FALSE POSITIVES IN FUNCTIONAL NEAR-
Ilias Tachtsidis1, Terence S. Leung1, Anchal Chopra1, Peck H. Koh1,
Caroline B. Reid1, and Clare E. Elwell1
Abstract: Functional cranial near-infrared spectroscopy (NIRS) has been widely used to
investigate the haemodynamic changes which occur in response to functional activation.
The technique exploits the different absorption spectra of oxy- and deoxy-haemoglobin
([HbO2] [HHb]) in the near-infrared region to measure the changes in oxygenation and
haemodynamics in the cortical tissue. The aim of this study was to use an optical
topography system to produce topographic maps of the haemodynamic response of both
frontal cortex (FC) and motor cortex (MC) during anagram solving while simultaneously
monitoring the systemic physiology (mean blood pressure, heart rate, scalp flux). A total
of 22 young healthy adults were studied. The activation paradigm comprised of 4-, 6- and
8- letter anagrams. 12 channels of the optical topography system were positioned over the
FC and 12 channels over the MC. During the task 12 subjects demonstrated a significant
change in at least one systemic variable (p≤0.05). Statistical analysis of task-related
changes in [HbO2] and [HHb], based on a Student’s t-test was insufficient to distinguish
between cortical haemodynamic activation and systemic interference. This lead to false
positive haemodynamic maps of activation. It is therefore necessary to use statistical
testing that incorporates the systemic changes that occur during brain activation.
When analysing cerebral haemodynamic activation data using functional
neuroimaging the task-specific activation observed is due to the existence of a close
coupling between regional changes in neuronal activation, brain tissue metabolism and
regional changes in cerebral blood flow (CBF). Cranial functional near-infrared
spectroscopy (NIRS) has been widely used to investigate the haemodynamic changes,
which occur in response to functional activation of specific regions of the cerebral cortex.
The technique exploits the different absorption spectra of oxy-haemoglobin (HbO2) and
1 Department of Medical Physics and Bioengineering, Malet Place Engineering Building,
University College London, Gower Street, London WC1E 6BT, UK
I. TACHTSIDIS ET AL.
deoxy-haemoglobin (HHb) in the near-infrared region to measure the changes in
oxygenation and haemodynamics in the brain cortical tissue. In order for this response to
be monitored unambiguously it is important that the haemodynamic task-related activity
is occurring on top of an unchanged global systemic and brain resting state.
We have previously reported that significant changes in mean blood pressure (MBP)
and heart rate (HR) occur during anagram activation tasks and observed that NIRS
haemodynamic changes were in some volunteers significantly correlated with changes in
these systemic variables.1 Most recently,2 we reported that during a frontal lobe anagram
activation task, task-related haemodynamic changes were observed both over the frontal
cortex (activated region) and motor cortex (control region). The task-related changes
were correlated with increases in MBP and scalp blood flow (flux) measured with laser
Doppler. This implies the possibility of some systemic “global interference” in our NIRS
measured data. It is possible that the anagram task elicits an emotional response, which
produces changes in blood pressure that are likely to cause passive changes in the scalp
blood flow. These changes can produce small task-related, but non cortical alterations in
the [HbO2] and [HHb] signals as measured by cranial NIRS.
Over the last decade or so, many studies have been published describing the use of
the optical topography (OT) technique to map functional brain activation.3-5 By making
simultaneous NIRS measurements at multiple brain sites, one can produce spatial maps
of the haemoglobin concentration changes that correspond to specific regions of the
cerebral cortex. OT can therefore potentially discriminate between regional activated
cortical areas and global haemodynamic changes.
The aim of this study is to investigate the functional haemodynamic changes during
frontal lobe anagram activation using optical topography both over the activated and
control area while continuously monitoring systemic and scalp blood flow changes.
2. MATERIAL AND METHODS
This study was approved by the UCL Research Ethics Committee. We studied 22
young healthy subjects with English as their first language (15 male, 7 female, median
age 22 years, range 20-39).
