Neural Correlates of Cognitive Workload and Anesthetic Depth:
fNIR Spectroscopy Investigation in Humans
A Thesis
Submitted to the Faculty
of
Drexel University
by
Kurtulus Izzetoglu
In partial fulfillment of the
requirements for the degree
of
Doctor of Philosophy
July 2008
ii
© Copyright 2008 Kurtulus Izzetoglu. All Rights Reserved.
iii
ACKNOWLEDGMENTS
This research work would not have been possible without the support of a great number
of people whose contributions to this thesis deserve special mention. It is a pleasure to
convey my gratitude to them for all their invaluable assistance, support and guidance.
First, I would like to thank my teacher, supervisor and mentor, Dr. Banu Onaral, for her
supervision, guidance and inspiration throughout this research. During times when I
lacked confidence and experienced self-doubt about my performance, she provided extra
support and motivation I needed to continue through the doctoral studies. She has always
been there for her students to support and teach in many ways and exceptionally inspire
and enrich our growth as a student, a researcher and more importantly as an human being
for a better society. I always feel blessed and special just by becoming one of her
students. My words in this acknowledgment will never be enough to thank and express
my gratitude to Dr. Banu Onaral.
I would like to extend my gratitude to Dr. Scott C. Bunce, who has been my co-advisor
for this research work. I am much indebted to Dr. Bunce for his valuable advice and
supervision from very early stage of this research. He constantly educated me
particularly in the field of neuroscience, provided continuous help, constructive criticism
and guided the novel ideas and analysis which made this research possible.
iv
Many thanks go in particular to Dr. Jay Horrow, Dr. Patricia A Shewokis and Dr. Kambiz
Pourrezaei. Dr. Jay Horrow guided and advised the anesthetic depth assessment studies.
His continuous support and help was essential in this research. I also would like to
acknowledge and express my gratitude to Dr. Jerry Levitt and Christopher Hoffman for
recruiting patients, collecting patients’ data and fNIR measurements for the anesthetic
depth assessment study. Christopher Hoffman was the study coordinator and handled all
aspects of human experimentation.
Dr. Patricia Shewokis provided a great deal of advice and guidance in the thesis outline
and supervision in all statistical analyses. Furthermore, she spent her precious times to
read this thesis and provided critical revisions and comments. I have always benefited
from Dr. Kambiz Pourrezaei’s advice and guidance, particularly in the area of fNIR
technology development. I sincerely thank Drs. Andres Kriete, Frank J. Lee, Jay Horrow,
Patricia Shewokis, and Kambiz Pourrezaei for serving in my thesis committee. Their
valuable time and input are greatly appreciated.
I gratefully thank and recognize my teammates at Optical Brain Imaging Lab: Hasan
Ayaz, Anna Merzagora, Frank Kepics who have always provided enormous support and
help more than one could ask for. I am thankful to Hasan Ayaz who developed the fNIR
data acquisition program and provided continuous technical support throughout this
research work. I also thank Frank Kepics who delivered the fNIR systems and made sure
that the systems complied with safety requirements for the operating room settings. I also
would like to thank Caryn Glaser, Lisa Williams, Natalia Broz, Steve Detofsky and Aylin
v
Sagay for their great help in making my life at School of Biomedical Engineering easier
and enjoyable.
I would like to thank my wife, Dr. Meltem Izzetoglu for her love, support and confidence
in me. In every step of this research, not only she encouraged me she also provided a
great deal of advice and help for devising the major thesis statements. I greatly appreciate
her guidance and help in signal processing algorithm developments. Special thanks to my
mother, father, my mother-in-law and father-in-law for being supportive and for taking
care of our daughters during this research.
This work has been sponsored in part by funds from the Defense Advanced Research
Projects Agency (DARPA) Augmented Cognition Program (AugCog), the Office of
Naval Research (ONR), Homeland Security and Wallace H. Coulter Foundation under
agreement numbers N00014-02-1-0524, N00014-01-1-0986 and N00014-04-1-0119. I
also acknowledge with thanks Mark St. John, David Kobus, Jeff Morrison, Gary
Kollmorgen and their colleagues at PSE, SPAWAR and BMH Inc. for organizing and
hosting the TIE sessions for the human performance assessment study; Gunay Yurtsever,
Ajit Devaraj and Frank Kepics for their hardware and software support in helping to
collect the fNIR data during the AugCog TIE sessions.
vi
TABLE OF CONTENTS
LIST OF TABLES ............................................................................................................. ix
LIST OF FIGURES ............................................................................................................ x
ABSTRACT ..................................................................................................................... xiii
CHAPTER 1: INTRODUCTION ....................................................................................... 1
1.1 Motivation ............................................................................................................ 1
1.1.1 Problem Statement of Human Performance Assessment .............................. 3
1.1.2 Problem Statement of Anesthetic Depth Assessment ................................... 4
1.2 Contributions ........................................................................................................ 5
1.3 Outline .................................................................................................................. 7
CHAPTER 2: fNIR SPECTROSCOPY.............................................................................. 8
2.1 Background .......................................................................................................... 8
2.1.1 Underlying Principles of fNIR in Brain Activity Assessment ..................... 9
2.1.2 CW fNIR System ........................................................................................ 14
2.2 The fNIR Studies in Working Memory and Attention ....................................... 18
2.2.1 Working Memory........................................................................................ 18
2.2.2 Attention ..................................................................................................... 21
CHAPTER 3: HUMAN PERFORMANCE ASSESSMENT ........................................... 23
3.1 Introduction ........................................................................................................ 23
vii
3.2 Objective, Hypothesis and Specific Aims .......................................................... 24
3.3 Methods .............................................................................................................. 25
3.3.1 Participants .................................................................................................. 25
3.3.2 Warship Commander Task ......................................................................... 25
3.4 Analyses, Results and Discussion ...................................................................... 28
3.4.1 Task Load and Performance Analysis in the Presence or Absence of
Divided Attention: Specific Aim 1 & 2 ..................................................................... 29
3.4.2 Individual Participant Analysis ................................................................... 35
CHAPTER 4: EXPLORATORY STUDY ON ANESTHETIC DEPTH ASSESSMENT37
4.1 Introduction ........................................................................................................ 37
4.2 Objective, Hypothesis and Specific Aims .......................................................... 39
4.3 Methods .............................................................................................................. 40
4.3.1 Patient Population and Anesthetic Procedures ........................................... 40
4.3.2 Procedure and Signal Acquisition ............................................................... 40
4.3.3 Data Analysis .............................................................................................. 42
4.4 Results ................................................................................................................ 45
4.4.1 Evaluation of Deep and Light Anesthesia Stages ....................................... 45
CHAPTER 5: DISCUSSION AND FUTURE WORK .................................................... 53
5.1 Discussion .......................................................................................................... 53
5.2 Future Work ....................................................................................................... 56
viii
5.2.1 Future Studies for Applied Research .......................................................... 57
5.3 Conclusion .......................................................................................................... 59
LIST OF REFERENCES .................................................................................................. 60
APPENDIX A: ICA AND PCA TESTS ........................................................................... 66
VITA ................................................................................................................................. 72
ix
LIST OF TABLES
1. Effects of wave size on the fNIR Left: Mean correlations between number of airplanes and the fNIR sensor. .....................................................................................36
2. Effects of wave size on the fNIR Right: Mean correlations between number of airplanes and the fNIR sensor. .....................................................................................36
3. Repeated measures ANOVA results. The within subject factors of Stages (Deep or Light) and Channels (1 to 16) were crossed for the factorial repeated-measures design.. .........................................................................................................................46
4. Paired t-test results.. .....................................................................................................47
5. Paired t-test comparisons for Hbt (df=25).. .................................................................48
6. Paired t-test comparisons for oxy-Hb (df=25).. ...........................................................48
7. Paired t-test comparisons for deoxy-Hb (df=25)... ......................................................49
x
LIST OF FIGURES
1. Absorption spectrum in NIR window: spectra of oxy-Hb and deoxy-Hb in the range of 700 to 900 nm allows spectroscopy methods to assess oxy-Hb and deoxy-Hb concentrations, whereas water absorption becomes substantial above 900 nm, and thus majority of photons are mainly absorbed by water. ............................................ 11
2. a. Photons that enter the tissue undergoes two types of interaction; scattering due to cell membranes and absorption due to two main chromophores in optical window (oxy-Hb and deoxy-Hb).b. Source-detector separation and ‘banana-shaped’ photon migration in tissue. ...................................................................................................... 12
3. Overview of the Drexel’s fNIR system ...................................................................... 14
4. Block diagram of fNIR sensor system: The control box hosts analog filters and amplifies; data acquisition board (DAQ) is used for switching the LED light sources and detectors, which collect the reflected light. .......................................................... 15
5. Flexible Sensor housing 4 LED light sources, 10 photodetectors and wires. ............. 16
6. Illustration of fNIR probe design. ............................................................................... 16
7. Spatial map of the 16-channel fNIR sensor on the curved brain surface, frontal lobe...................................................................................................................................... 17
8. fNIR results for channels over middle frontal gyrus during an n-back task. The y-axis is the mean oxygenation change acquired from subjects who performed well and had over 90% correct hits. ................................................................................................. 20
9. fNIR result from a subject who performed poorly during 3-back condition and had less than 90% correct hits. .......................................................................................... 20
10. Task presentation of target categorization. ‘XXXXX’ is the target and ‘OOOOO’ is for the context stimuli. ................................................................................................ 21
xi
11. (a) fNIR data on channel 11; (b) Location of the significant difference in fNIR measurements between target and context .................................................................. 22
12. A snapshot during WCT where air warfare management required the user to monitor “waves” of incoming airplanes, to identify the identity of the unknown ones (yellow) as friendly (blue) or hostile (red), and to warn and then destroy hostile airplanes using rules of engagement. ................................................................................................... 27
13. Averaged oxygenation versus Wave Size (n=8) ......................................................... 29
14. fNIR (Left) Averaged Oxygenation Data (n=8). ....................................................... 30
15. fNIR (Right) Averaged Oxygenation Data (n=8). ...................................................... 30
16. Wavesize by 2nd Verbal Task for Oxygenation Change. ............................................ 31
17. Pearson’s correlation between performance and oxygenation change. ...................... 33
18. Mean oxygenation change as a function of Wavesize, 2nd Verbal Task, and average Percentage of Game Score in the 24-plane waves. ..................................................... 33
19. fNIR (Left) measurements versus RTIFF (n=8). The plot reflects the correlation between rate of change in left prefrontal oxygenation and performance measure RTIFF for each of the 12 waves of the scenario. ........................................................ 34
20. fNIR (Right) measurements versus RTIFF (n=8). The plot reflects the correlation between rate of change in right prefrontal oxygenation and performance measure RTIFF for each of the 12 waves of the scenario. ........................................................ 35
21. The anesthesia stages and procedure markers ............................................................ 41
22. Flowchart for hemodynamic response calculation and noise removal procedures. MBLL: Modified Beer-Lamber Law; I(P)CA: Independent (Principal) Component Analysis; Ref: Reference Signal, NIR: Near-Infrared ................................................ 45
xii
23. Rate of hemodynamic changes within deep and light anesthesia stages: The time versus fNIR measurements for (a) Hbt, (b) oxy-Hb, and (c) deoxy-HB are shown for deep and light anesthesia stages .................................................................................. 51
24. Deoxy-Hb changes within deep and light anesthesia stages: deoxy-Hb averages displayed very slow rate of change in deep anesthesia, whereas this rate of change is drastically increased when the patient emerges to wakefulness. ................................ 52
25. A new split sensor design for the anesthetic depth assessment studies. One-source, four-detector can be used to scan each hemisphere. ................................................... 58
26. Examples of raw intensity measurements where (a) there was a significant correlation with the reference measurement and (b) there was almost no correlation. ................. 69
27. Mean and standard deviation of the correlation between reference signal and the original intensity measurement (whether at 730 or 850nm) rorig and the estimated intensity measurement rest after applying the ICA and PCA algorithms over 310 cases with rorig>0.7................................................................................................................ 71
xiii
ABSTRACT Neural Correlates of Cognitive Workload and Anesthetic Depth:
fNIR Spectroscopy Investigation in Humans Kurtulus Izzetoglu
Banu Onaral, Ph.D., Scott C Bunce, Ph.D
Functional near-infrared spectroscopy (fNIR), a non-invasive neuroimaging modality
designed to monitor the hemodynamic change, can help identify neural correlates of
human brain functioning mediated by different events. In this thesis, fNIR has been used
to monitor prefrontal cortical activity with the primary objective to determine a set of
neurophysiological markers that detect changes in neural activation elicited by levels of
mental engagement. Two studies were selected to assess change in the cognitive state of
mental engagement at both high and low levels of neural activation. At the high end of
neural activation, human performance studies were conducted to assess cognitive
workload, in particular signal changes associated with overload. At the low end of
cognitive engagement, the capacity of fNIR to detect changes associate with the depth of
anesthesia was investigated using patients undergoing general anesthesia. In the human
performance study, participants were cognitively challenged by a complex task. By
contrast, in the anesthetic depth assessment study, cognitive activity was deliberately
suppressed by anesthetic agents. In both studies, neurophysiological markers of
hemodynamic changes were extracted from the fNIR measurements.
The hypothesis underlying the human performance study is the positive correlation of
blood oxygenation in the prefrontal cortex with increasing task difficulty and sustained
cognitive effort. In addition, increased blood oxygenation demonstrates a positive
xiv
relationship with behavioral performance measures in this task. A naval air warfare
management and control task with varying levels of difficulty has been chosen to test this
engagement condition. The results of this study showed that changes in blood
oxygenation in relevant areas of the prefrontal cortex are associated with increasing
cognitive workload, defined as sustained attention in a verbal and spatial working
memory and decision-making task. The results suggest a reliable, positive association
between cognitive workload and increases in cortical oxygenation responses (r=0.89 &
p<0.001). The data analysis also supports the hypothesis that the rate of oxygenation
change in the dorsolateral prefrontal cortex as measured by fNIR can provide an index of
sustained attention in a complex working memory and decision-making task.
