying neural activity but may also contain multiple sources of systemic Each block consisted of 20 s of task followed by 10 s of rest, 12–14 effects. Two techniques are commonly used for removing during which participants were instructed to focus on a crosshair systemic components from fNIRS signals. The first is short on the screen and keep their hands still. There were 24 cues pre- channel separation, which has been shown to be able to remove sented every 0.83 s during the 20-s task block. a localized artifact that is non-hemodynamic in its temporal activation profile; however, removal of artifacts that are very similar in temporal response to neural activity, such as changes 2.3 Signal Acquisition in blood pressure, may also regress out true neural responses. Data were acquired using a multichannel, continuous wave The second technique is a principle components analysis Shimadzu LABNIRS system (LABNIRS, Shimadzu Corp., (PCA) spatial filter that removes activity distributed across Kyoto, Japan), which consists of emitters that connect to the entire cortex, which has been shown to be effective in iso- laser diodes at three wavelengths (780, 805, and 830 nm). lating neural responses. We evaluate this spatial filter to deter- Each participant was fitted with an optode cap with predefined mine how systemic components affect the reliability of fNIRS distances of 2.75 or 3 cm depending on the size of the individ- recordings. ual’s head. The cap was placed so that the most anterior optode- Here, we investigate the effects of signal and task type on the holder was positioned ∼1cm above nasion and the most pos- reliability of fNIRS data. The overall goal of this study was to terior opode holder 1 cm below inion. These anatomical land- compare the reliability of fNIRS signals in the motor cortex in a marks were chosen to maximize the chance that the cap was test–retest experiment. Specifically, we obtained whole-head placed on a single participant’s head the same way each day. fNIRS recordings during a series of digit manipulation tasks Hair was removed from the channel area prior to placement of in which participants perform stress ball squeezing, finger- each optode using a lighted fiber optic probe (Daiso, Hiroshima, thumb tapping, double finger-thumb tapping, and a finger- Japan). Thirty-two emitters and detectors were arranged in a thumb tapping task in which participants tapped specific digits against their thumb when cued by a number. Expected responses 105-channel layout covering the full head [Fig. 1(a)]. The resis- in the contralateral motor cortex were observed for all tasks tance in each channel was measured prior to recording and across participants on day one and day two. We compared adjustments were made until the channel resistance met the 4,17,18 reliability from day one to day two in OxyHb and deOxyHb minimum LABNIRS requirements. Signals were down- signals in all tasks both before and after spatial filtering of sampled 10-fold during the analysis for an effective sample systemic components. We specifically assessed whether rate of 1.0 s. OxyHb or deOxyHb was a more reliable signal for each of the four motor tasks. Finally, we determined how the OxyHb and deOxyHb signals varied in a longitudinal experiment in 2.4 Optode Localization which two participants repeated the same four motor tasks Following signal acquisition, the optodes were removed from for 10 days. the cap, but the cap was left on the participant for the purpose of optode localization. Anatomical locations of optodes with 2 Test–Retest Experiment Methods 19 respect to the standard 10 to 20 system head landmarks nasion, inion, Cz, T3 (left tragus), and T4 (right tragus) were determined 2.1 Participants using a Patriot 3-D Digitizer (Polhemus, Colchester, Vermont) 20–23 and previously described linear transform techniques. The Fourteen participants (4 male, 10 female; mean age: 26.9 þ ∕− 16 24 NIRS-SPM software was used with MATLAB (Mathworks, 9.5 years; 100% right-handed ) took part in the experiment Natick, Massachusetts) to determine Montreal Neurological over 2 days. Participants provided written informed consent in accordance with guidelines approved by the Yale University Institute (MNI) coordinates for each channel. The correspond- Human Investigation Committee (HIC #1501015178). All data ing anatomical locations for each channel were determined 25,26 were obtained at the Yale School of Medicine, New Haven, using the Talairach atlas. Neurophotonics 011006-2 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Fig. 1 Channel layouts. (a) For the test–retest experiment, 32 detectors and emitters were arranged in a 105 channel layout, represented by the blue circles. This layout covered the frontal, temporal, parietal, and occipital lobes. Black ovals surround 29 channels used for the ROI. (b) For the longitudinal experi- ment, 32 detectors and 29 emitters were arranged in a 98 channel layout. Black ovals surround 23 chan- nels used for the ROI. 2.5 Signal Processing 2.7 Test–Retest Reliability A modified Beer–Lambert equation was used to convert raw To evaluate the reliability of activity in the motor cortex from fNIRS data to deoxyhemoglobin and oxyhemoglobin concentra- day 1 to day 2, each participant’s channel locations were con- tions, and wavelet detrending was applied to