ThesisPDF Available

Electrophysiological, Hemodynamic, and Metabolic Effects of Transcranial Photobiomodulation (tPBM) on Topographical and Physiological Connectivity in the Human Brain

Authors:

Abstract and Figures

Transcranial photobiomodulation (tPBM) targets the human brain with near-infrared (NIR) light and is shown to affect human cognitive performance and neural electrophysiological activity as well as concentration changes of oxidized cytochrome-c-oxidase ([CCO]) and hemoglobin oxygenation ([HbO]) in human brain. Brain topographical connectivity, which shows the communication between regions of the brain, and its alteration can be assessed to quantify the effects of external stimuli, diseases, and cognitive decline, in resting-state or task-based measurements. Furthermore, understanding the interactions between different physiological representations of neural activity, namely electrophysiological, hemodynamic, and metabolic signals in the human brain, has been an important topic among researchers in recent decades. In my doctoral study, neurophysiological networks were constructed using frequency-domain analyses on oscillations of electroencephalogram (EEG), [CCO], and [HbO] time series that were acquired by a portable EEG and 2-channel broadband near-infrared spectroscopy (2-bbNIRS). Specifically, my dissertation included three aims. The first one was to examine how tPBM altered the topographical connectivity in the electrophysiological oscillations of the resting human brain. As the first step, I defined and found key regions and clusters in the EEG sensor space that were affected the most by tPBM during and after the stimulation using both cluster-based power analysis and graph-based connectivity analysis. The results showed that the right prefrontal 1064-nm tPBM modulates several global and regional electrophysiological networks by shifting the information path towards frontal regions, especially in the beta band. For the second aim, I performed 2-bbNIRS measurements from 26 healthy humans and developed a methodology that enabled quantification of the infra-slow oscillation (ISO) power and connectivity between bilateral frontal regions of the human brain in resting state and in response to frontal tPBM stimulation at different sites and laser wavelengths. As the result, several stable and consistent features were extracted in the resting state of 26 young healthy adults. Moreover, these features were used to reveal some effects of tPBM on prefrontal metabolism and hemodynamics, while illustrating the similarities and differences between different stimulation conditions. Finally, the third aim was to investigate the resting-state prefrontal physiological network and the corresponding modulation in response to left frontal 800-nm tPBM by determining the effective connectivity/coupling between each pair of the electrophysiological, hemodynamic, and metabolic ISO of the human brain. Complementary to the previous studies, my study showed that prefrontal tPBM not only modulates the information path between two locations of the prefrontal cortex, it can also induce unilateral alterations in interactions between neural activity, hemodynamics, and metabolism. Overall, my dissertation shed light on the mechanism of action of prefrontal tPBM.
Content may be subject to copyright.
Electrophysiological, Hemodynamic, and Metabolic Effects of
Transcranial Photobiomodulation (tPBM) on Topographical and
Physiological Connectivity in the Human Brain
Sadra Shahdadian
Presented to the Faculty of the Graduate School of
The University of Texas at Arlington in Partial Fulfillment
Of the Requirements
For the Degree of
DOCTOR OF PHILOSOPHY
THE UNIVERSITY OF TEXAS AT ARLINGTON
August 2022
ii
Copyright © by Sadra Shahdadian 2022
All Rights Reserved
iii
ACKNOWLEDGMENTS
First, I would like to acknowledge and thank my supervising professor, Dr. Hanli Liu for her
guidance, mentorship, and support towards completing the degree. Her professional guidance and
powerful encouragement made this work possible.
Second, I would like to thank my committee members, Dr. Alexandrakis, Dr. Busch, and Dr. Wang
for their time and guidance on this work.
Third, I would like to extend my appreciation to fellow doctoral, post-doctoral colleagues for their
assistance and support in my research including Dr. Xinlong Wang, Dr. Nghi Troung, Dr. Anqi
Wu, Dr. Parisa Rabbani, Dr. Hashini Wanniarachchi, Dr. Tyrell Pruitt, Dr. Akhil Chaudhari, Kang
Shu, Caroline Carter, Haylea Renguul, Sheetala Ranjitkar, Sylvine Ineza.
Finally, and most importantly, I would like to thank my parents, Dr. Javad Shahdadian and Dr.