Furthermore, this study reveals that an abrupt drop in the rate of oxygenation change in
dorsolateral prefrontal cortex under high workload conditions is associated with a user’s
decline in performance.
Awareness is an unintended mental state during general anesthesia. An accurate,
objective measure of return to consciousness would provide an important safeguard for
patients and physicians alike. This exploratory investigation on predicting awareness
under general anesthesia examines the hypothesis that the transition from deep to light
anesthetic stages is associated with reliable changes in oxygenated, deoxygenated, and
total hemoglobin in frontal cortex. Hemodynamic changes during deep and light
anesthesia were examined in 26 patients. The results suggest that the rate of
deoxygenated hemoglobin change can be used as a descriptive neuromarker to
differentiate between deep and light anesthesia stages (F1,25 = 7.61, p<0.01). This marker
xv
is proposed for further development as an index of the depth of anesthesia for the purpose
of monitoring awareness under general anesthesia.
In addition to the neuropsychological findings, this research demonstrates engineering
and signal processing solutions in the form of customized algorithms and procedures that
allow fNIR to measure usable signals under field conditions. Independent and principal
component analyses (ICA, PCA) were combined in a novel procedure that employed dark
current (i.e., signal from non-cortical sources) as a reference measurement. This method
provided improved signal-to-noise ratio for the hemodynamic measurements acquired in
the operating room, and can be used to increase the signal quality of fNIR for many other
applications and field situations.
1
CHAPTER 1: INTRODUCTION 1.1 Motivation
Human brain imaging techniques such as functional magnetic resonance imaging (fMRI)
and positron emission tomography (PET) have dramatically increased our knowledge
about the neural circuits that underlie human cognitive, emotional, and motivational
processes. (Cabeza et al. 2000, Davidson et al. 1998). However, these techniques are
expensive, highly sensitive to motion artifact, confine the participants to restricted
positions, and may expose individuals to potentially harmful materials (PET) or loud
noises (fMRI). These characteristics make these imaging modalities unsuitable for many
uses, including the monitoring of ongoing cognitive activity under routine working
conditions.
Over the last twenty years, near-infrared spectroscopy (NIRS) based optical imaging
systems have been widely used in functional brain studies as a noninvasive tool to
monitor changes in the concentration of oxygenated hemoglobin (oxy-Hb) and
deoxygenated hemoglobin (deoxy-Hb) (Delpy, 1988; Patterson, Chance and Wilson,
1989; Chance et al. 1993; Chance et al. 1998; Villringer et al. 1997; Obrig et al. 1997;
Boas et al. 2001; Izzetoglu K, et al., 2004). Based on the NIRS technique, the Drexel
Optical Brain Imaging team has developed a functional brain monitoring prototype,
called fNIR. The fNIR is a portable, safe, affordable and negligibly intrusive monitoring
system, which enables the study of cortical cognition-related hemodynamic changes in
various field conditions.
2
Neural activity has a direct relation with hemodynamic changes in the brain (Kruggel,
1999). Research on brain-energy metabolism has elucidated the close link between
hemodynamic and neural activity (Magistretti, 2000). Traditional neuroimaging
techniques, such as fMRI cannot be used to measure these hemodynamics for a variety of
real-life applications that could yield important discoveries and lead to novel uses. By
contrast, the fNIR system can be deployed to assess hemodynamic responses and help
understand human brain activation by providing neurophysiological markers derived
from neural responses to different experimental settings under field conditions. However,
the research should be conducted to establish the validity of the fNIR signal as well as to
demonstrate the acquisition of accurate and viable signals under real-life conditions. For
this reason, two experimental settings are selected to:
i) assess the high and the low ends of mental engagement: For the highest level
of activation, cortical areas are monitored while the participants are engaged
and overloaded with a complex task. By contrast, in an anesthetic depth
assessment study, cognitive activity is deliberately suppressed by anesthetic
agents. In both assessments, neurophysiological markers of hemodynamic
changes are extracted from the fNIR measurements.
ii) validate the fNIR signals under different field conditions: To advance the
fNIR technology into the application areas, it is critical to comprehend and
formulate unmet needs and field requirements for real time operation in
clinical settings. Examples include level of anesthesia monitoring in an
operating room or outside the laboratory settings such as ground control of
3
unmanned aerial vehicle (UAV). The research in this thesis is the first attempt
to deploy this emerging fNIR device as well as to tailor custom noise removal
and information processing algorithms to both human performance assessment
and anesthetic depth assessment studies.
1.1.1 Problem Statement of Human Performance Assessment
While performing informationally-dense tasks, frontal cortical areas that support
executive functions (attention, working memory, response monitoring) are activated. The
increase in difficulty of the task (task overload) may overwhelm the human brain. This
would lead to operator failure of task execution. Hence, this would have an effect on, for
instance, the safety of the operation. As an example, UAV ground operators are required
to perform more within high cognitively-demanding tasks. This workload increases both
the information processing and decision demands on operators. The report by Leduc et al
(U.S. Army Aeromedical Research Laboratory) indicates that 60-80 percent of non-
combat aircraft losses can be attributed to human error, and a large percentage of the
remaining losses have human error as a contributing factor (Leduc et al, 2005). Another
example is the military operations that require mental alertness and full engagement with
the task. To sustain and enhance soldier’s performance specifically when operating
advanced military equipment in a more complex environment, cognitive activities,
attentional and control resources, should be monitored in response to increases in
workload. Furthermore, the fNIR technology and workload assessment can be an
important tool during space flights to maintain safe and effective performance of
astronauts who need to perform high levels of cognitive information-processing.
4
1.1.2 Problem Statement of Anesthetic Depth Assessment
Awareness under general anesthesia is a rare condition that occurs when surgical patients
become conscious –or awake– and can recall events or conversations that happen in the
operation room. (Sigalovsky 2003, Pollard et al. 2007, Bruhn et al. 2006). Less
commonly noted is excruciating pain which may accompany unintended awareness
during general anesthesia. When using other kinds of anesthesia, such as local or regional
anesthesia with sedation, it is expected that patients may have some recollection of the
procedure which is neither expected nor desired under general anesthesia. Studies report
that approximately 0.13% of anesthesia awareness incidences at medical institutions,
predicting 26,000 cases yearly in the United States (Sigalovsky 2003, Sebel et al. 2004)
occurred during general anesthesia. Despite this is rare event, even one incident is
important to the clinician who recognizes that this can be a distressing or traumatic
experience for the patient and can leave a lifetime of residual emotional and
psychological problems ranging from sleep disorders, daytime anxiety and post-traumatic
stress disorder (Sigalovsky 2003, Avidan et al. 2008, Bruhn et al. 2006).
Prevention of awareness during anesthesia in a non-paralyzed patient requires only an
alert anesthesia provider. There has been interest in developing sophisticated and
automatic depth of anesthesia monitors that can continuously and reliably monitor the
anesthesia state during a surgical procedure. To date, such devices have largely been
based on the measurement of electrophysiological signals such as electrocardiographic
(ECG) signals, electroencephalographic (EEG) signals, auditory and somatosensory
evoked potentials, and craniofacial electromographic (EMG) signals. All of the above
5
biophysiological parameters involve measurable electric currents, whereas the fNIR
technology uses biological parameters (hemodynamic variables) which are not mediated
by measurable electric currents. A monitoring apparatus for objectively and continuously
quantifying the depth of anesthesia of a patient during surgery would not only reduce the
incidence of awareness, but also would allow the anesthesiologist to administer the
minimal dose of anesthetic required to achieve the desired depth of anesthesia.
Furthermore, since most of the opioids have direct effect on hemodynamic response
through neuro-vascular coupling, it would be highly advantageous to have a device and a
method for monitoring a patient's depth of anesthesia based on hemodynamic changes
rather than based on standard electrophysiological techniques.
1.2 Contributions
The present research extensively utilizes the fNIR technology which is being developed
by Drexel’s Optical Brain Imaging team in collaboration with Dr. Britton Chance. Dr.
Chance is the inventor of the modality and Drexel team has focused on new technology
development and deployment. The study presented in this thesis yields contributions to
the theoretical and applied research. Identification of changes and markers in neural
activation elicited by levels of mental engagement has been the theoretical outcome of
the research. The contributions to applied research can be detailed as:
i. A novel application of fNIR spectroscopy to human performance assessment is
performed through a feasibility study. During a complex and realistic cognitive
task based on naval air warfare management scenarios, physiological measures
6
based on fNIR are evaluated to predict changes in cognitive workload. The
research indicates that there is a positive correlation with performance measures
and blood oxygenation in the dorsolateral prefrontal cortex is very significant
where blood oxygenation has increased with increase in task difficulty and
workload provided that the participant was attentive and engaged in the task.
Furthermore, effects of divided attention via multi-tasking are studied.
ii. Feasibility of fNIR spectroscopy in the assessment of deep and light anesthesia in
the operating room during surgical procedures is demonstrated. During this
exploratory study, the hemodynamic measures obtained by the fNIR, total
hemoglobin, oxygenated and deoxygenated hemoglobin are analyzed to extract a
robust descriptor that differentiates between light and deep anesthesia stages.
iii. New techniques based on independent component analysis (ICA) and principal
component analysis (PCA) for the removal of non-physiological artifacts such as
motion artifact, table tilting, equipment or cable noise, and so forth are proposed
and tested. The primary novelty lies in the use of a reference measurement,
collected during dark current conditions, and the use of combined ICA and PCA
in artifact removal. The study in this thesis demonstrates that proposed techniques
can remove the artifacts accurately with no additional hardware requirements. In
addition, since the amount of time it takes to run an algorithm is negligible, ICA
and PCA techniques can be used to enhance signal quality under other field
situations and natural environments.
7
1.3 Outline
The five chapters discuss fNIR spectroscopy and its select applications in detail. Chapter
2 introduces background and principles of the technology being used in two basic
cognitive tasks, working memory and attention. Chapter 3 includes the discussion on the
human performance assessment study which is the high level of engagement. Specific
aims of the research and how they are achieved along with the results are presented in
detail. Chapter 4 introduces the exploratory study on anesthetic depth assessment. The
chapter reports the analyses and performance of total hemoglobin, oxygenated and
deoxygenated hemoglobin measurements in quantifying transition from deep anesthesia
to light anesthesia. The thesis is concluded by Chapter 5 which includes general
discussion, future work, and a conclusion. This last chapter also provides envisioned
future application areas.
8
CHAPTER 2: fNIR SPECTROSCOPY 2.1 Background
Since its first proposal (Jobsis, 1977), near infrared spectroscopy (NIRS) has been
increasingly applied for the noninvasive measurement of changes in the relative ratios of
deoxygenated hemoglobin (deoxy-Hb) and oxygenated hemoglobin (oxy-Hb) during
brain activation. In the late1980s, Delpy designed and tested an NIRS instrument on
newborn heads in neonatal intensive care (Delpy et al, 1987). In the late 1980s and early
1990s, Dr. Britton Chance and his colleagues, using pico-second long laser pulses,
spearheaded the development of time-resolved spectroscopy techniques in an effort to
obtain quantitative information about the optical characteristics of the tissue (Patterson,
Chance and Wilson,1989). These efforts by Chance, Delpy (Cope & Delpy, 1988) and
others (Fishkin & Gratton E, 1993), expedited the translation of NIRS based techniques
into a neuroimaging modality for various cognitive studies (Okada et al., 1996; Villringer
et al. 1997; Obrig, 1997; Chance et al, 1998).
The combined efforts of these researchers led to the development of three distinct NIRS
implementations, namely time resolved spectroscopy (TRS), frequency domain and
continuous wave (CW) spectroscopy (Strangman et al. 2002). In TRS systems, extremely
short incident pulses of light are applied to tissue, and the temporal distributions of
photons, which carry information about tissue scattering and absorption, are measured. In
frequency domain systems, the light source is amplitude modulated to the frequencies in
the order of tens to hundreds of megahertz. The amplitude decay and phase shift of the
detected signal with respect to the incident are measured to characterize the optical
9
properties of tissue (Boas, 2002). In CW systems, light is continuously applied to tissue
at constant amplitude. The CW systems are limited to measuring the amplitude
attenuation of the incident light.
CW systems have a number of advantageous properties that have resulted in wide use by
researchers interested in brain imaging relative to other near-infrared systems; it is
minimally intrusive and portable, affordable, and easy to engineer relative to frequency
and time domain systems (Chance et al. 1998, Boas et al. 2002). These CW systems hold
enormous potential for research studies and clinical applications that require the
quantitative measurements of hemodynamic changes during brain activation under
ambulant conditions in natural environments.
2.1.1 Underlying Principles of fNIR in Brain Activity Assessment
Understanding the brain energy metabolism and associated neural activity is of
importance for realizing principles of fNIR in assessing brain activity. The brain has
small energy reserves and a great majority of the energy used by brain cells are for
processes that sustain physiological functioning (Ames III, 2004). Ames III reviewed the
studies on brain energy metabolism as related to function and reported that the oxygen
(O2) consumption of the rabbit vagus nerve increased 3.4-fold when it was stimulated at
10 Hz and O2 consumption in rabbit sympathetic ganglia increased 40% with stimulation
at 15 Hz. Furthermore, glucose utilization by various brain regions increased several fold
in response to physiological stimulation or in response to pharmacological agents that
affect physiological activity (Ames III, Brain Research Reviews, 2004). These studies
10
provide clear evidence that large changes occur in brain energy metabolism in response
to changes in activity. The levels of compounds involved in energy metabolism and
energy metabolites can be outlined as:
Brain cells consume energy when activated. Oxygen is required to metabolize the
glucose. The concentration of oxygen in brain is about 0.1 μmol g−1 (Hansen,
1987) of which 90% is in oxy-Hb in brain capillaries. This concentration can
support the normal oxygen consumption (about 3.5 μmol g−1 min−1) for 2 seconds.
(Ames III, 2004). For that reason, increase in neural activity in the brain is
followed by the rise in local cerebral blood flow (CBF) (Buxton, 2004).
Oxygen is transported to neural tissue via oxygenated hemoglobin (oxy-Hb) in the
blood.
The oxygen exchange occurs in the capillary beds.