these values. A verted to MNI space. Each participant’s data were then regis- fourth-degree polynomial was used to model and remove the tered to the median channel location of both days using a baseline drift from the raw signal. For each participant, channels nonlinear interpolation method. Once in normalized space, reg- were automatically removed from the analysis if the root mean istered beta values were used to calculate test–retest reliability square of the raw data trace was 10 times that of the average for over 2 days. Beta values in all channels from all four tasks from that participant. Comparisons between “clean” and “raw” data both days were averaged, and the channel with the maximum refer to data that did or did not undergo global mean removal, beta value in a preidentified region of interest (ROI) was iden- respectively. To generate the “clean” data, global systemic tified for each participant [black ovals in Fig. 1(a)]. The ROI effects were removed using a spatial filter prior to hemo- comprised 29 channels in the left hemisphere, covering premo- dynamic modeling. The assumption underlying the use of a spa- tor, primary-motor, and supplementary motor areas. The chan- tial filter is that neural activity due to the task, in this case related nel with the maximum average beta value, the channel of to finger movements, would result in activity localized to the interest, differed across participants. Once the channel of interest contralateral motor cortex. Therefore, any activity present across was identified, beta values in that channel were extracted for a larger area of the brain is most likely due to global systemic each task from both days for each participant. The intraclass cor- effects. The algorithm used here utilizes PCA and a high-pass relation coefficient (ICC) was used to compare the degree of Gaussian spatial filter to remove components of the data that are reliability between the beta values on day 1 versus day 2. A present throughout the brain. Raw and clean data were reshaped MATLAB script was used to generate the ICCs for each signal into 4 × 4 × 4 × 133 images, and SPM8 was used for first-level type (deOxyHb and OxyHb) and each processing type (raw and general linear model (GLM) analysis. clean) for each digit manipulation task. 2.6 Contrast Comparisons 2.8 Similarity of Day-to-Day Cap Placement The GLM for fNIRS was used to generate contrast comparisons The reliability of the cap placement from day to day was con- for each task versus rest. The 30-s experimental blocks, which firmed by taking the channel of interest for each participant and included the 20-s task blocks and 10 s of rest, were convolved calculating the distance between the MNI coordinates for this with the hemodynamic response function and modeled to fit the channel on day 1 and the MNI coordinates for the same channel data. This resulted in individual beta values for each participant on day 2. The average distance between the channel of interest for every task. Beta values were obtained for all channels. One- on day 1 and day 2 was 9.5 6.7 mm, confirming that varia- tailed t-tests were used to generate group-level data in SPM8. tions in cap placements on both days were within the spatial Results were rendered at a threshold of p < 0.005. resolution of 3 cm. Neurophotonics 011006-3 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . In this study, the single channel with the maximum beta value for each signal type (OxyHb and deOxyHb, Fig. 2, rows). 13–15 in the ROI was used to identify beta values as the measure of Consistent with prior studies, raw OxyHb data showed a reliability because data from a single channel reflect the most distributed pattern of activity that became localized when the specific local neural activity from each participant. This was spatial filter was used, while deOxyHb signals were localized intended to eliminate variability in the group location of activity to motor cortex for both raw and clean results. due to variation in head shape across individuals. Variations in the location of the channel of interest for a single subject 3.2 Test–Retest Signal Reliability between day 1 and day 2 were within the spatial resolution of a single channel and, therefore, contributions to measures ICCs were used as a measure of test–retest signal reliability. The of reliability were not detectable. Within an individual subject, ICC values were determined for each task and both types of sig- neural activity related to the task was restricted to a few channels nal: deOxyHb versus OxyHb with and without global mean in the primary motor or premotor/supplementary motor cortex removal. The overall ICC value for each signal type was also with peak activity in one. Only in the group results, when the calculated using the beta values from all tasks from both data were interpolated across subjects, did the combined activity days. The ICCs are shown in Table 1. When collapsed across cover a larger area of cortex. Using an average of the beta values raw and clean data, the ICC values for the four tasks obtained in the entire area in a single individual would have therefore from the OxyHb signal were significantly greater than those resulted in averaging “zeros” from channels with no significant obtained from the deOxyHb signal (one-tailed paired t-test, activity, reducing the chance of comparing real activity from day p ¼ 0.048). When collapsed across OxyHb and deOxyHb to day. data, the ICC values for the four tasks using the raw data were significantly greater than the ICC values for the clean 3 Test–Retest Results data (one-tailed paired t-test, p ¼ 0.0086). 