NIRS measurements were conducted with the ETG-100 Optical Topography System
(Hitachi Medical Co., Japan) using two 12-channel arrays. Each optode array consisted of
5 source optodes (each delivering light at 780 and 830 nm) and 4 detector optodes. The
source-detector interoptode spacing was 30mm and data were acquired at 10Hz. The
optodes were placed over the subject’s left frontal cortex and positioned according to the
international 10-20 system of electrode placement such that channels 1-12 were centred
approximately over the frontopolar region (Fp) and channels 13-24 were centred
approximately over the left primary motor cortex (C3). A schematic illustration of optode
placement is show in Figure 1.
A Portapres® system (TNO Institute of Applied Physics) was used to continuously
and non-invasively measure MBP and HR from the finger. A laser Doppler probe
(FloLab, Moore Instruments) was placed over the forehead to monitor the changes in
scalp blood flow (flux).
FALSE POSITIVES IN FUNCTIONAL NEAR-INFRARED TOPOGRAPHY 3
Figure 1. A schematic diagram illustrating the approximate positions of the NIRS light sources, detectors and
locations of corresponding measuring positions/channels. One array was centred on the frontopolar region (Fp),
the other on the left motor cortex (C3).
All the volunteers were positioned in a comfortable sitting position. Data were
recorded during two minutes of the subject at rest (baseline), followed by 45 seconds of
the subject solving 4-letter anagrams (9 anagrams, 5 seconds per anagram), 30 seconds
rest, 45 seconds of solving 6-letter anagrams (5 anagrams, 9 seconds per anagram), 30
seconds of rest, 45 seconds of solving 8-letter anagram (5 anagrams, 9 seconds per
anagram) and 30 seconds of rest. Each anagram-solving period was repeated a total of
four times, with the study ending after a 2-minute rest period (total study time 19
minutes). In this study solving an anagram was defined as producing one coherent word
using only the letters from another word (e.g. icon – coin). Subjects were encouraged to
solve as many anagrams as possible and were instructed to say possible solutions out loud
All optical data were subjected to an identical processing procedure using the
functional Optical Signal Analysis program6 (fOSA, University College London, UK) to
convert the relative changes in light intensities to concentration changes in haemoglobin
(HbO2, HHb and their sum, HbT) using a differential pathlength factor correction of 6.26.
All the signals including MBP, HR and flux, were then decimated from 10Hz to 1Hz and
low pass filtered at 0.08Hz. The data were filtered using a 5th order low pass Butterworth
digital filter in forward backward directions to avoid introducing a phase delay. The last
pre-processing stage, prior to statistical analysis was to de-trend the time-course to
remove both drift introduced by the system and any slowly changing unrelated
physiological signals. A first-order linear baseline was drawn as the reference and then
subsequently subtracted from the activation signal.
The response to stimulation was calculated for each subject as the difference
between the average of 10 seconds worth of baseline data at the end of the rest period,
and the average of 10 seconds of data commencing 15 seconds after the onset of the 4, 6
or 8 letter anagram solving periods respectively. A ‘Student’s t-test’ was used to assess
the significance of these responses (the threshold of significance was set at p≤0.05 from
baseline). For the optical topographic data we then calculated the cumulative total
number of channels across subjects in which we observed activation. We define
activation as a statistical significant increase in [HbO2], a statistical significant decrease
or no change in [HHb] and a statistical significant increase in [HbT]. Systemic
interference was measured by using the Pearson correlation model to calculate
correlations between the systemic variables and changes in [HbO2] and [HHb] in all of
the OT channels.
I. TACHTSIDIS ET AL.
A summary of the activation data for the whole group is shown in Figure 2. Each
paradigm is shown separately and data are normalised to the number of valid channels.
Across paradigms similar activation response was observed in both frontal cortex and
motor cortex. Channels in which the highest number of subjects showed activation were
channel 23 (55.56%) for the 4-letter task, channel 1 (52.94%) for the 6-letter task, and
channels 6 and 21 (33.33%) for the 8-letter task. Taking into account all of the tasks, an
average of 30% of the subjects showed activation (range 25-35%) on the frontal cortex
and 27% (range 17-37%) on the motor cortex.
Analysis of the systemic variables show that at least 50% of the subjects
demonstrated a change in at least one systemic variable. Table 1 shows the mean changes
in each systemic variable for those subjects that showed a significant change.