Zohreh Esmaili, and my sister Ms. Sana Shahdadian who have always supported and encouraged
me during my long journey. Special thanks to my fiancé, Ms. Yasamin Mahboub for her
unconditional support and encouragement throughout the past year.
iv
ABSTRACT
Electrophysiological, Hemodynamic, and Metabolic Effects of
Transcranial Photobiomodulation (tPBM) on Topographical and
Physiological Connectivity in the Human Brain
Sadra Shahdadian
The University of Texas at Arlington, 2022
Supervising Professor: Dr. Hanli Liu
Transcranial photobiomodulation (tPBM) targets the human brain with near-infrared (NIR) light
and is shown to affect human cognitive performance and neural electrophysiological activity as
well as concentration changes of oxidized cytochrome-c-oxidase ([CCO]) and hemoglobin
oxygenation ([HbO]) in human brain.
Brain topographical connectivity, which shows the communication between regions of the brain,
and its alteration can be assessed to quantify the effects of external stimuli, diseases, and cognitive
decline, in resting-state or task-based measurements. Furthermore, understanding the interactions
between different physiological representations of neural activity, namely electrophysiological,
hemodynamic, and metabolic signals in the human brain, has been an important topic among
researchers in recent decades. In my doctoral study, neurophysiological networks were constructed
using frequency-domain analyses on oscillations of electroencephalogram (EEG), [CCO], and
[HbO] time series that were acquired by a portable EEG and 2-channel broadband near-infrared
spectroscopy (2-bbNIRS).
Specifically, my dissertation included three aims. The first one was to examine how tPBM
altered the topographical connectivity in the electrophysiological oscillations of the resting human
brain. As the first step, I defined and found key regions and clusters in the EEG sensor space that
v
were affected the most by tPBM during and after the stimulation using both cluster-based power
analysis and graph-based connectivity analysis. The results showed that the right prefrontal 1064-
nm tPBM modulates several global and regional electrophysiological networks by shifting the
information path towards frontal regions, especially in the beta band. For the second aim, I
performed 2-bbNIRS measurements from 26 healthy humans and developed a methodology that
enabled quantification of the infra-slow oscillation (ISO) power and connectivity between bilateral
frontal regions of the human brain in resting state and in response to frontal tPBM stimulation at
different sites and laser wavelengths. As the result, several stable and consistent features were
extracted in the resting state of 26 young healthy adults. Moreover, these features were used to
reveal some effects of tPBM on prefrontal metabolism and hemodynamics, while illustrating the
similarities and differences between different stimulation conditions. Finally, the third aim was to
investigate the resting-state prefrontal physiological network and the corresponding modulation in
response to left frontal 800-nm tPBM by determining the effective connectivity/coupling between
each pair of the electrophysiological, hemodynamic, and metabolic ISO of the human brain.
Complementary to the previous studies, my study showed that prefrontal tPBM not only modulates
the information path between two locations of the prefrontal cortex, it can also induce unilateral
alterations in interactions between neural activity, hemodynamics, and metabolism. Overall, my
dissertation shed light on the mechanism of action of prefrontal tPBM.
vi
Table of Contents
List of Figures .............................................................................................................................. xiii
List of Tables ................................................................................................................................ xx
Introduction ..................................................................................................................................... 1
1.1 Significance and Specific Aims ................................................................................................ 1
1.1.1 The need for a topographical mapping of the brain’s electrophysiological network
modulation in response to tPBM................................................................................................. 2
1.1.2 The need for a topographical mapping of the resting state prefrontal cortex
hemodynamic and metabolic network ........................................................................................ 3
1.1.3 The need for a topographical mapping of the prefrontal cortex hemodynamic and
metabolic network in response to different wavelengths and stimulation sites of tPBM ........... 4
1.1.4 The need for a physiological connectivity assessment of the brain’s
electrophysiological, hemodynamic, and metabolic network modulation in response to tPBM 6
Neuromodulation of brain topography and network topology by prefrontal transcranial
photobiomodulation ........................................................................................................................ 8
2.1 Introduction ............................................................................................................................... 8
2.2 Materials and methods ............................................................................................................ 12
2.2.1 Participants ....................................................................................................................... 12
2.2.2 Experimental setup and protocol ..................................................................................... 12
2.2.3 Overview of data processing steps ................................................................................... 14
2.2.4 Data pre-processing for EEG time series ......................................................................... 14
vii
2.2.5 EEG power spectral density and changes in power ......................................................... 15
2.2.6 Statistical analysis for EEG power topography ............................................................... 16
2.2.7 Amplitude and phase decomposition of EEG signal ....................................................... 17