3.1 Contrast Results 4 Longitudinal Experiment The group-level results for all combined tasks versus the rest in We compared the amount of systemic component versus neural the test–retest experiment are shown in Fig. 2 and areas in the signals in these tasks by conducting a second experiment left motor regions are reported in the tables in the Appendix. The (referred to as the longitudinal experiment) in which two par- results of the group-level contrast from each individual task are ticipants performed the same four tasks every day for 10 days. also shown in the Appendix. As predicted, each of the right- The basis for this experiment was the expectation that when a handed digit manipulation tasks resulted in activity in the left participant undergoes the same task every day, neural signals 28–30 premotor, primary motor, and supplementary motor cortices. will attenuate over time due to a repetition effect. The Results are presented for raw and clean data (Fig. 2, columns) hypothesis for this experiment was twofold. First, we predicted Fig. 2 Results of all tasks contrast, p < 0.005. The top row represents the group-averaged deOxyHb data; bottom row represents OxyHb data. (a) Results of raw deOxyHb data from day 1 and day 2. (b) Results of clean (spatial filter applied) deOxyHb data from day 1 and day 2. (c) Results of raw OxyHb data from day 1 and day 2. (d) Results of clean (spatial filter applied) OxyHb data from day 1 and day 2. Neurophotonics 011006-4 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Table 1 Test–retest ICC for each task and for the combination of all We compared the longitudinal beta values using the deOxyHb tasks. and OxyHb signals with and without the spatial filter from both of these tasks for each participant over the course of 10 days. DeoxyHb OxyHb Raw Clean Raw Clean 5.3 Regression Slope Tests Ball squeeze 0.5906 0.3851 0.8194 0.7086 The beta values in the channel of interest were plotted over the 10 days for each participant in each task. A trend line was fitted Double tap 0.5922 0.4539 0.8463 0.8307 to the data, and a regression slope test was performed on each trend line, to evaluate whether the slope was significantly differ- Single tap 0.6815 0.5687 0.7374 0.5142 ent from zero. In a regression slope test, a t statistic is obtained Follow the number 0.6580 0.6722 0.5878 0.5843 by dividing the slope of the line by the standard error of the slope. This t statistic was then converted to a p value using All tasks 0.6188 0.5020 0.7659 0.6588 the degrees of freedom (number of points -2). A p value less than 0.05 was considered to be significant, namely that the slope of the trend line over the 10 days was significantly differ- ent from zero. that tasks that generate more systemic artifacts would show less attenuation in the OxyHb data over the course of the 6 Longitudinal Experiment Results 10 days than tasks that generate fewer artifacts. Second, we Both participants completed the four tasks of experiment 1 every hypothesized that deOxyHb data would show similar attenua- day for 10 days. Here, we compare the results from the ball tion in all tasks, as this signal is less affected by systemic 12,14,15 squeeze task and the follow the number task using the artifacts. OxyHb and deOxyHb signals, with and without the spatial filter. The graphs in Fig. 3 show the beta values and trend lines for 5 Longitudinal Experiment Methods each signal type over the course of the 10 days for participants 5.1 Participants 1 and 2. Red points and lines represent data using the OxyHb signal and blue points and lines represent the deOxyHb signal. Two right-handed participants (one 25-year-old female and Triangular points and darker colors represent raw data and cir- one 42-year-old male) participated in the longitudinal experi- cular points and lighter colors represent clean (spatially filtered) ment and were tested for 10 days. As above, participants data. We refer to “attenuation” over the 10 days if the slope of provided written informed consent in accordance with guide- the line was negative and significantly different from zero lines approved by the Yale University Human Investigation (p < 0.05). Slopes of each trend line are shown in Table 2. Committee (HIC #1501015178), and all data were obtained For subject 1, no signal showed attenuation over the 10 days at the Yale School of Medicine, New Haven, Connecticut. during the ball squeeze task; although, the clean deOxyHb sig- The task paradigms, signal acquisition, and signal processing nals showed a negative trend (p ¼ 0.067). In the follow-the- methods used were the same as in the test–retest experiment (see number task, however, all signals showed attenuation Secs. 2.2–2.6 above). For this experiment, 32 emitters and 29 (p < 0.05) except the raw deOxyHb signal, which showed a detectors were arranged in a 98-channel layout [Fig. 1(b)], as trend (p ¼ 0.084). For subject 2, no signals showed attenuation, coverage of the occipital lobe was deemed unnecessary. The either in the ball squeeze or the follow-the-number task. ROI consisted of 23 channels in the left hemisphere [black ovals in Fig. 1(b)]. 