As oxy-hemoglobin gives up oxygen, it is transformed into deoxygenated
hemoglobin (deoxy-Hb).
Local cerebral blood flow (CBF) increases much more than the cerebral metabolic
rate of oxygen (CMRO2), therefore local blood is more oxygenated and less
deoxy-Hb present (Buxton, 2004).
Based on this brain energy metabolism, methods and imaging modalities, such as fNIR
and fMRI (Kwong et al., 1992; Ogawa et al., 1990) for measurements of deoxy-Hb
and/or oxy-Hb are implemented to provide correlates of brain activity through oxygen
consumption by neurons.
11
Because oxy-Hb and deoxy-Hb have characteristic optical properties in the visible and
near-infrared light range, the change in concentration of these molecules during increase
in brain activation can be measured using optical methods. Most biological tissues are
relatively transparent to light in the near infrared range between 700-900 nm, largely
because water, a major component of most tissues, absorbs very little energy at these
wavelengths (Figure 2.1). Within this window the spectra of oxy- and deoxy-hemoglobin
are distinct enough to allow spectroscopy and measures of separate concentrations of
both oxy-Hb and deoxy-Hb molecules (Cope 1991). This spectral band is often referred
to as the ‘optical window’ for the non-invasive assessment of brain activation (Jobsis,
1977).
(HbO 2)
Figure 2.1: Absorption spectrum in NIR window: spectra of oxy-Hb and deoxy-Hb in the range of 700 to 900 nm allows spectroscopy methods to assess oxy-Hb and deoxy-Hb concentrations, whereas water absorption becomes substantial above 900 nm, and thus
majority of photons are mainly absorbed by water.
‘Optical Window’
12
Photons introduced at the scalp pass through layers of tissue and are absorbed and
scattered mainly by oxy-Hb and deoxy-Hb molecules. Because a predictable quantity of
photons follows a ‘banana-shaped’ path and leaves the tissue, these photons can be
measured using photodetectors, as illustrated in Figure 2.2 (Gratton, Maier, Fabiani,
Mantulinm, & Gratton, 1994).
a*. b. Figure 2.2: a. Photons that enter the tissue undergoes two types of interaction; scattering due to cell membranes and absorption due to two main chromophores in optical window
(oxy-Hb and deoxy-Hb) b. Source-detector separation and ‘banana-shaped’ photon migration in tissue.(*Courtesy of Christopher H. Contag, Stanford University; J. Harris & O. Levi, NanoBioConvergence)
If wavelengths are chosen to maximize the amount of absorption by oxy-Hb and deoxy-
Hb, changes in these chromophore concentrations cause alterations in the number of
absorbed photons as well as in the number of scattered photons that leave the scalp.
These changes in light intensity measured at the surface of the scalp are quantified
using a modified Beer–Lambert law, which is an empirical description of optical
attenuation in a highly scattering medium (Cope & Delpy, 1988; Cope, 1991). By
measuring absorbance/scattering changes at two (or more) wavelengths, one of which is
Source
Detector
Source
Detector
Source
Detector
Source
Detector
Scattering
Absorption
13
more sensitive to oxy-Hb, the other to deoxy-Hb, changes in the relative concentration of
these chromophores can be calculated. Using these principles, researchers have
demonstrated that it is possible to assess hemodynamic changes in response to brain
activity through the intact skull in adult human subjects (Chance, Zhuang, UnAh, Alter,
& Lipton 1993; Gratton, Corballis, Cho, Fabiani, & Hood, 1995; Hoshi & Tamura, 1993;
Kato, Kamei, Takashima, & Ozaki, 1993; Villringer, Planck, Hock, Schleinkofer, &
Dirnagl, 1993; Izzetoglu K., Bunce, et al., 2004)
Typically, an optical apparatus consists of a light source by which the tissue is radiated
and a light detector that receives light after it has interacted with the tissue. Photons that
enter tissue undergo two different types of interaction, namely absorption and scattering
(Figure 2.2.a). According to the modified Beer-Lambert Law (Cope 1991), the light
intensity after absorption and scattering of the biological tissue is expressed by the
equation:
I=GIoe-(
HBC
HB+
HBO2C
HBO2)*L (2.1)
where G is a factor that accounts for the measurement geometry and is assumed constant
when concentration changes. Io is input light intensity, HB and HBO2 are the molar
extinction coefficients of deoxy-Hb and oxy-Hb, CHB and CHBO2 are the concentrations of
chromophores, deoxy-Hb and oxy-Hb, respectively and L is the photon path which is a
function of absorption and scattering coefficients a and b .
By measuring optical density (OD) changes at two wavelengths, the relative change of
oxy- and deoxy-hemoglobin versus time can be obtained. If the intensity measurement at
14
an initial time is bΙ (baseline), and at another time is I , the OD change due to variation in
HBC and 2HBOC during that period is:
2210 HBOHBOHBHB
b CCI
IOD log (2.2)
Measurements performed at two different wavelengths allow the calculation of HBC and
2HBOC . Change in oxygenation and blood volume or total hemoglobin (Hbt) can then be
deduced:
HBHBO CCnOxygenatio 2
(2.3)
(2.4)
2.1.2 CW fNIR System
The fNIR system used in this study was originally described by Chance et al. (Chance et
al. 1993). The current generation, flexible headband sensor developed in the Drexel’s
Optical Brain Imaging laboratory, consists of 4 LED light sources and 10 detectors
Figure 2.3: Overview of the Drexel’s fNIR system
HBHBO CCeBloodVolum 2
15
Figure 2.4 below shows a block diagram of the CW fNIR sensor system to monitor brain
activity. The main components are the sensor that covers the entire forehead of the
participant, a control box for data acquisition, a power supply for the control box, and a
computer for the data analysis software. The communication between the data analysis
computer and the task presentation computer is established via serial port to time-lock the
fNIR measures to the task events.
Figure 2.4: Block diagram of fNIR sensor system: The control box hosts analog filters
and amplifies; data acquisition board (DAQ) is used for switching the LED light sources and detectors, which collect the reflected light.
The flexible sensor is a modular design consisting of two parts: a reusable, flexible circuit
board that carries the necessary infrared sources and detectors, and a disposable, single-
use cushioning material that serves to attach the sensor to the participant’s forehead (see
Figure 2.5). The flexible circuit provides a reliable integrated wiring solution, as well as
consistent and reproducible component spacing and alignment. Because the circuit board
and cushioning material are flexible, the components move and adapt to the various
contours of the participant’s head, thus allowing the sensor elements to maintain an
fNIR
Sensor
Data Acquisition Card
Computer for Task
Presentation
Amplifiers &
Analog Filters
Computer for Acquisition
& Analysis
via Serial/Parallel Port
Control Box
16
orthogonal orientation to the skin surface, dramatically improving light coupling
efficiency and signal strength.
Figure 2.5: Flexible Sensor housing 4 LED light sources, 10 photodetectors and wires.
The flexible sensor (Figure 2.5) covers the forehead using 16 channels (Figure 2.6), with
a source-detector separation of 2.5 cm. The light sources (manufactured by Epitex Inc.;
type no: L4X730/ 4X805/4X850-40Q96-I) contain 3 built in LEDs having peak
wavelengths at 730, 805, 850 nm with an overall outer diameter 9.2 ± 0.2 mm. The
photodetectors (manufactured by Bur Brown; type no: OPT101) are monolithic
photodiodes with single supply trans-impedance amplifier having the size of 0.90 X 0.90
inch (Bunce et al. 2006).
Figure 2.6: Illustration of fNIR probe design.
: source : detector
12
34
56
78
910
1112
1314
1516
Source DetectorSource Detector
17
The fNIR sensor, illustrated in Figures 2.5 and 2.6, reveal information in localizing brain
activity, particularly in dorsolateral prefrontal cortex. Figure 2.7 shows spatial map of the
16-channel fNIR sensor on the curved brain surface, frontal lobe.
Figure 2.7: Spatial map of the 16-channel fNIR sensor on the curved brain surface, frontal lobe.
Neuroimaging studies using positron emission tomography (PET) or functional magnetic
resonance imaging (fMRI) have shown that the prefrontal cortex is the structure that
helps humans to engage in executive process components of working memory and
attention functions (Smith & Jonides, 1997; McCarthy et al.1997). In an attempt to
validate the fNIR system in monitoring these executive functions, attention and working
18
memory studies were performed on healthy subjects by placing the fNIR sensor (Figure
2.5) on a subject’s forehead (Figure 2.7).
2.2 The fNIR Studies in Working Memory and Attention
The fNIR technology has been tested by carrying out studies using memory and attention
tasks (Izzetoglu et al. 2003b, Bunce et al. 2006, Izzetoglu M et al. 2007). During these
experiments in healthy subjects, the n-back task is used for working memory assessment
and an odd-ball paradigm is implemented for attention monitoring.
2.2.1 Working Memory
To assess working memory, a standard “n-back” task was used. The n-back task is a
sequential letter task with varied workload conditions that has frequently been used to
assess working memory by cognitive psychologists and neuroscientists (e.g., Braver et
al., 1997, Smith et al. 1997). In this study, the stimuli were single consonants presented
centrally, in pseudorandom sequences, on a computer monitor. Similar to Smith’s task
design (Smith & Jonides, 2007), the stimulus duration was 500 ms, with a 2500-ms
interstimulus interval. Four conditions were used to incrementally vary working memory
load from zero to three items. In the 0-back condition, subjects responded to a single pre-
specified target letter (e.g., ‘‘X’’) with their right hand by pressing a button to identify the
stimulus. In the 1-back condition, the target was defined as any letter identical to the one
immediately preceding it (i.e., one trial back). In the 2-back and 3-back conditions, the
targets were defined as any letter that was identical to the one presented two or three
trials back, respectively. Subjects were asked to press one button for target stimuli
19
(approximately 33% of trials) and another button for non-target stimuli. This strategy
incrementally increased working memory load from the 0-back to the 3-back condition.
The brain hemodynamic responses for each of the 20 single trials were obtained within
each n-back block (the order across subjects was counterbalanced) using the
hemodynamic model and fNIR oxygenation data were measured. Peak amplitudes were
extracted and averaged for each subject, workload condition, and channel. The results for
channels over the middle frontal gyrus were in agreement with fMRI data suggesting that
increased working memory workload is associated with increased oxygenation in the
dorsolateral prefrontal cortex (Braver et al., 1997, Smith et al. 1997) (see Figure 2.8).
However, more importantly, there were individual differences at the highest level of
working memory (the 3-back condition). Participants who continued to work hard and
perform well (over 90% correct hits) in the 3-back condition exhibited greater
oxygenation in the 3-back condition than the 2-back condition. However, for participants
who lost their concentration during the 3-back condition and performed poorly (less than
90% correct hits), a significant decrease in their performance was observed, as their
oxygenation levels dropped (see Figure 2.9).
20
Figure 2.8: fNIR results for channels over middle frontal gyrus during an n-back task. The y-axis is the mean oxygenation change acquired from subjects who performed well
and had over 90% correct hits.
Figure 2.9: fNIR result from a subject who performed poorly during 3-back condition and had less than 90% correct hits.
Oxy
gena
tion
Cha
nge
N-Back (n=0,1,2,3)
2.2
2.4
2.6
2.8
3
Sample Subject 1
0 Back
1 Back
2 Back
3 Back
Oxy
gen
atio
n C
han
ge
2.2
2.4
2.6
2.8
3
Sample Subject 1
0 Back
1 Back
2 Back
3 Back
Oxy
gen
atio
n C
han
ge
21
2.2.2 Attention
The protocol used to measure attention is a common visual oddball paradigm modified
for fMRI by McCarthy et al.1997 (Izzetoglu et al. 2003b, Bunce et al. 2006, Izzetoglu M
et al. 2007). The stimuli were two strings of white letters (XXXXX and OOOOO)
presented against the center of a dark background. A total of 516 stimuli were presented,
480 context stimuli (OOOOO) and 36 targets (XXXXX). Stimulus duration was 500ms,
with an interstimulus interval of 1500ms (Figure 2.10). Target stimuli were presented
randomly with respect to context stimuli with a minimum of 12 context stimuli between
successive targets to allow the hemodynamic response an opportunity to return to
baseline between target presentations.
Figure 2.10: Task presentation of target categorization. ‘XXXXX’ is the target and ‘OOOOO’ is for the context stimuli.
O O O OO O O O
X X X XO O O O
O O O OO O O O
500 ms
1500 ms
Target
22
Fifteen right-handed participants (4 females and 11 males with average age 20.8±4.2
years) were required to press one of two buttons on a response pad after each stimulus,
while both fNIR and EEG were recorded simultaneously. One button was pressed in
response to targets (X’s), and another button was pressed in response to context stimuli
(O’s).
Repeated-measures ANOVA computed on the fNIR oxygenation data revealed that
oxygenation values were greater in response to targets than to controls in channel 11,
located over middle frontal gyrus of the right hemisphere (see Figure 2.11.a).
Differentiation occurred between 6 and 9s post-stimulus (see Figure 2.11.b). These
results are consistent with the fMRI literature for visual target categorization with respect
to increased oxygenation in response to targets, cortical location, and time course
(McCarthy et al.1997, Ardekani et al. 2002).
(a) (b)
Figure 2.11: (a) fNIR data on channel 11; (b) Location of the significant difference in
fNIR measurements between target and context
23
CHAPTER 3: HUMAN PERFORMANCE ASSESSMENT 3.1 Introduction
This chapter describes the deployment and statistical analysis of fNIR measurements
acquired during a cognitive workload assessment study which was designed to
investigate transitions during high levels of mental engagement. At this level of
engagement, ability of the fNIR sensor to assess cognitive workload is evaluated while
the participants are cognitively challenged by performing a complex task. The central
hypothesis is that blood oxygenation in the prefrontal cortex would show a positive
correlation with increasing sustained cognitive effort as related to task difficulty. In
addition, increased blood oxygenation is expected to have a positive correlation with
behavioral performance measures in this task.
Monitoring the cognitive and the physiological state of the user, especially during
sensitive and cognitively demanding operations are critical for successful task execution.