Correlation analysis of the NIRS and systemic data shows a large variability across
different OT channels and across subjects. Figure 3 shows the results of the correlation
analysis between MBP and the NIRS data, across all channels for (a) subject 3 who
showed generally high correlations (r>0.5), and (b) subject 18 who showed generally
low correlations (r<0.5). Both subjects showed significant changes in systemic variables
during the anagram tasks and both subjects had channels that showed activation. This
trend was observed across subjects. The correlation between the systemic data and the
NIRS data from the frontal cortex channels show no difference from the correlation
between the systemic data and the NIRS data from the motor cortex channels.
Table 1. Group changes from rest to activation are presented as mean ± standard
deviation for those subjects that demonstrated a significant change.
Systemic Variables 4-letter task
(n=4) 14.3±31.1 (n=3) 20.3±10.3 (n=1) -17.8
In this study we used an optical topography system to investigate the changes in
[HbO2] and [HHb] during anagram solving over the frontal lobe (activated area) and
motor cortex (control area) while simultaneously monitoring systemic variables. We used
a Student’s t-test to define significant changes in [HbO2], [HHb] and [HbT] for each OT
channel and for each subject during the different anagram solving tasks and used these
data to define where and when activation was detected. The same analysis was performed
on the systemic variables. We observed a large variability in activated OT channels
across subjects. The OT results failed to define specific regional areas of activation. 50%
of subjects showed a significant change in at least one systemic variable. These systemic
changes appear in some subjects to correlate with the observed functional changes in
[HbO2] and [HHb] across the OT channels. Figure 4 shows an example of changes in
[HbO2] and [HHb] from an OT channel over the frontal cortex and an OT channel over
the motor cortex with the simultaneously recorded changes in MBP and scalp flux.
FALSE POSITIVES IN FUNCTIONAL NEAR-INFRARED TOPOGRAPHY 5
Clearly systemic interference during the anagram task can lead to false positives in
defining activated OT channels.
Figure 2. Group analysis shows the percentage of subjects that demonstrated activation in specific channels
during the three different anagram solving paradigms.
Scale of number of
I. TACHTSIDIS ET AL.
Figure 3. Individual correlation coefficients between MBP and ∆[HbO2] and MBP and ∆[HHb] across all
channels for (a) subject 3 and (b) subject 18.
In this study we used the classical approach to define significant changes in
haemoglobin concentrations by employing a “Student’s t-test”. This approach compares
two different states of the brain, i.e. “rest” versus “activation”. The “rest” period is
usually defined as a baseline period before the stimulus onset and the “activation” period
is defined as the period 10-20 seconds after the onset of the stimulus. By keeping the rest
and activation periods constant across subjects one can investigate the functional
response to specific tasks. Whilst a simplistic approach of this kind helps to provide a
quick assessment of the haemodynamic response to the task it does not consider any
spatial coherence in the OT data. It also assumes that the measured changes in
haemoglobin concentrations are due solely to the neuronal activation, and that there are
no tasks-related systemic effects. We have shown that this latter assumption is not true
for all subjects performing an anagram solving task. One can include a priori information
regarding systemic changes and can de-correlate the physiological noise (cardiac,
respiratory and vasomotion related fluctuations) from the evoked haemodynamic
response, by using techniques such as Principal Component Analysis,7 Independent
Component Analysis,8 and more recently Statistical Parametric Mapping (SPM).6 SPM
has been widely used for the analysis of functional activation data from other
neuroimaging modalities such as the BOLD response in fMRI studies.9 SPM uses mass
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
MBP and ∆ ∆[HbO2]
MBP and ∆ ∆[HHb]
12345678 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
FALSE POSITIVES IN FUNCTIONAL NEAR-INFRARED TOPOGRAPHY 7
univariate approach to modelling the spatiotemporal neuroimaging data by assigning a
statistic value to every brain voxel. It enables the construction of spatial statistical
processes to test hypotheses about regional specific effects in the brain. Unlike the
classical approach mentioned earlier, where the two different time courses compared,
SPM employs a modelling approach for each brain voxel. In our study all of the
explanatory variables (HbO2, MBP, HR and flux) were treated as regressors in the linear
model. To treat the variability of haemodynamic responses arising from different events
between different brain voxels, SPM allows the modelling of latency and dispersion
derivatives as additional regressors to its canonical response function. The associated
parameter estimates are the coefficients for each of the regressors that best model the
observed response for the voxel in question (here a voxel is defined as an OT channel).