7 Discussion To study the test–retest reliability of fNIRS signals, we asked 5.2 Intraparticipant Signal Consistency participants to perform four different motor tasks on two sepa- The same channel registration method described in the first rate days. Overall, the OxyHb signal was shown to be more reli- experiment was used to register the data from all 10 days to able than the deOxyHb signal, and the reliability was also higher one set of channel locations for each participant. To evaluate for the raw signal than for signals that had undergone a spatial interscan variability, registered beta values from each channel filter. To test the extent to which these signals were stable over were averaged over 10 days for each participant. The channel time, a longitudinal study was conducted in which two partic- with the maximum average beta value in the left motor cortex ipants performed the same four motor tasks for 10 days. A com- ROI was identified for each participant, and beta values in this parison of two representative tasks with the highest and lowest channel were identified for every task performed over 10 days test–retest reliability in the clean deOxyHb signals (follow the for each participant using raw and clean signal data. As with the number and ball squeeze, respectively) showed that these tasks test–retest experiment, the channel of interest was selected for elicit different levels of neural and global components in some each participant. Main effects from the ball squeeze and follow participants. For the “ball squeeze” task, which had the lowest the number tasks are shown below. These two tasks were chosen test–retest reliability using the clean deOxyHb data, neither par- because they showed the greatest difference in test–retest reli- ticipant showed attenuation in any signal, OxyHb or deOxyHb, ability in the clean, deOxyHb signal. Out of the four tasks, the with or without the spatial filter over the course of the 10 days. ball squeeze task showed the lowest reliability (ICC ¼ 0.3567) However, for the “follow the number” task, which showed the and the follow-the-number task showed the highest reliability highest test–retest reliability in the clean deOxyHb signal, in one (ICC ¼ 0.6264) using the deOxyHb signal with the spatial filter, participant, signals showed attenuation over 10 days, indicating which was consistent with the least amount of systemic artifact. that this task may elicit less systemic noise in some subjects. Neurophotonics 011006-5 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Fig. 3 Results of 10-day longitudinal study. Top row represents data from participant 1; bottom row rep- resents data from participant 2. Left column shows result of the ball squeeze task; right column shows result of the follow the number task. Daily beta values and trend lines are shown. Red represents OxyHb; blue represents deOxyHb. Dark colors and triangles represent raw data; lighter colors and circles re- present clean data. suggests that systemic effects are more similar day to day than Table 2 Slopes of functions in Fig. 3. Asterisks (*) indicate signifi- neural effects. Reduced reliability for the clean signals relative cance (p < 0.05); (−) indicates negative slope. to the raw signals also supports this interpretation. This reliabil- ity, however, comes at the cost of functional specificity, as the DeOxyHb OxyHb activity represented by the raw OxyHb signal was widely dis- tributed rather than limited to left motor cortex and adjacent Raw Clean Raw Clean areas. Applying a spatial filter to these results improved the functional specificity but still showed higher reliability values Subject 1 Ball squeeze −5.87 −8.37 −1.36 −3.09 when compared to the deOxyHb signal, which is known to * * * Follow the number −6.58 −11.27 −13.00 −9.70 be less susceptible to systemic effects. The group findings also support the conclusion that the Subject 2 Ball squeeze 0.41 3.93 −3.98 −6.31 deOxyHb signal, while less reliable, was less affected by sys- temic components than OxyHb, suggested by the more localized Follow the number −8.31 −3.97 −13.37 −9.44 region of activity in Fig. 2. The reduced reliability observed for the deOxyHb signal, assumed to be primarily neural in origin, raises important questions about the nature of the interaction This study is the first to our knowledge to systematically analyze between systemic and neural effects between tasks and across the reliability of OxyHb and deOxyHb signals with and without days, and about strategies to improve signal acquisition and a spatial filter, and our findings suggest that systemic compo- processing. nents present in fNIRS signals may be individually specific and For the Ball squeeze task in the longitudinal experiment, the inflate day-to-day reliability relative to the underlying neural beta values for the OxyHb were higher than for the deOxyHb components. Further, the global mean may inform physiological throughout the 10 days for all tasks in both subjects. Even when processes associated with specific functional tasks and neural the spatial filter was applied to the OxyHb signals, the beta val- mechanisms, adding insight to our understanding of neurovas- ues remained high in both participants