Because the fNIR technology allows the design of portable, safe, affordable, and
negligibly intrusive monitoring systems, cortical monitoring of brain hemodynamics via
the fNIR can be of value in assessing the cognitive state of the user during mentally
demanding operations.
The experimental protocol, namely ‘Warship Commander Task’ (WCT), was developed
by Pacific Science & Engineering Group under the direction of Space and Naval Warfare
Systems Center - San Diego. The WCT is designed as a cognitive multitasking
environment to manipulate workload while simulating actual military commands and
24
tasks (St. John et al. 2002). In the WCT, various levels of task difficulty and task load can
be manipulated in the presence or absence of an auditory memory task.
3.2 Objective, Hypothesis and Specific Aims
The primary objective of this study is to use physiological measures based on fNIR to
predict changes in cognitive workload during a complex cognitive task. The principal
hypothesis, derived from preliminary data on a working memory task (the n-back task as
explained in section 2.2.1 (Izzetoglu K et al. 2003a, Izzetoglu M et al 2005a), is:
Blood oxygenation in the dorsolateral prefrontal cortex, as assessed by fNIR, would
increase with increasing task difficulty and sustained cognitive effort and would
demonstrate a positive relationship with performance measures.
Therefore we expect blood oxygenation to increase with increasing workload as long as
the participant was attentive and engaged in the task. However, at the point when the
task became too difficult, and the participant lost attention or engagement in the task, the
blood oxygenation would decrease. Similar to the Yerkes-Dodson law (Yerkes &
Dodson, 1908), blood oxygenation is expected to rise with increasing workload to the
point of overload, after which it would decline. This hypothesis is consistent with
preliminary data using fNIR in the n-back task. Participants in the 3-back condition
reported losing their concentration in this taxing working memory task, which tended to
coincide with a decrease in blood oxygenation.
25
The specific aims to achieve the objective are as follows:
1) Establish the relationship between cognitive workload, the participant’s
performance, and changes in blood oxygenation levels within the dorsolateral
prefrontal cortex.
2) Determine the effect of divided attention as manipulated by the secondary
component of the WCT (the auditory task).
3.3 Methods
3.3.1 Participants
A total of eight healthy participants, three females and five males, ranging in age from 18
– 50 years, participated in the study (arranged and organized by the Defense Advanced
Research Projects Agency (DARPA) Augmented Cognition Program). The study was
conducted at Pacific Science & Engineering in San Diego, CA (Izzetoglu K, 2004). The
participants, recruited by Pacific Science & Engineering, had a range of experience with
the WCT, ranging from 8 hours to 300 hours. Prior to the study, all participants signed
informed consent statements approved by the Human Subjects Institutional Review
Board at Drexel University.
3.3.2 Warship Commander Task
This study is based on data collected during the Augmented Cognition - Technical
Integration Experiment session. The experimental protocol for this session used a
complex task resembling a videogame called the Warship Commander Task (WCT). The
WCT was designed to approximate naval air warfare management. The WCT has been
26
described as a quasi-realistic, ship-based navy command and control environment task
that requires spatial and verbal working memory and decision-making processes (St. John
et al. 2002). The version of the WCT employed in the current experiment used was
comprised of two component tasks: Air Warfare Management and the Ship Status (SS)
task. Air Warfare Management required the user to monitor “waves” of incoming
airplanes, identify them as friendly or hostile, and to warn and then destroy hostile
airplanes using rules of engagement. Each wave lasted 75 seconds. The level of
cognitive effort in the Air Warfare Management component based on task difficulty and
task load is manipulated by 1) increasing the number of planes per wave (6, 12, 18 or 24
planes) that had to be managed at a given time, and 2) changing the proportion of planes
whose identity was unknown, i.e., whether hostile or friendly compared to known.
Airplanes with unknown status require more decision and processing time, and therefore
make the task more difficult. The WCT has two levels of difficulty relative to the
proportion of unknown planes, low in which 33% of the planes are unknown, and high, in
which 67% of the planes are unknown. A snapshot during WCT is shown in Figure 3.1.
27
Figure 3.1: A snapshot during WCT where air warfare management required the user to monitor “waves” of incoming airplanes, to identify the identity of the unknown ones
(yellow) as friendly (blue) or hostile (red), and to warn and then destroy hostile airplanes using rules of engagement.
A third task load factor used to manipulate cognitive workload is the presence or absence
of a secondary verbal task labeled the Ship Status (SS) Task. The SS task is comprised of
auditory messages containing information about the ship and its operation (encoding),
and periodic queries, or “recall” about earlier auditory messages. In this auditory task, the
participants are required to listen to sporadic auditory messages and memorize various
ship status data while answering periodic queries that appeared on the computer screen.
The Air Warfare Management task involves spatial and verbal working memory and
decision-making processes, whereas the Ship Status Task is primarily a verbal memory
task. When both tasks are operational, the WCT becomes a divided attention task.
For this study, each participant completed four sets of WCT. Each set was comprised of
12 waves, 3 repetitions of each wave size in the order of 6, 18, 12, and 24 planes. The
factors of Wavesize (6,12,18, or 24 planes), Complexity (high versus low percentage of
28
unknown airplanes), and full versus divided attention (secondary Ship Status Task On or
Off) were crossed to create a 4 x 2 x 2 factorial repeated-measures design.
Performance was assessed using two measures. Reaction Time to Identify Friend or Foe,
(RTIFF) indicated the time from when a plane appeared on the screen until the participant
selected the plane and pressed the identification button. The Percent Game Score
(PctGS) was calculated as the percentage of game points that a participant accumulated
during any given wave relative to the total game points that were possible for that wave.
Previous work on the WCT has demonstrated that RTIFF is a reasonable predictor of
cognitive task load and of overall performance which can be used as a behavioral
measure of participant workload (DARPA Report 2003).
3.4 Analyses, Results and Discussion
For each wave of 75 seconds, the rate of change in the oxygenation was calculated from
the fNIR measurements. For the purpose of these analyses, blood oxygenation values
were averaged across the 8 channels or pixels covering left and right hemispheres. To test
the primary hypothesis that increased workload would be associated with relative
increases in blood oxygenation in the frontal poll and dorsolateral prefrontal cortex, a 4 X
2 X 2 (Wave Size) x (High vs Low Complexity) x (Second Verbal Task) repeated
measures analysis of variance (ANOVA) was computed. To test the hypothesis that blood
oxygenation would predict performance, within-subject Pearson’s product-moment
correlations were computed between blood oxygenation levels and their RTIFF scores for
each wave.
29
3.4.1 Task Load and Performance Analysis in the Presence or Absence of Divided Attention: Specific Aim 1 & 2
In support of our primary hypothesis, the results indicated a main effect for wave size
across both hemispheres (F3,21 = 16.24, p<0.001). The analyses also revealed that the 6,
12, and 18-plane waves differed from each other; 12-plane > 6-plane, t62=2.29, p= 0.03;
18-plane > 12-plane, t62=3.52, p= 0.002. The 18- and 24-plane waves did not differ
(Figure 3.2) (Izzetoglu K et al. 2004).
Figure 3.2: Averaged oxygenation versus Wave Size (n=8)
Independent analyses of left (Fleft,3,21 = 14.87, p<0.01) and right (Fright,3,21 = 11.73,
p<0.01) hemispheres indicated that both left and right dorsolateral prefrontal cortex were
30
responsive to wave size (see Figures 3.3 and 3.4). No other main effect obtained
significance (all p-values > 0.10).
Figure 3.3: fNIR (Left) Averaged Oxygenation Data (n=8).
(Error bars: +/- 1 SE)
Figure 3.4: fNIR (Right) Averaged Oxygenation Data (n=8).
(Error bars: +/- 1 SE)
-20.00
-15.00
-10.00
-5.00
0.00
5.00
10.00
15.00
20.00
Low Task Load - No Audio
Low Task Load -Audio
High Task Load - No Audio
High Task Load - Audio
Oxy
gen
atio
n C
han
ge
WCT Scenarios
fNIR - Left ( n=8 )
6-Wave
12-Wave
18-Wave
24-Wave
-20.00
-15.00
-10.00
-5.00
0.00
5.00
10.00
15.00
20.00
Low Task Load - No Audio
Low Task Load -Audio
High Task Load - No Audio
High Task Load - Audio
Oxy
gen
atio
n C
han
ge
WCT Scenarios
fNIR - Right (n=8)
6-Wave
12-Wave
18-Wave
24-Wave
31
The main effect for Wavesize was qualified by one significant interaction between
Wavesize and the Secondary Verbal Task (F = 3.10, p<0.05); no other interaction obtain
conventional levels of significance (all p-values > 0.10). Post-hoc analyses of the
Wavesize x Secondary Verbal Task interaction revealed that the primary effect occurred
in the 24-plane wave. When the Secondary Verbal Task was off (the less difficult
condition), blood oxygenation exhibited a linear relationship with Wavesize. In contrast,
when the Secondary Verbal Task was on, blood oxygenation exhibited a quadratic
relationship with Wavesize, with the mean for the 24-plane wave dropping below that of
the 18-plane wave (see Figure 3.5).
Figure 3.5: Wavesize by 2nd Verbal Task for Oxygenation Change.
In line with the stated hypothesis, a preliminary interpretation of this finding was that a
number of participants had reached their maximal level of performance in this most
difficult level of the task, lost their concentration/effort in the task, and consequently their
blood oxygenation dropped. The hypothesis also predicts that individuals who were able
-8
-6
-4
-2
0
2
4
6
8
6 12 18 24
Wave Size
2nd Verbal High
2nd Verbal Of f
Oxy
gen
atio
n C
han
ge
32
to stay on task and continue to perform in this difficult workload condition should
demonstrate increased oxygenation relative to both 1) their own oxygenation levels in the
18-plane wave and 2) individuals who became overwhelmed and disengaged in the 24-
wave condition. Because sustained concentration and engagement in the task should
result in increased performance, a positive relationship between performance and blood
oxygenation would provide support for this interpretation. A Pearson’s product-moment
correlation indicated a very strong positive relationship between blood oxygenation and
performance, indexed by the Percentage Game Score, in the 24-plane condition
(Pearson’s r = 0.89, p = .003; see Figure 3.6). Given the bimodal distribution of the
scores, a median split on the Percentage Game Score, as depicted in Figure 3.6, provided
further evidence of the hypothesized relationship between cognitive effort and the blood
oxygenation response. As seen in Figure 3.7, the mean levels of oxygenation were higher
for both high and low performers in the 24-plane wave than the 18-plane wave when the
Secondary Verbal Task was off, the easier condition. However, when Secondary Verbal
Task was on, the more difficult condition, the individuals who performed well on the 24-
plane wave showed a higher mean level of oxygenation for the 24-plane wave than for
the 18-plane wave, whereas those who performed more poorly showed a decrease in
oxygenation relative to the 18-plane wave.
33
Figure 3.6: Pearson’s correlation between performance and oxygenation change.
Figure 3.7: Mean oxygenation change as a function of Wavesize, 2nd Verbal Task, and average Percentage of Game Score in the 24-plane waves.
-8
-6
-4
-2
0
2
4
6
8 10
12
6 12 18 24 6 12 18 24
Low Performance High
Oxy
gen
atio
n C
han
ge
2nd Verbal On 2nd Verbal Off
34
Performance measures computed for each wave, including RTIFF, warn and destroy
airplanes and total game scores (as a percentage of possible points) correlate directly with
workload levels (number of airplanes) and can be used as behavioral measures of
participants’ workload (St John 2002). A Pearson’s correlation between fNIR gauge
output and the RTIFF performance measure confirmed a positive relationship between
prefrontal blood oxygenation and performance across all conditions (Left Hemisphere:
Pearson’s r = 0.737; Right Hemisphere: Pearson’s r =0.758). These data are graphically
represented by wave size in Figure 3.8 and 3.9.
Figure 3.8: fNIR (Left) measurements versus RTIFF (n=8). The plot reflects the
correlation between rate of change in left prefrontal oxygenation and performance measure RTIFF for each of the 12 waves of the scenario.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
-15.00
-10.00
-5.00
0.00
5.00
10.00
15.00
6 18 12 24 6 18 12 24 6 18 12 24
RT
IFF
Oxy
gen
atio
n C
han
ge
Wave Size
fNIR vs RTIFFfNIR- LeftRTIFF
35
Figure 3.9: fNIR (Right) measurements versus RTIFF (n=8). The plot reflects the
correlation between rate of change in right prefrontal oxygenation and performance measure RTIFF for each of the 12 waves of the scenario.
3.4.2 Individual Participant Analysis
To measure the relationship between number of airplanes and the fNIR sensor values for
each participant, we calculated 48 mean values of the blood oxygenation change rates for
3 waves of 6, 12, 18 and 24 airplanes within each WCT session. These values were
analyzed using repeated measures Analysis of Variance (ANOVA) and the results are
presented in Table 3.1 and Table 3.2.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
-15.00
-10.00
-5.00
0.00
5.00
10.00
15.00
6 18 12 24 6 18 12 24 6 18 12 24
RT
IFF
Oxy
gen
atio
n C
han
ge
Wave Size
fNIR vs RTIFFfNIR- Right
RTIFF
36
Table 3.1. Effects of wave size on the Table 3.2. Effects of wave size on the
fNIR (Left) fNIR (Right)
Subjects F p rP1 14.929 0.000 0.621P2 19.980 0.000 0.672P3 3.155 0.034 0.363P4 1.931 0.138 0.306P5 5.258 0.030 0.428P6 0.312 0.817 -0.023P7 2.769 0.053 0.395P8 3.597 0.021 0.300
Subjects F p rP1 13.196 0.000 0.649P2 8.331 0.000 0.537P3 0.574 0.635 0.064P4 3.006 0.040 0.342P5 2.527 0.070 0.308P6 1.151 0.339 0.169P7 1.491 0.230 0.290P8 2.152 0.107 0.281
* r: Mean correlations between number of airplanes and the fNIR sensor.
The individual analysis in tables 3.1 and 3.2 suggests that the fNIR gauge is significantly
sensitive to number of airplanes per wave in at least one hemisphere for 7 participants.