2.2.8 Imaginary part of coherence as EEG connectivity measure ............................................ 17
2.2.9 Global and nodal graphical metrics selected for GTA ..................................................... 19
2.3 Results ..................................................................................................................................... 21
2.3.1 Topographic changes in power between tPBM and sham stimulations .......................... 22
2.3.2 Global graphical metrics of functional connectivity altered by tPBM ....................... 24
2.3.3 Nodal graphical metrics of functional connectivity altered by tPBM ........................ 25
2.4 Discussion ............................................................................................................................... 27
2.4.1 tPBM-induced alterations on EEG
mPower in clusters of electrodes in frontoparietal
network ..................................................................................................................................... 28
2.4.2 tPBM-induced alterations in global measures of functional network in beta band ......... 28
2.4.3 tPBM-induced alterations in nodal graphical measures in beta band .............................. 30
2.4.4 The role of the beta band in tPBM-induced network modulation and its relation to
enhancement of human cognition ............................................................................................. 32
2.4.5 Comparisons to two other publications ............................................................................ 33
2.4.6 Limitations and future work ............................................................................................. 35
2.5 Conclusion .............................................................................................................................. 36
viii
Prefrontal Cortical Connectivity and Coupling of Infraslow Oscillation in the Resting Human
Brain:............................................................................................................................................. 37
3.1 Introduction ............................................................................................................................. 37
3.1.1 Infra-slow Oscillation of the Human Brain ..................................................................... 37
3.1.2 Exploration of the Prefrontal Cortical Connectivity and Coupling of ISO ..................... 38
3.2 Materials and Methods ............................................................................................................ 40
3.2.1 Participants ....................................................................................................................... 40
3.2.2 Experiment Setup and Protocol ....................................................................................... 41
3.2.2 Data Analysis ................................................................................................................... 42
3.2.2.1 Step 1: Quantification of [HbO] and [CCO] time series ...................................... 44
3.2.2.2 Step 2: Multi-taper method for spectral analysis of [HbO] and [CCO] ............... 44
3.2.2.3 Step 3: Quantification of SA in E/N/M Bands.......................................................... 45
3.2.2.4 Step 4: Hemodynamic and Metabolic Connectivity and Coupling by Coherence ... 45
3.3 Results ..................................................................................................................................... 48
3.3.1 Time Series of [HbO] and [CCO] versus Their Spectral Analysis ............................. 48
3.3.2 ISO Spectral Amplitudes of Prefrontal Δ[HbO] and Δ[CCO] in the Resting Brain ........ 49
3.3.3 ISO Coherence of Prefrontal Δ[HbO] and Δ[CCO] in the Resting Human Brain........... 52
3.4 Discussion ............................................................................................................................... 56
3.4.1 ISO Spectral Amplitudes of Prefrontal [HbO] and [CCO] as Brain-state Features .... 57
3.4.2 Cerebral Hemodynamic and Metabolic ISO Connectivity/Coupling as Features ........... 59
ix
3.4.3 Eight Measurable Features of Prefrontal ISO .................................................................. 60
3.4.4 Limitations ....................................................................................................................... 61
3.4.5 Future work ...................................................................................................................... 62
3.5. Conclusion ............................................................................................................................. 63
3.6. Effect of Gender on Measurable Features of Prefrontal ISO ............................................. 64
Wavelength- and Site-Specific Effects of Prefrontal Photobiomodulation in vivo on Bilateral
Metabolic Connectivity and Unilateral Metabolic-Hemodynamic Coupling in Humans ............ 65
4.1 Introduction ........................................................................................................................ 65
4.2 Materials and Methods ....................................................................................................... 68
4.2.1 Participants ................................................................................................................. 68
4.2.2 Experiment Setup and Protocol .................................................................................. 69
4.2.3 Data Analysis .............................................................................................................. 71
4.2.3.1 Amplitude and Phase Decomposition ....................................................................... 72
4.2.3.2 Hemodynamic and Metabolic Connectivity/Coupling Quantification ..................... 73
4.2 Results ................................................................................................................................ 75
4.2.3 tPBM-induced Alterations in Spectral Amplitude of Δ[HbO] and Δ[CCO] ISO ...... 75
4.2.4 tPBM-induced Alterations in ISO Coherence of Prefrontal Δ[HbO] and Δ[CCO] .... 76
4.3 Discussions ........................................................................................................................ 79