without evidence of cular interactions. attenuation. Similarly, the deOxyHb signal did not show signifi- Prior studies have shown that the OxyHb signal is more cant attenuation for either participant during this task, even after 12,14,15 susceptible to systemic artifacts than the deOxyHb signal application of the spatial filter. One interpretation of this result is and our results additionally indicate that the OxyHb signal is that both signals obtained during this task were more influenced more reliable than the deOxyHb signal. One interpretation by systemic components. This is consistent with the group Neurophotonics 011006-6 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . results for this task, which show widespread, non-specific acti- Findings of this study suggest that day-to-day reliability of vation in both hemispheres in the raw, OxyHb signals (see fNIRS recordings depends on both the signal and task used and Appendix, Fig. 4). It is possible that the global components that reliability may be inflated by systemic factors rather than in the signals elicited by this task masked any putative signal neural activity. In general, neural and systemic components dif- attenuation over the course of the 10 days. ferentially affect OxyHb and deOxyHb signals. The deOxyHb By contrast, the task with the highest reliability in the test– signal was less reliable but the task-based effect was spatially retest experiment using the clean, deOxyHb signal was follow specific. These factors are especially important in the design the number, which had a cognitive component that the other of pre/postintervention experiments as they influence the tasks did not. For the follow the number task, participants likelihood that changes in signals reflect neural effects of the were asked to move a specific finger in response to the num- treatment or intervention. bered cue, requiring increased attention to execute an unpre- dictable sequence of finger movements. In the test–retest experiment, this task showed low reliability in the OxyHb sig- nals, but higher reliability in the deOxyHb and especially in the Appendix: Group-level Task Contrast Results clean data. This is consistent with the hypothesis that global Group-level results from each of the motor tasks in the test– components in the data lead to higher reliability, and this retest experiment are shown below (Figs. 4–7). Areas in the task may either elicit less systemic artifacts or the global com- left motor cortex from each contrast are listed in Table 3 ponents may be more effectively separated from neural signals (day one) and Table 4 (day two). in this task using our method of global mean removal. In the longitudinal experiment, for one participant, this task showed attenuation over 10 days in all signal types, supporting the Disclosures hypothesis that it may not elicit as much global signal as the ball squeeze task for this subject. However, this attenuation This work was partially supported by NIH Medical Scientist was not present in the data from the second participant, Training Program Training Grant T32GM007205. The work showing that this effect was variable across participants. also received partial supported by the National Institute of While the global systemic artifact increases the reliability of Mental Health of the National Institutes of Health under the data, using a task that elicits a less global signal may result award number R01MH107513 (PI JH). The content is solely in signals that are more likely to be neural in origin. However, the responsibility of the authors and does not necessarily re- further studies that directly record systemic measures are nec- present the official views of the National Institutes of Health. essary to determine what, if any, task features cause increases Authors report no conflict of interest. The data reported in or decreases in the systemic artifact accompanying neural this paper are available upon request. signals. Fig. 4 Results of ball squeeze task, p < 0.005. The top row represents the group-averaged deOxyHb data; bottom row represents OxyHb data. (a) Results of raw deOxyHb data from day 1 and day 2. (b) Results of clean (spatial filter applied) deOxyHb data from day 1 and day 2. (c) Results of raw OxyHb data from day 1 and day 2. (d) Results of clean (spatial filter applied) OxyHb data from day 1 and day 2. Neurophotonics 011006-7 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Fig. 5 Results of double tap task, p < 0.005. The top row represents the group-averaged deOxyHb data; bottom row represents OxyHb data. (a) Results of raw deOxyHb data from day 1 and day 2. (b) Results of clean (spatial filter applied) deOxyHb data from day 1 and day 2. (c) Results of raw OxyHb data from day 1 and day 2. (d) Results of clean (spatial filter applied) OxyHb data from day 1 and day 2. Fig. 6 Results of single tap task, p < 0.005. The top row represents the group-averaged deOxyHb data; bottom row represents OxyHb data. (a) Results of raw deOxyHb data from day 1 and day 2. (b) Results of clean (spatial filter applied) deOxyHb data from day 1 and day 2. (c) Results of raw OxyHb data from day 1 and day 2. (d) Results of clean (spatial filter applied) OxyHb data from day 1 and day 2. Neurophotonics 011006-8 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Fig. 7 Results