For participant P6, both left and right fNIR measurements failed. The correlation between
fNIR measurements and number of airplanes per wave is provided in the third column of
tables 2.1, and 2.2 designated by r. Two individual participants (P1 and P2) show
significantly strong correlations (Pearson’s r > 0.6, p<0.01).
37
CHAPTER 4: EXPLORATORY STUDY ON ANESTHETIC DEPTH ASSESSMENT
4.1 Introduction
This chapter provides an exploratory study and the fNIR data analysis to compare
neurophysiological markers of hemodynamic changes in response to deep and light
anesthetic depth. In contrast to the human performance assessment study presented in
Chapter 3, ability of the fNIR sensor to differentiate between deep and light anesthesia
stages is investigated while the patients underwent general anesthesia and cognitive
activity was suppressed by anesthetic agents. The primary hypothesis for this preliminary
study is that the hemodynamic response is a sensitive measure of anesthetic depth, in
particular when the hemodynamic response changes during the transitioning from deep to
light anesthesia stages.
Anesthesia awareness occurs when anesthetic medication to maintain unconsciousness is
not sufficient (Sigalovsky 2003, Pollard et al. 2007, Bruhn et al. 2006). In rare instances,
technical failure (i.e., the equipment that delivers the anesthetic to your body
malfunctions) or human error (i.e., the anesthesiologist misjudges the amount of
medication needed to keep you unconscious) may contribute to unexpected episodes of
awareness. When awareness during general anesthesia occurs, it is usually immediately
before the anesthetic takes effect or as the patient is emerging from anesthesia. In rare
instances, it can occur during the surgery when insufficient medication for the
maintenance of deep anesthesia is administered. Awareness can range from brief, hazy
recollections to some specific awareness of surroundings during surgery, and in some
38
cases to feeling pain (American Society of Anesthesiologists [ASA] Report, 2005).
During general anesthesia, clinicians use multiple ways to ensure that the patient receives
sufficient amount of anesthetic medication and remain unconscious by relying on their
clinical experience, training and judgment combined with continuous patient monitoring.
There has been interest in developing sophisticated and automatic depth of anesthesia
monitors that can continuously and reliably monitor the anesthesia state during a surgical
procedure. To date, such devices have largely been based on the measurement of
electrophysiological signals such as electrocardiographic (ECG) signals,
electroencephalographic (EEG) signals, auditory and somatosensory evoked potentials,
and craniofacial electromographic (EMG) signals. In the clinical setting, there are two
currently available devices to assess the anesthesia awareness: the Bispectrum Index
(BIS®)monitor (Aspect Medical Systems, Norwood, MA) and the A-line Auditory
Evoked Potential Index (AAI, Danmeter, Odense, Denmark). The BIS® utilizes a
proprietary algorithm that combines measurements of patient movement with the
amplitude and phase information from Fourier-transformed electroencephalogram epochs
to a number between 0 and 100, with 40-60 indicating the target for awareness prevention
(Heinke et al. 2005, Sigalovsky 2003, Avidan et al 2008, Ebert 2005, Singh 1999,Bruhn
et al. 2006). In contrast, the AAI measures changes in cortical auditory evoked potentials.
While each device appropriately tracks increasing depth from sedative class drugs and
volatile anesthetic agents, both devices underestimate depth of anesthesia when
adjunctive opioids are administered during general anesthesia (Manyam, et al, 2007).
39
The fNIR spectroscopy detects the hemodynamic changes in the cerebral cortex. It is
safe, portable and rugged which makes it suitable for applications in the operating room.
As shown by BIS, PET and fMRI studies, neuronal activity is inhibited and oxygen
consumption reduced by the administration of anesthetic drugs and hence brain
oxygenation is increased during anesthesia (Heinke et al. 2005, Alkire et al 1995, Kaisti
et al. 2005, Morgan et al. 1996) This effect can be acquired by fNIR measures in terms of
oxy-Hb, deoxy-Hb and total hemoglobin (Hbt).
A limited number of fNIR spectroscopy studies in anesthesia have been reported in the
literature (Lovell et al. 1999, Iwasaki et al. 2003). These fNIR studies primarily
investigate the effects of different types of anesthetic drugs i.e. propofol, etomidate,
sevoflurane, etc. on cerebral metabolism and hemodynamics (Lovell et al. 1999, Iwasaki
et al. 2003). Currently marketed cerebral oximetry devices (FORE-SIGHT®, CAS
Medical Systems, Branford, CT and INVOS®, Somanetics, Troy, MI) also employ fNIR
technology to make inferences about cerebral blood flow; neither company has directed
signal processing efforts to elucidate inferences about anesthesia depth. These oximetry
devices, FORE-SIGHT® and INVOS® are known to elicit physiological changes rather
than the changes linked to or associated with cognitive activity.
4.2 Objective, Hypothesis and Specific Aims
The primary objective of this exploratory study is to analyze the fNIR measures to
determine neuromarker(s) that can differentiate deep and light anesthesia stages. The long
40
term goal is to deploy fNIR in the operating room to further aid the clinician in the
detection of awareness under general anesthesia.
To achieve the primary objective, the specific aims are to:
1) Identify hemodynamic measures and robust marker(s), such as total hemoglobin,
oxygenated or deoxygenated hemoglobin that are capable of differentiating
between deep and light anesthesia stages
2) Obtain robust fNIR measurements that can be continuously monitored in real time
within highly noisy environments such as the operating room
4.3 Methods
4.3.1 Patient Population and Anesthetic Procedures
The analyses were performed on 26 patients for this exploratory study. The study was
conducted at Drexel University College of Medicine. Prior to the study, all participants
signed informed consent statements during their preoperative visit with the
anesthesiologist using a form approved by the Human Subjects Institutional Review
Board at Drexel University.
4.3.2 Procedure and Signal Acquisition
In the operating room, patient routine monitors for surgery were positioned. The fNIR
sensor was placed and a preliminary signal was obtained prior to administration of any
medication. Induction of anesthesia using mainly intravenous propofol occurred once a
41
satisfactory fNIR baseline signal had been achieved and recorded. The fNIR signal was
recorded continuously beginning one minute prior to injection. Intraoperative data
included times of anesthetic induction, first surgical incision, and wound closure as well
as administration of medication including intravenous drug doses, such as Fentanyl,
Propofol, etc. and inhalational drugs, such as Sevoflurane and Desflurane. Upon
achieving end-tidal inhalational agent concentrations below 0.1 MAC, patients were
asked to follow two separate simple commands (“open your eyes” and “squeeze my
fingers”) every 30 seconds, and their response and emergence from anesthesia were
continuously recorded during acquisition of the fNIR signal.
Figure 4.1: The anesthesia stages and procedure markers
Deep anesthesia was defined as the 4 minute time interval prior to wound closure (Figure
4.1). Light anesthesia was defined as the 4 minute time interval prior to eye opening. The
Procedure Markers
‘Wake’
…………………….fNIR Recording…..............…………………..…………………….
fNIR 4-minute 4-minute
placed… interval interval
Baseline selected selected
recorded for ‘deep’ for ‘light’
anesthesia anesthesia
Wound ClosureInduction ‘Eye Opening’Incision
Maintenance‘Deep’ Anesthesia
Emergence‘Light’ Anesthesia
Procedure Markers
‘Wake’
…………………….fNIR Recording…..............…………………..…………………….
fNIR 4-minute 4-minute
placed… interval interval
Baseline selected selected
recorded for ‘deep’ for ‘light’
anesthesia anesthesia
Wound ClosureInduction ‘Eye Opening’Incision
Maintenance‘Deep’ Anesthesia
Emergence‘Light’ Anesthesia
42
baseline fNIR signal was acquired before the induction, as described above. These
baseline values were used to calculate relative values of the hemodynamic responses.
4.3.3 Data Analysis
4.3.3.1 Hemodynamic Response Calculation and Noise Removal Procedures
fNIR measurements can be corrupted by a number of different physiological and/or non-
physiological noise sources such as respiration and heart pulsation signals, equipment
noise, motion artifact. To obtain more robust and reliable outcomes, these artifacts need
to be removed from the signal of interest. A set of noise removal procedures was
performed on the raw light intensity measurements before extracting the hemodynamic
signals using the modified Beer-Lambert Law for analysis, and then for comparison
between deep and light anesthetic states.
For the hemodynamic response calculations and noise removal algorithms (see flowchart
in Figure 4.2), the following procedures were implemented:
i) The light sources in fNIR measurements have three built in LEDs (730, 805 and
850nm) to allow for spectroscopic measurements.
ii) Calculations for hemodynamic signals are carried out using the measurements
obtained from two of the three wavelengths; 730 nm which is associated with
absorption due to deoxy-hemoglobin, and 850 nm which is associated with
absorption due to oxy-hemoglobin.
iii) 805nm is used in the dark-current condition, i.e., no light is emitted at this
wavelength, and therefore the detector only captures signals that are generated by
43
noise within that local region. Since the noise is local, it provides reliable
information about potential contaminates of the neural signal.
iv) Both independent component analysis (ICA) and principal component analysis
(PCA) are implemented to identify and to extract the measured correlated noise.
(The ICA and PCA algorithms as well as their performances are explained in
Appendix B)
v) To increase robustness and performance, the correlation between the dark current
measurement and the recordings obtained at two wavelengths during the real-time
measurement are calculated.
vi) The correlation threshold is set to ( r = 0.7) and the reference signal intensity should
be above the dark current level for the application of ICA and PCA algorithms for
noise removal.
vii) The method (ICA or PCA) that performs better is selected for automatic removal of
all artifacts.
a. ICA and PCA Approach (Appendix B): For each wavelength measurement
either at 730nm or 850nm, a separate ICA and PCA algorithm is performed
where the measurement vector x(t) is two dimensional containing the
measurement obtained either at 730 or at 850nm and the one at the dark
current.
b. The unknown independent/uncorrelated source signals s(t) that is estimated by
the ICA/PCA algorithm is also two dimensional which will be the non-
physiological noise signal and the clean raw intensity signal related with
hemodynamic changes.
44
c. Once the unknown A matrix and the unknown source signals s(t) are
extracted, the noise signal is selected by correlating the independent
components with the reference signal separately and selecting the one analysis
giving the highest correlation. After the selection of the independent
component corresponding to the noise signal, the noise is removed from the
original 730nm or 850 nm recordings by subtracting it from that measurement
with an appropriate amount found from the estimated A matrix.
d. Correlation between the dark current measurement and the noise removed
intensity measurements via ICA and PCA algorithms are calculated
separately. The outcome of the best performing algorithm that generates
lowest correlation is selected as the cleaned raw intensity measurement.
45
Figure 4.2: Flowchart for hemodynamic response calculation and noise removal
procedures. MBLL: Modified Beer-Lamber Law; I(P)CA: Independent (Principal) Component Analysis; Ref: Reference Signal, NIR: Near-Infrared
4.4 Results
4.4.1 Evaluation of Deep and Light Anesthesia Stages
To test the primary hypothesis, two periods were chosen to represent deep and light
anesthetic stages; four minutes prior to wound closure was chosen for deep anesthesia,
and four minutes prior to eye opening was chosen to assess light anesthesia and the return
Execute Signal Level &
CorrelationCheck
Calculate Hemodynamics via
MBLL
Apply ICA and PCA
Iref > dark level & r > 0.7?
No
Yes
No
The noise component is removed from 730nm and
850nm
Execute Algorithm Selection
r[ICA,Ref] < r[PCA, Ref]
No Calculate Hemodynamics via
MBLL[PCA]
Yes [ICA]
Calculate Hemodynamics via
MBLL
Reference signal should be higher than 400 AND its correlation with 730 and 850 nm should be high
Acquire 730nm, 850nm and Reference NIR Intensities
46
to conscious awareness (Figure 4.1). The 4-minute epochs were extracted from the raw
intensity measurements and cleaned from physiological and non-physiological artifacts
following the procedures described in 4.3.3.1. Relative to the fNIR baseline measures,
recorded prior to the induction, these measurements at 730 nm and 850 nm were used to
calculate oxy-Hb, deoxy-Hb and Hbt (oxy-Hb+deoxy-Hb). The Hbt, oxy-Hb and deoxy-
Hb values were calculated for deep and light anesthetic stages, which are defined in 4.3.2
based on the described markers, i.e., wound closure and eye-opening, respectively. These
values are plotted for each minute within the corresponding anesthesia stage.
The ability of three dependent measures, oxy-Hb, deoxy-Hb, and Hbt to differentiate
between deep and light stages of anesthesia were tested using 2 (Stage; Deep vs Light) x
16 (Channels) repeated measures ANOVAs. As tabulated in Table 4.1, the main effect
for anesthetic state was significant for deoxy-Hb (F1,25=7.61, p=0.010). This effect did
not interact with channel, indicating a consistent effect across the imaging area.
Inspection of the means indicated that light anesthesia was associated with relatively less
deoxy-Hb than deep anesthesia. The main effect for oxy-Hb was not significant, and Hbt
showed a trend that was driven primarily by the deoxy-Hb findings (see Table 4.1).
Table 4.1: Repeated measures ANOVA results df F-Ratio P value
Hbt 1,25 2.76 0.109
Oxy-Hb 1,25 0.53 0.471
Deoxy-Hb 1,25 7.61 0.010*
47
Because this was an exploratory study to determine if fNIR spectroscopy could be a
useful monitor for anesthetic depth in the operating room, we refined these analyses in
two ways: First, we used paired t-tests in the post-hoc analyses to provide a map of the
channels that were carrying the effects, as is characteristic in fMRI studies. Second, we
examined changes in deoxy-Hb in each of the four minutes leading up to eye-opening to
determine with greater specificity the predictive ability of hemodynamic measure to
determine return to consciousness.
Paired t-tests revealed that there was a predominance of right hemisphere activation that
differentiated between deep and light anesthesia stages (Table 4.3, 4.4. and 4.5). Channel
12 and 4 (see Table 4.2), which underlie attentional processes, were found to provide the
best separation of anesthetic stages, along with Channels 14 and 15. Channel 12 was also
found to be most significant region in the attention study summarized in section 2.2.2.