To account for the spatial coherence of the functional data, SPM provides the necessary
family-wise correction based on the theory of Gaussian random field to resolve the
multiple comparison problem.
Figure 4. Results from one subject showing changes in; (a) NIRS data from channel 3 (frontal cortex) (b) NIRS
data from channel 15 (motor cortex); (c) MBP and (d) scalp flux.
As an example of this method we have used fOSA-SPM software6 to analyse NIRS
and systemic data from one subject collected during the 6-letter anagram solving task.
Using the “Student’s t-test” analysis, this subject demonstrated activation across all OT
channels. Figure 5 shows the results of the SPM analysis on the same subject’s data.
These are presented as an SPM t-result for the HbO2 signal over all channels and show a
spatial localisation of the haemodynamic response. Unlike the “Student’s t-test” approach
which compares the difference between two specific physiological states, SPM offers a
more rigorous approach to analysing functional OT data by taking into account the global
Time (seconds) Time (seconds)
Concentrations (µ µM)
Concentrations (µ µM)
Mean Blood Pressure (mmHg)
Scalp Flux (a.u.)
4-Letters 6-Letters 8-Letters
I. TACHTSIDIS ET AL.
systemic effects by means of fitting a haemodynamic response function and performing
spatial correlations across all channels.
Figure 5. The SPM t-results with a threshold of significance of p≤0.05; darker pixels correspond to higher
In conclusion, when analysing OT data for evidence of functional activation the
effect of task-related changes in systemic variables should be taken into account. SPM
may be a useful tool for analysing simultaneously measured multi-channel OT NIRS data
and systemic variables.
The authors would like to acknowledge the EPSRC (Grant No EP/D060982/1).
1. I. Tachtsidis, T.S. Leung, L. Devoto, D.T. Delpy, and C.E. Elwell, Measurement of frontal lobe functional
activation and related systemic effects: a near-infrared spectroscopy investigation, Adv. Exp. Med. Biol. In
2. I. Tachtsidis, T.S. Leung, M.M. Tisdall, D. Presheena, M. Smith, D.T. Delpy, and C.E. Elwell,
Investigation of frontal cortex, motor cortex and systemic haemodynamic changes during anagram solving,
Adv. Exp. Med. Biol. In Press (2008).
3. Y. Hoshi, B. H. Tsou, V. A. Billock, M. Tanosaki, Y. Iguchi, M. Shimada, T. Shinba, Y. Yamada, and I.
Oda, Spatiotemporal characteristics of hemodynamic changes in the human lateral prefrontal cortex
during working memory tasks, NeuroImage 20(3), 1493-1504 (2003).
4. B. Chance, S. Nioka, S. Sadi, and C. Li, Oxygenation and blood concentration changes in human subject
prefrontal activation by anagram solutions, Adv. Exp. Med. Biol. 510, 397-401 (2003).
5. R.P. Kennan, D. Kim, A. Maki, H. Koizumi, and R.T. Constable, Non-invasive assessment of language
lateralization by transcranial near infrared optical topography and functional MRI, Hum. Brain Mapp.
16(3), 183-189 (2002).
6. P.H. Koh, D.E. Glaser, G. Flandin, S. Kiebel, B. Butterworth, A. Maki, D.T. Delpy, and C.E. Elwell,
Functional optical signal analysis (fOSA): a software tool for NIRS data processing incorporating
statistical parametric mapping (SPM), JBO In Press (2007).
7. X. Zhang, V. Toronov, and A. Webb, Simultaneous integrated diffuse optical tomography and functional
magnetic resonance imaging of the human brain, Opt. Express 13(14), 5513-5521 (2005).
8. I. Schiessl, M. Stetter, J.E.W. Mayhew, N. McLoughlin, J.S. Lund, and K. Obermayer, Blind signal
separation from optical imaging recordings with extended spatial decorrelation, IEEE Transactions on
Biomedical Engineering, 47(5), 573-577 (2000).
9. K.J. Friston, A.P. Holmes, J.B. Poline, P.J. Grasby, S.C. Williams, R.S. Frackowiak, and R. Turner,
Analysis of fMRI time-series revisited, NeuroImage 2(1), 45-53 (1995).