of follow the number task, p < 0.005. The top row represents the group-averaged deOxyHb data; bottom row represents OxyHb data. (a) Results of raw deOxyHb data from day 1 and day 2. (b) Results of clean (spatial filter applied) deOxyHb data from day 1 and day 2. (c) Results of raw OxyHb data from day 1 and day 2. (d) Results of clean (spatial filter applied) OxyHb data from day 1 and day 2. Table 3 Group-level GLM results from day one. Clusters of positive activity in the motor cortex and surrounding areas are listed. Horizontal lines separate results from each contrast (in bold). (−) on the x axis indicates left side. BA = Brodmann’s area. MNI coordinates were converted to Talairach coordinates to generate cluster labels. Signal Peak MNI coordinates Peak T P # voxels BA Anatomical area Probability Ball squeeze task OxyHb, Raw −48 −18 62 6.40 0.00001 19143 3 Primary somatosensory cortex 0.31 6 Premotor and supplementary motor cortex 0.29 1 Primary somatosensory cortex 0.15 2 Primary somatosensory cortex 0.13 4 Primary motor cortex 0.10 OxyHb, clean −42 −8 64 6.66 0.00000 804 6 Premotor and supplementary motor cortex 0.79 3 Primary somatosensory cortex 0.12 DeOxyHb, raw −44 −24 66 3.96 0.00063 202 3 Primary somatosensory cortex 0.27 6 Premotor and supplementary motor cortex 0.25 1 Primary somatosensory cortex 0.16 2 Primary somatosensory cortex 0.15 4 Primary motor cortex 0.13 Neurophotonics 011006-9 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Table 3 (Continued). Signal Peak MNI coordinates Peak T P # voxels BA Anatomical area Probability DeOxyHb, clean −44 −6 62 4.65 0.00016 364 6 Premotor and supplementary motor cortex 0.80 3 Primary somatosensory cortex 0.11 Double tap task OxyHb, raw −40 −22 68 4.87 0.00010 4802 6 Premotor and supplementary motor cortex 0.40 3 Primary somatosensory cortex 0.25 4 Primary motor cortex 0.17 1 Primary somatosensory cortex 0.12 OxyHb, clean −40 −4 64 4.26 0.00034 698 6 Premotor and supplementary motor cortex 0.92 DeOxyHb, raw −38 −14 62 5.43 0.00004 1811 6 Premotor and supplementary motor cortex 0.71 3 Primary somatosensory cortex 0.19 DeOxyHb, clean −44 −20 64 6.89 0.00000 2005 6 Premotor and supplementary motor cortex 0.35 3 Primary somatosensory cortex 0.28 1 Primary somatosensory cortex 0.16 4 Primary motor cortex 0.11 Single tap task OxyHb, raw −26 −22 70 4.93 0.00009 4536 6 Premotor and supplementary motor cortex 0.58 4 Primary motor cortex 0.27 3 Primary somatosensory cortex 0.14 OxyHb, clean −44 −10 60 4.50 0.00021 705 6 Premotor and supplementary motor cortex 0.67 3 Primary somatosensory cortex 0.21 DeOxyHb, Raw −50 −10 56 4.20 0.00038 661 6 Premotor and supplementary motor cortex 0.58 3 Primary somatosensory cortex 0.21 DeOxyHb, clean −44 −18 54 4.86 0.00010 1102 3 Primary somatosensory cortex 0.36 6 Premotor and supplementary motor cortex 0.25 1 Primary somatosensory cortex 0.19 4 Primary motor cortex 0.13 Follow the number task OxyHb, raw −44 −8 62 6.33 0.00001 5429 6 Premotor and supplementary motor cortex 0.74 3 Primary somatosensory cortex 0.16 OxyHb, clean −56 −4 48 5.34 0.00004 734 6 Premotor and supplementary motor cortex 0.79 DeOxyHb, raw −46 −8 60 5.44 0.00003 846 6 Premotor and supplementary motor cortex 0.70 3 Primary somatosensory cortex 0.20 DeOxyHb, clean −46 −8 60 5.76 0.00002 911 6 Premotor and supplementary motor cortex 0.70 3 Primary somatosensory cortex 0.20 Neurophotonics 011006-10 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Table 3 (Continued). Signal Peak MNI coordinates Peak T P # voxels BA Anatomical area Probability All tasks OxyHb, raw −48 −20 62 6.43 0.00001 14279 3 Primary somatosensory cortex 0.31 6 Premotor and supplementary motor cortex 0.23 1 Primary somatosensory cortex 0.18 2 Primary somatosensory cortex 0.16 OxyHb, clean −42 −8 64 5.89 0.00001 947 6 Premotor and supplementary motor cortex 0.79 3 Primary somatosensory cortex 0.12 DeOxyHb, raw −46 −10 60 4.98 0.00008 1375 6 Premotor and supplementary motor cortex 0.62 3 Primary somatosensory cortex 0.22 4 Primary motor cortex 0.10 DeOxyHb, clean −32 −6 66 6.26 0.00001 1571 6 Premotor and supplementary motor cortex 1.00 Table 4 Group-level GLM results from day two. Clusters of positive activity in the motor cortex and surrounding areas are listed. Horizontal lines separate results from each contrast (in bold). (−) on the x axis indicates left side. BA = Brodmann’s area. MNI coordinates were converted to Talairach coordinates to generate cluster labels. Signal Peak MNI coordinates Peak T P # Voxels BA Anatomical area Probability Ball squeeze task OxyHb, raw 34 −42 70 6.69 0.00000 31062 5 Somatosensory association cortex 0.35 2 Primary somatosensory cortex 0.19 3 Primary somatosensory cortex 0.16 7 Somatosensory association cortex 0.15 OxyHb, clean −42 2 48 3.20 0.00296 61 6 Premotor and supplementary motor cortex 0.88 8 Includes frontal eye fields 0.12 DeOxyHb, raw −50 −22 60 3.87 0.00075 206 3 Primary somatosensory cortex 0.31 2 Primary somatosensory cortex 0.21 1 Primary somatosensory cortex 0.19 6 Premotor and supplementary motor cortex 0.11 DeOxyHb, clean −52 −22 58 4.08 0.00049 189 3 Primary somatosensory cortex 0.31 2 Primary somatosensory cortex 0.22 1 Primary somatosensory cortex 0.17 40 Supramarginal gyrus part of Wernicke’s area 0.11 Neurophotonics 011006-11 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Table 