Table 4.2: Paired t-test results df T p value
Hbt 25 3.64 p<0.01
Oxy-Hb 25 1.87 p< 0.05
Deoxy-Hb 25 4.42 p<0.01
48
Table 4.3: Paired t-test comparisons for Hbt (df=25)
Significantly different channels for
Hbt
Table 4.4: Paired t-test comparisons for oxy-Hb (df=25)
Significantly different channels for oxy-Hb
Channels *: significant
Deep vs. Light Anesthesia (1 min prior)
1 t=1.44, p>0.05 2 t=-0.30, p>0.05 3 * t=1.97, p<0.05 4 t=1.61, p>0.05 5 t=0.09, p>0.05 6 * t=2.24, p<0.05 7 t=1.09, p>0.05 8 t=-0.29, p>0.05 9 t=0.32, p>0.05 10 t=-0.35, p>0.05 11 * t=1.76, p<0.05 12 * t=3.64, p<0.01 13 * t=1.82, p<0.05 14 * t=2.59 p<0.01 15 * t=2.33, p<0.05 16 * t=2.25, p<0.05
Channels *: significant
Deep vs. Light Anesthesia (1 min prior)
1 * t=1.74, p<0.05 2 t=-0.93, p>0.05 3 t=0.24, p>0.05 4 t=-0.38, p>0.05 5 t=-0.42, p>0.05 6 t=-0.04, p>0.05 7 t=-0.61, p>0.05 8 t=-0.72, p>0.05 9 t=-1.65, p>0.05 10 t=-0.74, p>0.05 11 t=0.25, p>0.05 12 * t=1.87, p<0.05 13 t=0.09, p>0.05 14 t=0.46, p>0.05 15 t=-0.88, p>0.05 16 t=0.41, p>0.05
49
Table 4.5: Paired t-test comparisons for deoxy-Hb (df=25)
Significantly different channels for deoxy-Hb
fNIR sensor locations that differentiated light from deep anesthesia (Table 4.2) were used
to investigate the time course for the rate of change in light vs. deep anesthesia. An
optimal measure of awareness for use in anesthesia would allow some capacity to predict
potential awakening with sufficient time to adjust the anesthetic agents to prevent
awareness. To examine this potential, the time versus hemodynamic responses for Hbt,
oxy-Hb and deoxy-HB within deep and light anesthesia stages is shown in Figures 4.3.a,
4.3.b, and 4.3.c, respectively. During deep anesthesia, deoxy-Hb averages (4.3.c)
displayed a very slow rate of change (3.4%). In contrast, as the patient emerges to
wakefulness, this rate of change increases drastically (48.8%). This change far exceeded
values determined in Hbt and oxy-Hb. Comparison study to determine rate of change
between deep (1 minute before wound closure) and light anesthesia times from 1 minute
to 4 minute prior to eye-opening revealed 293, 210 170 and 164 percent changes,
Channels *: significant
Deep vs. Light Anesthesia (1 min prior)
1 t=-0.99, p>0.05 2 t=1.22, p>0.05 3 t=1.36, p>0.05 4 * t=4.24, p<0.01 5 t=0.40, p>0.05 6 * t=3.66, p<0.01 7 * t=1.94, p<0.05 8 t=1.30, p>0.05 9 t=1.36, p>0.05 10 t=0.32, p>0.05 11 * t=1.92, p<0.05 12 * t=4.42, p<0.01 13 * t=1.98, p<0.05 14 * t=4.21, p<0.01 15 * t=4.15, p<0.01 16 * t=2.91, p<0.01
50
respectively (see Figure 4.4). Two (Stage, Deep vs Light) x 4 (4, 3, 2, and 1 minutes prior
to wound closure or eye opening) repeated measures ANOVAs revealed these changes
were significant for two channels (Channel 4: F3,75 = 5.10, p=.003; Channel 15: F3,75 =
3.62, p=.017) and Channel 12 showed a trend in the same direction. As plotted in
Figures 4.3.c and 4.4, deoxy-Hb decreased across the 4 minutes of light anesthesia, but
did not across the deep anesthesia. Post-hoc analysis for the light anesthetic state alone
showed that all three channels showed this same pattern of decreasing deoxy-Hb
(Channel 4: F3,75 = 3.93, p=.012; Channel 12: F3,75= 2.67, p=.053; Channel 15: F3,75=
3.32, p=.024).
a. Hbt
Hbt Changes in Channel 12
0.000
1.000
2.000
3.000
4.000
5.000
6.000
7.000
8.000
9.000
X-4min X-3min X-2min X-1min X Y-4min Y-3min Y-2min Y-1min Y
Time
mic
roM
ola
r
Wound ClosureEye Opening
………..
7.0% {
15.2% {Deep Anesthesia
Light Anesthesia
Hbt Changes in Channel 12
0.000
1.000
2.000
3.000
4.000
5.000
6.000
7.000
8.000
9.000
X-4min X-3min X-2min X-1min X Y-4min Y-3min Y-2min Y-1min Y
Time
mic
roM
ola
r
Wound ClosureEye Opening
………..
7.0% {
15.2% {Deep Anesthesia
Light Anesthesia
51
b. oxy-Hb
c. deoxy-Hb
Figure 4.3: Rate of hemodynamic changes within deep and light anesthesia stages: The
time versus fNIR measurements for (a) Hbt, (b) oxy-Hb, and (c) deoxy-HB are shown for deep and light anesthesia stages
oxy-Hb Changes in Channel 12
0.000
1.000
2.000
3.000
4.000
5.000
6.000
X-4min X-3min X-2min X-1min X Y-4min Y-3min Y-2min Y-1min Y
Time
mic
roM
ola
rWound Closure
Eye Opening
………..
9.7% {
7.3% {
Deep Anesthesia Light Anesthesia
oxy-Hb Changes in Channel 12
0.000
1.000
2.000
3.000
4.000
5.000
6.000
X-4min X-3min X-2min X-1min X Y-4min Y-3min Y-2min Y-1min Y
Time
mic
roM
ola
rWound Closure
Eye Opening
………..
9.7% {
7.3% {
Deep Anesthesia Light Anesthesia
deoxy-Hb Changes in Channel 12
0.000
0.500
1.000
1.500
2.000
2.500
3.000
3.500
X-4min X-3min X-2min X-1min X Y-4min Y-3min Y-2min Y-1min Y
Time
mic
roM
ola
r
Wound Closure Eye Opening
………..
3.4% {
48.8%{Deep Anesthesia Light Anesthesia
deoxy-Hb Changes in Channel 12
0.000
0.500
1.000
1.500
2.000
2.500
3.000
3.500
X-4min X-3min X-2min X-1min X Y-4min Y-3min Y-2min Y-1min Y
Time
mic
roM
ola
r
Wound Closure Eye Opening
………..
3.4% {
48.8%{Deep Anesthesia Light Anesthesia
52
Figure 4.4: Deoxy-Hb changes within deep and light anesthesia stages: deoxy-Hb averages displayed very slow rate of change in deep anesthesia, whereas this rate of
change is drastically increased when the patient emerges to wakefulness.
Research shows that inhalant anesthetics such as sevoflurane and desflurane have direct
cerebral vasodilatory effects and increase cerebral blood flow (CBF; Kaisti et al. 2003,
Bedforth et al. 2000, Marcar et al. 2006, Heath et al 1997, Matta et al. 1999). Increases in
CBF are generally followed by increases in cerebral blood volume (CBV; Morgan et al.
1996, Kaisti et al. 2003). Therefore, excessive amount of Hbt, oxy-Hb and deoxy-Hb
during deep anesthesia can be explained by ‘luxury perfusion,’ a combination of the
decrease in neuronal metabolic demand coupled with an increase in CBF (metabolic
supply; Heath et al. 1997, Duffy et al. 2000). Decreases in the rate of deoxy-Hb changes
during deep anesthesia are ascribed to cerebral metabolic rate (demand) suppression by
the administered anesthetic agents (Figure 4.3.c; Matta et al, 1995; Matta et al, 1999).
53
CHAPTER 5: DISCUSSION AND FUTURE WORK 5.1 Discussion
fNIR is a wearable neuroimaging device that enables the continuous, non-invasive, and
portable monitoring of changes in hemodynamic responses during human brain function.
The purpose of the first study in this thesis is to study high level of mental engagement
by using the fNIR as a gauge of cognitive workload in a complex, realistic, command and
control task. Changes in blood oxygenation in relevant areas of the prefrontal cortex are
associated with increasing cognitive workload, defined as attention in a verbal and spatial
working memory and decision making task. The results suggest a reliable, positive
association between cognitive workload and increases in the oxygenation responses.
The main hypothesis underlying the human performance study is that blood oxygenation
in the prefrontal cortex, as assessed by fNIR, would rise with increasing task load and
would exhibit a positive correlation with performance measures. Results indicate that the
rate of change in blood oxygenation is significantly sensitive to task load changes and
correlated with performance variables (r=0.89, p<0.001). The results support our main
hypothesis and suggest that fNIR can provide an index of sustained attention in a
complex working memory and decision making task. The results of this study further
indicated that a drop in the rate of oxygenation change in dorsolateral prefrontal cortex
under high workload conditions can predict a participant’s performance decline. One
assumption in this interpretation of the data is that individuals may become overwhelmed
and that a mental shift results in a decrease in oxygenation.
54
The main effects for complexity and for the Secondary Verbal Task, despite increasing
perceived cognitive effort (St. John & Kobus, et al., 2002), were not associated with
significant changes in blood oxygenation. A number of potential reasons for this
discrepancy is: (i) the sample size was small, which limits the power of these analyses;
(ii) few features were analyzed. The current analyses focused on only average change in
oxygenation. It is possible that other parameters, such as peak amplitude, pulse width,
latency, etc. could add predictive power in these complex cognitive tasks; (iii) the current
sensor was applied over a limited area of the prefrontal cortex. Some of these
manipulations may have had effects in areas of the cortex that are accessible to fNIR, but
were not measured with the current sensor. This question remains for future generations
of sensor to determine.
In the human performance assessment study, significant effects were found across both
hemispheres (Izzetoglu, K. et al, 2004). Although fNIR has many advantages over fMRI
for field deployment, it does not have the same spatial resolution as fMRI or PET. In
addition, because the depth of fNIR imaging is dependent on the distance between the
source and detector, these distances must be fixed. The current sensor is fixed with
respect to all sources and detectors across both hemispheres. As forehead dimensions
differed among individuals, the additional variance associated with different spatial
locations combined with variations among individual brains, may have contributed to
these widespread effects. Having more independent source-detector arrays in future
generations of the sensor, and more precise localization algorithms, may help to mitigate
this limitation. Finally, the Warship Commander Task itself is complex, and numerous
55
cognitive and emotional functions are occurring during the execution of the task. It is
possible that these various tasks have differential effects on the hemodynamic response.
For example, recent research using PET indicates that various areas of cortex show
increases in oxygenation during a divided attention task relative to a full attention task,
whereas other areas demonstrate decreases in oxygenation during the same task (Iidaka et
al. 2000). Further work is needed to more fully explicate our understanding of brain
function during what may be common everyday, and yet extremely complex tasks.
At the low end of cognitive engagement, the goal is to investigate the capacity of fNIR to
detect changes associate with the depth of anesthesia. In the anesthetic depth assessment
study, the fNIR measures from patients undergoing general anesthesia were analyzed to
examine the following hypothesis: The transition from deep to light anesthetic stages is
associated with reliable changes in oxygenated, deoxygenated, and total hemoglobin in
frontal cortex. The results support this hypothesis and reveal that rate of change in
deoxygenated hemoglobin obtained through the use of fNIR technology can differentiate
deep and light anesthesia stages (p<0.01). The increases in oxy-Hb, deoxy-Hb and total
blood volume during deep anesthesia are due to luxury perfusion, a combination of the
decrease in neuronal metabolic demand coupled with an increase in cerebral blood flow
(CBF), or metabolic supply (Heath et al. 1997, Duffy et al. 2000). The cerebral blood
flow increases are due to the direct cerebral vasodilatory effects of the inhalant
anesthetics such as sevoflurane and desflurane (Kaisti et al. 2003, Bedforth et al. 2000,
Marcar et al. 2006, Heath et al 1997, Matta et al. 1999), followed by an increase in
cerebral blood volume (Morgan et al. 1996, Kaisti et al. 2003). The results of this
56
exploratory study are consistent with this model, as deoxy-Hb levels are correlated with
the level of anesthetic depth, suggesting that the rate of deoxy-Hb changes can be used as
a neuromarker to detect the emergence from deep to light anesthesia. The slower rate of
change in deoxy-Hb may be due to cerebral metabolic rate (demand) suppression during
deep anesthesia, i.e., neuronal activity is inhibited and oxygen consumption reduced by
the administration of anesthetic drugs (Heinke et al. 2005, Alkire et al 1995, Kaisti et al.
2005).
Furthermore, this anesthetic depth assessment research reveals that robust fNIR
measurements can be continuously monitored through the implementation of novel
procedures and algorithms, ICA or PCA in real time even in highly noisy environments
such as the operating room. These preliminary results suggested that fNIR spectroscopy
has the potential to be deployed in the operating room to further aid the anesthesiologists
in the detection of awareness under general anesthesia in the future.
5.2 Future Work
A number of individual differences could play a role in the current results. One
individual difference of particular interest for the human performance assessment study is
that of expertise, or practice. Overall “expertise” was not quantified, although
participants had a wide range of exposure to the task. However, performance in a
complex, rule-driven task like the WCT can reasonably be assumed to be highly
influenced by practice. As such, practice would appear to be an important variable for
this type of task, and should be quantified in future studies. Nevertheless, oxygenation
57
levels in both skilled and novice participants appeared to conform to the hypotheses in
the current study: Those with less skill attain their maximal level of performance sooner
than those with better skills, with an associated drop in cortical oxygenation.