4.3.3 Alterations in ISO Prefrontal SAHbO and SACCO in Response to Different tPBM
Conditions ................................................................................................................................. 80
x
4.3.4 Alterations in ISO Prefrontal bCON and uCOP in Response to Different tPBM
Conditions ................................................................................................................................. 81
4.3.5 Application of Different tPBM Conditions ................................................................ 83
4.3.6 Limitations and Future Work ..................................................................................... 84
4.4 Conclusion ......................................................................................................................... 85
Prefrontal transcranial photobiomodulation alters the physiological network in the prefrontal
cortex of healthy adults; A directed neurovascular, neurometabolic, and metabolic-vascular
coupling analysis .......................................................................................................................... 87
5.1 Introduction ............................................................................................................................. 87
5.2 Materials and Methods ............................................................................................................ 91
5.2.1 Experiment Setup and Protocol ....................................................................................... 91
5.2.2 Overview of data processing steps ................................................................................... 93
5.2.3 bbNIRS data preprocessing .............................................................................................. 94
5.2.4 EEG data preprocessing ................................................................................................... 95
5.2.5 EEG data down-sampling ................................................................................................ 96
5.2.6 Topographical directed hemodynamic and metabolic connectivity ................................ 97
5.2.7 physiological network construction ................................................................................. 97
5.2.8 tPBM-induced modulation in metabolic-vascular and physiological network ................ 98
5.3 Results ..................................................................................................................................... 99
5.3.1 Resting-state prefrontal metabolic-vascular network ...................................................... 99
xi
5.3.2 Resting-state unilateral physiological network .............................................................. 100
5.3.3 tPBM-induced directed alterations in bilateral metabolic and hemodynamic connectivity
................................................................................................................................................. 101
5.3.4 tPBM-induced directed alterations in unilateral physiological network ........................ 102
5.4 Discussion ............................................................................................................................. 103
5.4.1 Resting-state prefrontal metabolic-vascular network .................................................... 103
5.4.2 Resting-state prefrontal unilateral physiological network ............................................. 105
5.4.3 Alterations in bilateral metabolic and hemodynamic connectivity, modulated by
prefrontal tPBM ...................................................................................................................... 106
5.4.4 Alterations in the unilateral physiological network, modulated by prefrontal tPBM .... 107
5.4.5 Combination of dual-mode data and MVAR model-based connectivity analysis ......... 108
5.4.6 Limitations and future works ......................................................................................... 109
5.5 Conclusion ............................................................................................................................ 110
Conclusion .................................................................................................................................. 112
6.1 Summary of the Dissertation ................................................................................................ 112
6.2 Limitations and Future Works .............................................................................................. 115
References ................................................................................................................................... 117
Appendix A ................................................................................................................................. 128
Appendix B ................................................................................................................................. 130
B.1 Steps for Frequency-Domain Data Analysis of Prefrontal [HbO] and [CCO] at Rest 130
xii
B.2 Statistical Analysis for the Test of Equivalence Using the Two One-Sided Tests (TOST)
................................................................................................................................................. 131
B.3 Decomposition of a [HbO] Time Series into Three ISO Frequency Bands ..................... 132
xiii
List of Figures
Figure 2.1 A crossover experimental protocol for tPBM and sham experiments (n=45) with
simultaneous EEG recording. The participants were at wakeful resting state with eyes closed. . 12
Figure 2. 2 A data processing flow chart, including steps for (1) data pre-processing (blue
boxes), (2) PSD-based analysis and permutation tests to form power topographies (orange
boxes), (3) graphical edge formation based on the “imaginary part of coherence” analysis, (4)
GTA-based assessment for global graphical connectivity metrics (pink boxes), and (5) GTA-
based assessment for nodal graphical connectivity metrics (gray boxes). .................................... 14
Figure 2. 3 A layout of 10 clusters for the 64 EEG electrodes. Circles represent electrodes; lines
separate different clusters. The 10 medial electrodes are grouped in both left and right regions;
the overlapped regions are marked by the two dashed lines. LF: Left frontal; RF: Right frontal;