4 (Continued). Signal Peak MNI coordinates Peak T P # Voxels BA Anatomical area Probability Double tap task OxyHb, raw −60 −20 48 4.73 0.00013 2604 6 Premotor and supplementary motor cortex 0.24 2 Primary somatosensory cortex 0.23 3 Primary somatosensory cortex 0.19 1 Primary somatosensory cortex 0.17 40 Supramarginal gyrus part of Wernicke’s area 0.11 OxyHb, clean −54 −12 52 4.74 0.00013 485 6 Premotor and supplementary motor cortex 0.52 3 Primary somatosensory cortex 0.19 4 Primary motor cortex 0.12 DeOxyHb, raw −40 −20 68 4.03 0.00055 548 6 Premotor and supplementary motor cortex 0.48 3 Primary somatosensory cortex 0.24 4 Primary motor cortex 0.16 1 Primary somatosensory cortex 0.10 DeOxyHb, clean −40 −20 68 4.37 0.00028 786 6 Premotor and supplementary motor cortex 0.48 3 Primary somatosensory cortex 0.24 4 Primary motor cortex 0.16 1 Primary somatosensory cortex 0.10 Single tap task OxyHb, raw −50 −16 58 4.51 0.00021 7521 6 Premotor and supplementary motor cortex 0.33 3 Primary somatosensory cortex 0.28 2 Primary somatosensory cortex 0.14 1 Primary somatosensory cortex 0.14 4 Primary motor cortex 0.10 OxyHb, clean −52 −14 56 4.87 0.00010 280 6 Premotor and supplementary motor cortex 0.42 3 Primary somatosensory cortex 0.22 1 Primary somatosensory cortex 0.13 2 Primary somatosensory cortex 0.12 4 Primary motor cortex 0.11 DeOxyHb, raw −60 −14 46 4.21 0.00038 440 6 Premotor and supplementary motor cortex 0.48 3 Primary somatosensory cortex 0.19 1 Primary somatosensory cortex 0.15 2 Primary somatosensory cortex 0.11 Neurophotonics 011006-12 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . Table 4 (Continued). Signal Peak MNI coordinates Peak T P # Voxels BA Anatomical area Probability DeOxyHb, clean −60 −14 46 4.63 0.00016 605 6 Premotor and supplementary motor cortex 0.48 3 Primary somatosensory cortex 0.19 1 Primary somatosensory cortex 0.15 2 Primary somatosensory cortex 0.11 Follow the number task OxyHb, raw −44 −14 64 6.62 0.00000 3870 6 Premotor and supplementary motor cortex 0.56 3 Primary somatosensory cortex 0.27 OxyHb, clean −24 −10 74 4.95 0.00009 643 6 Premotor and supplementary motor cortex 0.98 DeOxyHb, raw −64 −20 34 4.91 0.00009 1306 40 Supramarginal gyrus part of Wernicke’s area 0.26 6 Premotor and supplementary motor cortex 0.23 2 Primary somatosensory cortex 0.16 1 Primary somatosensory cortex 0.14 DeOxyHb, Clean −64 −16 32 4.88 0.00010 643 6 Premotor and supplementary motor cortex 0.31 2 Primary somatosensory cortex 0.14 40 Supramarginal gyrus part of Wernicke’s area 0.14 43 Subcentral area 0.13 1 Primary somatosensory cortex 0.11 All tasks OxyHb, raw −46 −32 54 6.71 0.00000 19650 40 Supramarginal gyrus part of Wernicke’s area 0.45 2 Primary somatosensory cortex 0.27 1 Primary somatosensory cortex 0.18 OxyHb, clean −52 −14 56 4.75 0.00013 480 6 Premotor and supplementary motor cortex 0.42 3 Primary somatosensory cortex 0.22 1 Primary somatosensory cortex 0.13 2 Primary somatosensory cortex 0.12 4 Primary motor cortex 0.11 DeOxyHb, Raw −60 −14 46 3.87 0.00075 911 6 Premotor and supplementary motor cortex 0.48 3 Primary somatosensory cortex 0.19 1 Primary somatosensory cortex 0.15 2 Primary somatosensory cortex 0.11 6 Premotor and supplementary motor cortex 0.48 DeOxyHb, clean −60 −14 46 4.05 0.00052 883 3 Primary somatosensory cortex 0.19 1 Primary somatosensory cortex 0.15 2 Primary somatosensory cortex 0.11 Neurophotonics 011006-13 Jan–Mar 2018 Vol. 5(1) Dravida et al.: Comparison of oxyhemoglobin and deoxyhemoglobin signal reliability. . . 20. A. T. Eggebrecht et al., “A quantitative spatial comparison of high-den- Acknowledgments sity diffuse optical tomography and fMRI cortical mapping,” The authors thank Jenny Park and Pawan Lapborisuth for assis- NeuroImage 61(4), 1120–1128 (2012). tance with data collection. 21. S. L. Ferradal et al., “Atlas-based head modeling and spatial normali- zation for high-density diffuse optical tomography: in vivo validation against fMRI,” NeuroImage 85(Part 1), 117–126 (2014). References 22. M. Okamoto and I. Dan, “Automated cortical projection of head-surface locations for transcranial functional brain mapping,” NeuroImage 26(1), 1. D. A. Boas et al., “Twenty years of functional near-infrared spectroscopy: 18–28 (2005). introduction for the special issue,” NeuroImage 85(Part 1), 1–5 (2014). 23. A. K. Singh et al., “Spatial registration of multichannel multi-subject 2. D. Leff et al., “Comparing prefrontal responses evoked by a visuomotor fNIRS data to MNI space without MRI,” NeuroImage 27(4), 842– task and a cognitive problem solving task: a longitudinal fNIRS evalu- 851 (2005). ation in surgical trainees,” NeuroImage 47, S190 (2009). 24. J. C. Ye et al., “NIRS-SPM: statistical parametric mapping for near- 3. S. Koike et al., “Association between rostral prefrontal cortical activity infrared spectroscopy,” NeuroImage 44(2), 428–447 (2009). and functional outcome in first-episode psychosis: a longitudinal func- 25. J. L. Lancaster et al., “Automated labeling of the human brain: a tional near-infrared spectroscopy study,” Schizophr. Res. 170(2–3), preliminary report on the development and evaluation of a forward- 304–310 (2016). transform method,” Hum. Brain Mapp. 5(4), 238–242 (1997). 