Moreover, control experiments are needed to develop an index for depth of anesthesia
monitoring. This exploratory study indicates that the deoxygenated hemoglobin has a
strong association with changes in deep and light anesthesia stages. Future experiments,
employing control groups, will allow for sensitivity and specificity analyses on
deoxygenated hemoglobin, and help quantify the fNIR performance during stages of
anesthesia. In these control experiments, there will be no surgical procedures and deep
anesthesia will be ensured by adequate sedation via the observer’s assessment of
alertness/sedation score (OAA/S score). To help delineate the effects of conscious
cognitive processes from those of strictly physiological origin (with no effect on
consciousness), a cognitive “probe” will be used to evaluate changes in the processing of
external stimuli. Oxy-Hb and deoxy-Hb changes in response to this cognitive probe can
be identified using the same methods presented in this thesis. During the external stimuli
presentation, the hemodynamic responses are expected to change in response to
conscious awareness, not to background physiological changes.
5.2.1 Future Studies for Applied Research
In both studies, the fNIR sensor is fixed with respect to all sources and detectors across
both hemispheres. As forehead dimensions differ among individuals, the additional
variance could be associated with different spatial locations combined with variations
58
among individual brains. In addition, more than 10% of the subjects in anesthetic depth
assessment study have not been evaluated, because the data was corrupted by noise or the
sensor was saturated. In some cases, very low signal levels were acquired. Therefore, a
new modular split-sensor design (Figure 5.1) that can be attached to left and right
hemispheres may reduce the variance among individuals as well as reduce data loss and
can result in more robust fNIR recording in the future.
Figure 5.1: A new split sensor design for the anesthetic depth assessment studies. One-source, four-detector can be used to scan each hemisphere.
The fNIR technology and methods presented in this research can be particularly deployed
in following application areas:
1. UAV ground control stations and federal aviation administration (FAA) to
monitor operator’s workload for safe navigation.
2. Training facilities to objectively assess a trainees’ level of expertise development
3. Operating room (OR), intensive care unit (ICU), and a range of other settings
where sedation is given to monitor drug emergence from anesthesia and during
administration of supplemental opioid medication.
59
5.3 Conclusion
Existing neuroimaging technologies, such as fMRI and PET, have both strengths and
limitations, and new techniques and technologies are continuously being developed to
extend our knowledge about the neural circuits that underlie human cognitive and
emotional processes. Functional near-infrared spectroscopy (fNIR) is an emerging
neuroimaging technology that uses light in the near-infrared spectrum to provide a
continuous measure of hemodynamic changes, allowing the imaging of changes
associated with neural activity in the cortex of the adult human. The purpose of this
dissertation is to identify markers from the fNIR measurements during change in the
cognitive state of mental engagement at both high and low levels of neural activation. At
the high end of neural activation, the fNIR was used to assess workload in a simulated
command/control work environment. This human performance assessment study
demonstrated that oxygenation increased with rise in task load and was correlated with
performance variables. Furthermore, an abrupt drop in oxygenation under high workload
conditions can be used to predict a participant’s decline in performance. At the low end
of neural activation, awareness under general anesthesia was studied. This exploratory
investigation on predicting awareness under general anesthesia examined the hypothesis
that the transition from deep to light anesthetic stages is associated with reliable changes
in oxygenated, deoxygenated, and total hemoglobin in frontal cortex. The results
suggested that deoxy-Hb can be further developed as an index of the depth of anesthesia
for the purpose of monitoring awareness under general anesthesia.
60
LIST OF REFERENCES
Alkire MT, Haier RJ, Barker SJ, et al. (1995). Cerebral metabolism during propofol anesthesia in humans studied with positron emission tomography. Anesthesiology. 82:393– 403.
Anesthesia Brochure. (2005). Patient awareness under general anesthesia—what is it?
Prepared by the American Society of Anesthesiologists and the American Association of Nurse Anesthetists.
Ardekani BA, Choi SJ, Hossein-Zadeh G, Porjesz B, Tanabe JL, Lim KO, Bilder R,
Helpern JA, Begleiter H. (2002). Functional magnetic resonance imaging of brain activity in the visual oddball task. Cognitive Brain Res. 14:347–356.
Avidan M, Zhang L, et. al. (2008). Anesthesia Awareness and the Bispectral Index. The
new England journal of medicine. 358(11):1097-1108 Bedforth NM, Hardman JG, Nathanson MH. (2000).Cerebral Hemodynamic Response to
the Introduction of Desflurane: A Comparison with Sevoflurane. Anesth Analg. 91:152–5.
Bell AJ, Sejnowski TJ. (1995). An Information-Maximization Approach to Blind
Separation and Blind Deconvolution. Neural Computation. 7(6): 1129-1159. Bell, A. J. and Sejnowski, T. J. (1997). The 'independent components' of natural scenes are
edge filters. Vision Research, 37(23) 3327-3338. Belouchrani A, Meraim KA, Cardoso JF, Moulines E. (1993). Second Order Blind
Separation of Correlated Sources. Proc. of Int. Conf. on Digital Sig. Processing. 346-351. Belouchrani A, Abed-Meraim K, Cardoso JF, Moulines E. (1997). A Blind Source
Separation Technique Using Second-Order Statistics. IEEE Trans. on Signal Processing. 45(2):434-444.
Boas DA, et al. (2001). The accuracy of near infrared spectroscopy and imaging during
focal changes in cerebral hemodynamics. NeuroImage, 13, 76.
Boas DA, Franceschini MA, Dunn AK, Strangman G (2002). Non-Invasive imaging of cerebral activation with diffuse optical tomography. In-Vivo Optical Imaging of Brain Function. Boca Raton: CRC Press, pp 193–221.
Braver TS, Cohen JD, Nystrom LE, Jonides J, Smith EE, Noll DC. (1997). A Parametric
Study of Prefrontal Cortex Involvement in Human Working Memory. NeuroImage. 5:49-62.
61
Broke, B., Battmann, W. (1992). The arousal-activation theory of extraversion and neuroticism: A systematic analysis and principal conclusions. Advances in Behaviour Research & Therapy.14, 211-246. Pergamon Press Inc, US.
Bruhn J, Myles PS, Sneyd R, Struys MMRF. (2006). Depth of anesthesia monitoring:
what’s available, what’s validated and what’s next? Br. J of Anaesthesia. 97(1): 85-94. Bunce S, Izzetoglu M, Izzetoglu K, Onaral B, Pourrezaei K, (2006). Functional Near
Infrared Spectroscopy: An Emerging Neuroimaging Modality. IEEE Engineering in Medicine and Biology Magazine, Special issue on Clinical Neuroengineering, 25(4):54 - 62.
Cabeza, R & Nyberg, L (2000). Imaging Cognition II: An empirical Review of 275 PET
and fMRI Studies. J. Cognitive Neuroscience, 12. 1-47. Cardoso JF. (1998). Blind Signal Separation: Statistical Principles. Proc. of the IEEE.
9(10): 2009-2025. Cenic A, Craen RA, Howard-Lech VL, Lee TY, Gelb AW. (2000).Cerebral blood volume
and blood flow at varying arterial carbon dioxide tension levels in rabbits during propofol anesthesia. Neurosurgical Anesthesia. Anesth Analg. 90:1376–83.
Chance, B. Zhuang Z, UnAh C, Alter C, Lipton L (1993). Cognition-activated low-
frequency modulation of light absorption in human brain. Proceedings of the National Academy of Science of the United States of America 90, 3770–3774.
Chance B, Anday E, Nioka S, Zhou S, Hong L, Worden K, Li C, Murray T, Ovetsky Y,
Pidikiti D, Thomas R (1998). A novel method for fast imaging of brain function, non-invasively, with light. Optics Express, 2, 10.
Cope, M. (1991). The Development of a Near-Infrared Spectroscopy System and Its
Application for Noninvasive Monitoring of Cerebral Blood and Tissue Oxygenation in the Newborn Infant. Univ.College London, London.
DARPA Augmented Cognition Technological Integration Experiment Report (2003). July
7 2003. Davidson RJ, & Sutton, SK (1998). Affective neuroscience: The emergence of a
discipline. Current Opinions in Neurobiology, 5, 217-224 Devaraj A, Izzetoglu M, Izzetoglu K, Onaral B, Pourrezaei K, (2004). Arterial Pulsation
and Respiration Artifact Cancellation in Functional NIR Spectroscopy. Proc. of BMES 2004.
Duda RO, Hart PE, Stork G. (2001) Pattern Classification. John Wiley & Sons.
62
Duffy CM, Matta BF. (2000). Sevoflurane and Anesthesia for Neurosurgery:A Review. Journal of Neurosurgical Anesthesiology. 12(2): 128–140
Ebert TJ. (2005). Sympathetic and Hemodynamic Effects of Moderate and Deep Sedation
with Propofol in Humans. Anesthesiology. 103:20–4. Haykin S. (2001). Adaptive filter theory. Prentice Hall, Upper Saddle River. Heath KJ, Gupta S, Matta BF. (1997).The effects of sevoflurane on cerebral hemodynamics
during propofol anesthesia. Anesthesia & Analgesia. 85:1284-1287. Heinke W, Koelsch S. (2005).The effects of anesthetics on brain activity and cognitive
function. Current Opinion in Anaesthesiology . 18:625–631. Hyvarinen A, Oja E. (1997). A Fast Fixed Point Algorithm for Independent Component
Analysis. Neural Computation. 9(7):1483-1492. Hyvarinen A. (1999). Survey on Independent Component Analysis. Neural Computating
Surveys. 2:94-128. Hyvarinen A, Karhunen J, Oja E. (2001). Independent Component Analysis. John Wiley &
Sons. Iidaka T, Anderson ND, Kapur S, Cabeza R, Craik FIM (2000). The effect of divided
attention on encoding and retrieval in episodic memory revealed by positron emission tomography. Journal of Cognitive Neuroscience, 12:2, 267 –280.
Iwasaki K, NomotoY, Ishiwata1 M, Yokota T, Ogawa R. (2003). Vital capacity induction
with 8% sevoflurane and N2O causes cerebral hyperemia. J Anesth. 17:3–7. Izzetoglu, K., Yurtsever, G., Bozkurt, A., Bunce, S. (2003a) Functional brain monitoring
via NIR based optical spectroscopy. Proceedings of IEEE 29th Annual Bioengineering Conference, 335-336
Izzetoglu K, Yurtsever G, Bozkurt A, Yazici B, Bunce S, Pourrezaei K, Onaral B (2003b).
NIR Spectroscopy Measurements of Cognitive Load Elicited by GKT and Target Categorization. Proceedings of the 36th Hawaii International Conference on System Sciences, 129 –134.
Izzetoglu K, Bunce S, Onaral B, Pourrezaei K, Chance B, (2004). Functional Optical Brain
Imaging Using Near-Infrared During Cognitive Tasks. Int. J. of Human-Comp. Int., 17(2):211-227.
Izzetoglu M, Izzetoglu K, Bunce S, Onaral B, Pourrezaei K, (2005a). Functional Near-
Infrared Neuroimaging. IEEE Trans. on Neural Systems and Rehabilitation Engineering, 13(2):153-159.
63
Izzetoglu M, Devaraj A, Bunce S, Onaral B, (2005b). Motion Artifact Cancellation in NIR
Spectroscopy Using Wiener Filtering. IEEE Transactions on Biomedical Engineering, 52(5):934-938.
Izzetoglu M, Bunce S, Izzetoglu K, Onaral B, Pourrezaei K, (2007). Functional Brain
Imaging Using Near Infrared Technology for Cognitive Activity Assessment. IEEE Engineering in Medicine and Biology Magazine, Special issue on on the Role of Optical Imaging in Augmented Cognition, 26(4):38 – 46.
Kailath T., Sayeed A.H., Hassibi B. (2000). Linear Estimation. Prentice-Hall, Upper Saddle
River. Kaisti KK, LångsjöJW, Aalto S, Oikonen V, Sipilä H, Teräs M, Hinkka S, Metsähonkala L,
Scheinin H.(2003). Effects of Sevoflurane, Propofol, and Adjunct Nitrous Oxide on Regional Cerebral Blood Flow, Oxygen Consumption, and Blood Volume in Humans. Anesthesiology. 99:603–13.
Kerssens , Hamann S, Peltier S, Hu XP, Byas-Smith MG, Sebel PS. (2005). Attenuated
Brain Response to Auditory Word Stimulation with Sevoflurane: A Functional Magnetic Resonance Imaging Study in Humans. Anesthesiology. 103:11-19.
Lahti KM, Ferris CF, Li F, Sotak CH & King JA (1999) Comparison of evoked cortical
activity in conscious and propofol-anesthetized rats using functional MRI. Magn Reson Med 41(2): 412-6.
Lay D. (2000). Linear Algebra and It’s Applications. Addison-Wesley, New York. Leduc, Patricia A., Rash, Clarence E., Manning, Sharon D.(2005). Human Factors in
UAV Accidents. Special Operations Technology. Volume: 3 Issue: 8. Leon-Carrion J, Damas J, Izzetoglu K, Pourrezai K, Martin-Rodriguez JF, Martin JM,
Dominguez-Morales MR. (2006). Differential Time Course and Intensity of PFC Activation for Men and Women in Response to Emotional Stimuli: A Functional Near-Infrared Spectroscopy (fNIRS) Study. Neuroscience Letters, 403(1-2):90-95.
Leon-Carrion J, Martin-Rodriguez JF, Damas-Lopez J, Pourrezaei K, Izzetoglu K, Martin
JM, Dominguez-Morales MR. (2007a). Does Dorsolateral Prefrontal Cortex (DLPFC) Activation Return to Baseline When Sexual Stimuli Cease? The Role of DLPFC in Visual Sexual Stimulation. Neuroscience Letters, 416:55-60.
Leon-Carrion J, Martin-Rodriguez JF, Damas-Lopez J, Pourrezai K, Izzetoglu K, Martin
JM, Dominguez-Morales MR. (2007b). A Lasting Post-Stimulus Activation on Dorsolateral Prefrontal Cortex is Produced When Processing Valence and Arousal in Visual Affective Stimuli. Neuroscience Letters, 422(3):147-52.
64
Levendowski DJ, Berka C, et al. (2002). Event-related potentials during a test of working memory differentiate sleep apnea patients from healthy subjects. Sleep, 25, A460-A461.
Lovell AT, Owen-Reece H, Elwell CE, Smith M, Goldstone JC. (1999). Continuous
Measurement of Cerebral Oxygenation by Near Infrared Spectroscopy During Induction of Anesthesia. Anesth Analg, 88:554-557.
Marcar VL, Schwarz U, Martin E, Loenneker T. (2006). How Depth of Anesthesia
Influences the Blood Oxygenation Level–Dependent Signal from the Visual Cortex of Children. Am. J. Neuroradiol. 27:799-805.
Marshall SP, Pleydell-Pearce CW, Dickson BT (2003). Integrating psychophysiological
measures of cognitive workload and eye movements to detect strategy shifts. Proceedings of the 36th Annual Hawaii International Conference on System Sciences, 130 –135.
Matta BF. (1995). Direct cerebrovasodilatory effects of halothane, isoflurane, and desflurane during propofol-induced isoelectric electroencephalogram in humans. Anesthesiology - 83(5): 980-5.
Matta BF, Heath KJ, Tipping K, Summors AC. (1999). Direct Cerebral Vasodilatory
Effects of Sevoflurane and Isoflurane. Anesthesiology. 91(3):677. McCarthy G, Luby M, Gore J, Goldman-Rakic P.(1997). Infrequent events transiently
activate human prefrontal and parietal cortex as measured by functional MRI. J. Neurophysiol. 77:1630-1634.
Minna M. (2004) EEG and BOLD-contrast fMRI in Brain: Cerebrovascular reactivity,
suppression of neuronal activity, global and local brain injury. Academic Dissertation. University of Oulu Hospital.
Morgan, Jr GE, Mikhail MS, Murray MJ. (1996). Clinical Anesthesiology, 4th Edition. Mc
Graw Hill. (Access Medicine) Obrig H, Villringer A (1997). Near-infrared spectroscopy in functional activation studies.
Can NIRS demonstrate cortical activation? Adv. Exp. Med. Biol., 413, 113. Ochoa JB. (2002). EEG Signal Classification for Brain Computer Interface Applications.
Ecole Polytechnique Federale De Laisanne. Diploma Project. Parra L, Alvino C, Tang A, Pearlmutter B, Yeung N, Osman A, Sajda P. (2002). Linear
Spatial Integration for Single Trial Detection in Encephalography. Neuroimage. 17: 223-230.
Pollard RJ, Coyle JP, Gilbert RL, Beck JE, (2007). Intraoperative Awareness in a Regional
Medical System A Review of 3 Years’ Data. Anesthesiology. 106:269–74.
65
Razoumnikova, O. (2003). Interaction of personality and intelligence factors in cortex activity modulation. Personality & Individual Differences, 35, 135-162.
Schmorrow DD, Kruse AA (2002). DARPA's augmented cognition program tomorrow's
human computer interaction from vision to reality: building cognitively aware computational systems. Proceedings of the IEEE 7th Conference on Human Factors and Power Plants; 7:1-4.
Sebel PS, Bowdle TA, Ghoneim MM, et al. (2004). The incidence of awareness during
anesthesia: a multicenter United States study. Anesth Analg. 99:833-9. Sigalovsky N. (2003) Awareness under anesthesia. AANA Journal. 71(5): 373-379. Singh H. (1999). Bispectral index (BIS) monitoring during propofol-induced sedation and
anesthesia. European J of Anesthesiology. 16:31-36. Smith EE, Jonides J. (1997). Working Memory: A View from Neuroimaging. Cognitive
Psychology. 33:5–42. St. John M, Kobus DA, et al. (2002). A Multi-Tasking Environment for Manipulating and
Measuring Neural Correlates of Cognitive Workload,” Proceedings of the IEEE 7th Conference on Human Factors and Power Plants; 7:10 –14.
Strangman, G., Boas, D.A., Sutton, J.P. (2002) Non-invasive neuroimaging using near-
infrared light. Biological Psychiatry. 52(7), 679-93. Tang AC, Pearlmutter BA, Malaszenko NA, Phung DB, Reeb BC. (2002). Independent
Components of Magnetoencephalography. Neural Computation. 14:1827-1858. Vandesteene A, Trempont V, Engelman E, Deloof T, Focroul M, Schoutens A, de Rood M.
(1988). Effect of propofol on cerebral blood flow and metabolism in man. Anaesthesia. 43(1):42-43.
Villringer A, Chance B (1997). Non-invasive optical spectroscopy and imaging of human
brain function. Trends in Neuroscience, 20, 435-442. Yerkes R.M and Dodson J (1908) The Relation of Strength of Stimulus to Rapidity of
Habit-Formation. Journal of Comparative Neurology and Psychology, 18, 459-482.
66
APPENDIX A: ICA AND PCA TESTS A.1. Independent Component Analysis
Independent Component Analysis (ICA) (Duda 2001, Cadoso 1998, Hyvarinen 1999,
Hvarinen et al. 2001) is a relatively recent method for blind source separation which has
been used in several different application areas involving EEG, MEG, MRI, etc. (Parra et
al. 2002, Tang et al. 2002, Belouchrani et al. 1997, Ochoa 2002). ICA has been
developed from the well-known principal component analysis (PCA). ICA uses higher-
order statistical information to find a suitable basis in such a way that the statistical
independence between the projections of the signal onto these basis vectors is
maximized. In contrast, PCA uses only second order statistics to separate a signal into
uncorrelated components.
ICA assumes the existence of n signals that are linear mixtures of m unknown
independent source signals (Duda 2001, Hyvarinen 1999, Hvarinen et al. 2001). At time
instant t, the observed n-dimensional data vector x(t)=[x1(t)…xn(t)]T is given by the
model:
( ) ( )t tx = As (A.1)
where both the independent source signals s(t)=[s1(t)…sm(t)]T and the coefficients of the
mixing matrix A are unknown. Conditions for the existence of a solution are (1) n≥m
(there are at least as many mixtures as the number of independent sources), and (2) up to
one source may be Gaussian (Hyvarinen 1999). Under these assumptions the ICA seeks a
solution of the form:
67
ˆ( ) ( )t ts = Bx (A.2)
where 1ˆ B = A is called the separating matrix.
Although ICA has many advantages over existing signal separation methods (for example
it does not rely on the availability of clean reference channels to separate signals as in
adaptive filtering), its performance depends highly on the ICA method selected. Several
ICA algorithms such as second-order blind identification (SOBI) (Belouchrani et al.
1993), Infomax (Bell et al. 1995), and fICA (Hyvarinen et al 1997) have been developed
and applied to EEG and MEG data. In this application, we decide to use fICA method
since it outperformed the other methods.
A. 2. Principal Component Analysis
Principal component analysis (PCA) solves the same problem as in ICA where the
observed n-dimensional data vector x(t)=[x1(t)…xn(t)]T is again given by a linear model,
( ) ( )t tx = As and where both the source signals s(t)=[s1(t)…sm(t)]T and the coefficients of
the mixing matrix A are unknown. Again as in ICA, PCA seeks a solution of the form:
ˆ( ) ( )t ts = Bx where 1ˆ B = A is called the separating matrix. Unlike the much stronger
condition in ICA for the sources to be statistically independent, in PCA a weaker
condition is assumed and the sources should be uncorrelated. Therefore, PCA estimates
the unknown sources and the separating matrix using only second order statistics based
on singular value decomposition (Bell et al. 1997, Lay 2000). The principal components
come from the eigenvectors of the covariance matrix of x(t) which are the columns of a
matrix E satisfying
68
EDE-1=<xxT> (A.3)
where D is the diagonal matrix of eigenvalues. Then the separating matrix B is then
found as B=D(-1/2)ET
A. 3. ICA and PCA Performances
We performed the aforementioned ICA and PCA algorithms to remove the motion
artifact from the raw fNIR measurements obtained at 730 and 850nm where their
correlation with the measurement obtained at dark current condition is above rorig>0.7 and
where the measurement at dark current condition is above the dark current intensity
levels on 26 patient and 16 channel recordings. We first performed this correlational
analysis and then checked on intensity level of the reference signal to not induce any
additional noise to the data due to the application of unnecessary signal processing
algorithms. Intensity level of the reference signal should be checked because in some
cases there are no correlated changes or no high intensity levels in the reference signal
accounting for a possible noise with the other raw intensity measurements. Two cases in
each category where there was a significant correlation between the reference signal
(green) and other raw intensity measurements (blue for 730nm and red for 850nm) and
where there was none is shown in Figure A.1(a) and (b) respectively.
69
(a) (b) Figure A.1: Examples of raw intensity measurements where (a) there was a significant
correlation with the reference measurement and (b) there was almost no correlation.
In the cases shown in Figure A.1(a) even though there was no light induced to the tissue,
there were fluctuations in the reference which may be the effects of noise due to motion
or coming from the wires, the ambient light and so forth. Similar type of fluctuations
were also apparent in the 730nm and 850nm measurements which were also reflected in
the correlation assessment. In the cases shown in Figure A.1(b), the intensity level in the
reference signal was low and there were no correlated fluctuations in the reference
measurement with the other intensity measurements. This type of pattern for the
0 50 100 150 200 250 300 350 400 450
500
1000
1500
2000
2500
3000
3500
Subject 15, Channel 12
730nm, Correlation with Reference R=0.90
Reference Measurement850nm, Correlation with Reference R=0.90
0 50 100 150 200 250 300 350 400 450
400
500
600
700
800
900
1000
1100
1200
1300Subject 3, Channel 8
730nm, Correlation with Reference R=-0.04
Reference Measurement850nm, Correlation with Reference R=0.07
0 50 100 150 200 250 300 350 400 450
400
600
800
1000
1200
1400
1600
1800
2000Subject 5, channel 16
730nm, Correlation with Reference R=0.08
Reference Measurement850nm, Correlation with Reference R=0.34
0 50 100 150 200 250 300 350 400 450
500
1000
1500
2000
2500
3000
3500
Subject 8, Channel 4
730nm, Correlation with Reference R=0.90
Reference Measurement850nm, Correlation with Reference R=0.73
70
reference signal and correlated function may mean that in this patient possibly no artifact
had occurred.
In order to be able to quantitatively compare the performances of the proposed algorithms
in noise removal, we used a performance measure based on the correlation between the
reference signal (the true noise) and the noise suppressed intensity signal at 730 or 850nm
estimated by the selected algorithm, namely (rest). Since the noise within the original raw
intensity measurement will be reduced in the estimated intensity signal, the correlation of
it with the reference signal is expected to be the lowest in the best performing algorithm.
The mean value and the standard deviation of (rorig )and (rest ) after the application of ICA
and PCA algorithms on 310 measures are presented in Figure A.2. Both algorithms were
successful in noise suppression where the correlation of the reference signal with the
original noisy raw intensity measurement (rorig) was very high with a mean value
approximately 0.89. After the application of the proposed algorithms, the correlation of
the reference signal with noise suppressed intensity measurement, ( rest )dropped in both
ICA and PCA as expected. The mean value of (rest ) for both the algorithms was found to
be very close suggesting they are not different from each other in terms of their
performance in noise suppression.
71
Figure A.2: Mean and standard deviation of the correlation between reference signal and
the original intensity measurement (whether at 730 or 850nm) rorig and the estimated intensity measurement rest after applying the ICA and PCA algorithms over 310 cases
with rorig>0.7
Correlation between Reference Signal and the Original and Estimated Intensity Measurements using Different Algorithms
0.00
0.20
0.40
0.60
0.80
1.00
1.20
R_orig R_ICA R_PCA
72
VITA
Kurtulus Izzetoglu EDUCATION/TRAINING
INSTITUTION AND LOCATION DEGREE
(if applicable)
YEAR(s) FIELD OF
STUDY
Middle East Technical University, Ankara, Turkey
B.S. 1992 Electrical Engineering
Middle East Technical University, Ankara, Turkey
M.S.
1995
Electrical Engineering
POSITIONS AND EMPLOYMENT 1997-2000 Senior Programmer/Analyst, Premier Data Corporation, Springfield, IL 2001-2002 Application Developer, MEDIS medical imaging systems, Leiden, The
Netherlands 2002- Research Engineer, School of Biomedical Engineering, Drexel University,
Philadelphia, PA SELECT PUBLICATIONS Izzetoglu K, Bunce S, Onaral B, Pourrezaei K, Chance B, (2004). Functional Optical
Brain Imaging Using Near-Infrared During Cognitive Tasks. Int. J. of Human-Comp. Int., 17(2):211-227.
Izzetoglu M, Izzetoglu K, Bunce S, Onaral B, Pourrezaei K, (2005). Functional Near-Infrared Neuroimaging. IEEE Trans. on Neural Systems and Rehabilitation Engineering, 13(2):153-159.
Bunce S, Izzetoglu M, Izzetoglu K, Onaral B, Pourrezaei K, (2006). Functional Near Infrared Spectroscopy: An Emerging Neuroimaging Modality. IEEE Engineering in Medicine and Biology Magazine, Special issue on Clinical Neuroengineering, 25(4):54 - 62
Leon-Carrion J, Damas J, Izzetoglu K, Pourrezai K, Martin-Rodriguez JF, Martin JM, Dominguez-Morales MR (2006). Differential time course and intensity of PFC activation for men and women in response to emotional stimuli: A functional near-infrared spectroscopy (fNIRS) study. Neuroscience Letters.
Leon-Carrion J, Martin-Rodriguez JF, Damas-Lopez J, Pourrezai K, Izzetoglu K, Barroso y Martin JM, Dominguez-Morales MR (2007). Lasting post-stimulus activation on dorsolateral prefrontal cortex is produced when processing valence and arousal in visual affective stimuli. Neurosci Lett. 2007 Jul 18;422(3):147-52.
Leon-Carrion J, Martin-Rodriguez JF, Damas-Lopez J, Pourrezai K, Izzetoglu K, Barroso Y Martin JM, Dominguez-Morales MR (2007). Does dorsolateral prefrontal cortex (DLPFC) activation return to baseline when sexual stimuli cease? The role of DLPFC in visual sexual stimulation. Neurosci Lett. 2007 Apr 6;416(1):55-60.