4. Y. Ono et al., “Frontotemporal oxyhemoglobin dynamics predict perfor- 26. J. L. Lancaster et al., “Automated Talairach atlas labels for functional mance accuracy of dance simulation gameplay: temporal characteristics brain mapping,” Hum. Brain Mapp. 10(3), 120–131 (2000). of top-down and bottom-up cortical activities,” NeuroImage 85, 27. K. J. Friston et al., “Statistical parametric maps in functional imaging: a 461–470 (2014). general linear approach,” Hum. Brain Mapp. 2(4), 189–210 (1994). 5. L. G. Portney and M. P. Watkins, Foundations of Clinical Research: 28. R. N. Henson and M. D. Rugg, “Neural response suppression, haemo- Applications to Practice, Prentice Hall, Upper Saddle River, New dynamic repetition effects, and behavioural priming,” Neuropsychologia Jersey (2000). 41(3), 263–270 (2003). 6. J. P. Weir, “Quantifying test-retest reliability using the intraclass corre- 29. K. Grill-Spector, R. Henson, and A. Martin, “Repetition and the brain: lation coefficient and the SEM,” J. Strength Cond. Res. 19(1), 231–240 neural models of stimulus-specific effects,” Trends Cognit. Sci. 10(1), (2005). 14–23 (2006). 7. K. Yoshitani et al., “A comparison of the INVOS 4100 and the NIRO 30. A. Pannese and J. Hirsch, “Self-face enhances processing of immedi- 300 near-infrared spectrophotometers,” Anesth. Analg. 94(3), 586–590 ately preceding invisible faces,” Neuropsychologia 49(3), 564–573 (2002). (2011). 8. J. Menke et al., “Reproducibility of cerebral near infrared spectroscopy in neonates,” Biol. Neonate 83(1), 6–11 (2003). Swethasri Dravida is a graduate student at Yale School of Medicine. 9. A. Watanabe et al., “Cerebrovascular response to cognitive tasks and She received her BS degree in mathematics and brain and cognitive hyperventilation measured by multi-channel near-infrared spectros- sciences from Massachusetts Institute of Technology in 2013. Her copy,” J. Neuropsychiatry Clin. Neurosci. 15(4), 442–449 (2003). current research interests include using functional near-infrared spec- 10. M. M. Plichta et al., “Event-related functional near-infrared spectros- troscopy and EEG to study social interaction, especially in clinical copy (fNIRS): are the measurements reliable?” NeuroImage 31(1), contexts, such as autism. 116–124 (2006). 11. M. Plichta et al., “Event‐related functional near‐infrared spectroscopy Jack Adam Noah received his PhD in biomedical sciences from (fNIRS) based on craniocerebral correlations: reproducibility of activa- Marshall University School of Medicine in 2003. He is an associate tion?” Hum. Brain Mapp. 28(8), 733–741 (2007). research scientist at Yale School of Medicine in the Department of 12. I. Tachtsidis and F. Scholkmann, “False positives and false negatives in Psychiatry and the Brain Function Laboratory. His research interests functional near-infrared spectroscopy: issues, challenges, and the way include functional near-infrared spectroscopy and integration of other forward,” Neurophotonics 3(3), 030401 (2016). multimodal and behavioral recording techniques for applications in 13. G. Strangman et al., “A quantitative comparison of simultaneous BOLD communication and social interactions, neurofeedback, and cognitive fMRI and NIRS recordings during functional brain activation,” neuroimaging. NeuroImage 17(2), 719–731 (2002). 14. X. Zhang, J. A. Noah, and J. Hirsch, “Separation of the global and Xian Zhang received his PhD in psychology and visual science from Columbia University in New York in 2003. He is an associate research local components in functional near-infrared spectroscopy signals using scientist in the Brain Function Laboratory in the Department of principal component spatial filtering,” Neurophotonics 3(1), 015004 Psychiatry, Yale School of Medicine. His research interests include (2016). computational neuroscience, signal processing, and neuroimaging 15. E. Kirilina et al., “The physiological origin of task-evoked systemic technologies, such as EEG, fNIRS, and fMRI and their applications artefacts in functional near infrared spectroscopy,” NeuroImage 61(1), in psychiatry, vision science, social interactions, and decision making. 70–81 (2012). 16. R. C. Oldfield, “The assessment and analysis of handedness: the Joy Hirsch received her PhD in psychology and visual science from Edinburgh inventory,” Neuropsychologia 9(1), 97–113 (1971). Columbia University and is now a professor of psychiatry and neuro- 17. J. A. Noah et al., “fMRI validation of fNIRS measurements during a biology, Yale School of Medicine, and a professor of neuroscience, naturalistic task,” J. Visualized Exp. 100, e52116 (2015). University College London. She is also the director of the Brain 18. A. Tachibana et al., “Parietal and temporal activity during a multimodal Function Laboratory at Yale University. Her research is focused on dance video game: an fNIRS study,” Neurosci. Lett. 503(2), 125–130 investigations of neural circuitry that underlies human–social inter- (2011). actions using multimodal neuroimaging techniques, including fNIRS, 19. G. H. Klem et al., “The ten-twenty electrode system of the International fMRI, EEG, eye-tracking, and behavioral measures. Prior to recruit- Federation,” Electroencephalogr. Clin. Neurophysiol. 52(3), 3–6 ment to Yale, she was a director of the fMRI research center at (1999). Columbia University. Neurophotonics 011006-14 Jan–Mar 2018 Vol. 5(1)
Journal
Neurophotonics – SPIE
Published: Jan 1, 2018