LC: Left central; RC: Right central; LP: Left parietal; RP: Right parietal; LT: Left temporal; RT:
Right temporal; LO: Left occipital; RO: Right occipital. ............................................................. 20
Figure 2.4 Topographic maps of group-averaged (n=45), baseline-normalized, and sham-
subtracted changes in mPowerss (see eq. (2-2)) in delta (1-4 Hz), theta (4-8 Hz), alpha (8-13
Hz), beta (13-30 Hz), and gamma (30-70 Hz) bands during the first 4 minutes of tPBM (Stim1),
second 4 minutes of tPBM (Stim2), and post tPBM period. Also, statistical results after the
cluster-based permutation testing are superimposed in each topographical map, showing
significant differences in mPower between the tPBM and sham stimulations during respective
three time segments and in five frequency bands with corrected significance levels of p < 0.05
(×) and p < 0.01 (*). ...................................................................................................................... 23
Figure 2.5 Three GTA-derived global graphical metrics, namely, synchronization (the 1st
column), global efficiency (the 2nd column), and small-worldness (the 3rd column), of the EEG
xiv
brain network in the beta band (13-30 Hz) under both active tPBM and sham stimulation during
Stim1 (the 1st row), Stim2 (the 2nd row), and the post period (the 3rd row). In each panel, the y
axis denotes respective metric values while the x axis presents sparsity values with an increment
of 5%. The grey bars mark sparsity values at which the corresponding graphical metrics were
altered significantly by tPBM with respect to sham based on paired t-tests (p < 0.05). Error bars
represent standard error of the mean. ............................................................................................ 25
Figure 2.6 Comparative topographical maps of 10-Cluster-distributed nodal network metrics at
the beta band (13-30 Hz). The comparison was made between tPBM and sham stimulation
conditions for each of the five nodal metrics, namely, the clustering coefficient (the 1st column),
nodal local efficiency (the 2nd column), nodal efficiency (the 3rd column), betweenness centrality
(the 4th column), and degree centrality (the 5th column) of the EEG brain network during Stim1
(the 1st row), Stim2 (the 2nd row), and the post period (the 3rd row). LF: Left frontal, RF: Right
frontal, LC: Left central, RC: Right central, LP: Left parietal, RP: Right parietal, LT: Left
temporal, RT: Right temporal, LO: Left occipital, RO: Right occipital. Red color represents
tPBM > sham; blue color indicates tPBM < sham with p < 0.05 (FDR corrected). ..................... 27
Figure 3. 1 (a) Dual-mode (bbNIRS and EEG) head probe setup, showing two separate channels
with two sets of fiber bundles that were connected to (b) the 2-channel bbNIRS. While an EEG
cap on the head is observable, the EEG data are not the topic/subject of this paper. The bbNIRS
datasets used for this study were taken during 7-min eyes-closed conditions with the setup shown
above. ............................................................................................................................................ 42
Figure 3. 2 A data processing flow chart with five steps. Step 1: [HbO] and [CCO]
quantification at each time point and time series (blue boxes); Step 2: amplitude and phase
xv
decomposition using multi-taper method (yellow box); Step 3: quantification of spectral
amplitudes (SA) for endogenic, neurogenic, and myogenic (E/N/M) frequency bands (orange
box); Step 4: determination of four types of coherences for each E/N/M bands (green box).
Steps 1 to 4 were repeated for each of 26 participants (outlined by the dotted box) and then for 5
sets of the measurements (outlined by the solid box). The bottom dashed box marks Step 5,
showing several statistical analyses, including one-way ANOVA, paired t-tests, and two one-
sided tests (TOST) used to identify group-level features for SA and respective coherence
parameters (gray box). .................................................................................................................. 43
Figure 3. 3 (a) and (b) illustrate an example of time-domain representation of [HbO] and
[CCO] signals, respectively, with a frequency band of 0.005-0.2 Hz over a period of 7 min.
This set of time series was derived after processing Step 1 from one channel of the subject’s
dataset. (c) and (d) show the frequency-domain spectral amplitudes for [HbO] and [CCO],
respectively, quantified using Steps 2 and 3. Blue, green, and red indicate endogenic, neurogenic,
and myogenic bands, respectively. ............................................................................................... 49
Figure 3. 4 Resting-state prefrontal SAHbO (in µM) of the left and right forehead averaged over
(a) a combined set of grand/total measurements (n=130) and (b) each individual set of five
measurements (n=26 per set) at endogenic (E; 0.005-0.02 Hz), neurogenic (N; 0.02-0.04 Hz),
and myogenic (M; 0.04-0.2 Hz) frequency bands. Similarly, resting-state prefrontal SACCO (in
µM) of the left and right forehead averaged over (c) the combined set of measurements (n=130)
and (d) each individual set of five measurements (n=26 per set) at E/N/M bands. p-values shown
for each group of bars in (b) and (d) represent ANOVA results. All error bars are based on the
standard error of the mean. *: p<0.05. I values represent intraclass correlation coefficients for
each group. .................................................................................................................................... 50
xvi
Figure 3. 5 Resting-state prefrontal bCONHbO and bCONCCO averaged over (a) the combined set
of measurements (n=130) and (b) each separate set of five measurements (n=26 per set) over E
(0.005-0.02 Hz), N (0.02-0.04 Hz), and M (0.04-0.2 Hz) bands. p-values shown on top of each
group of the bars in (b) represent one-way ANOVA results. All error bars indicate the standard
error of the mean. ***: p<0.001. I values represent intraclass correlation coefficients for each
group. ............................................................................................................................................ 53
Figure 3. 6 Left and right resting-state prefrontal uCOPHbO-CCO obtained from (a) combined
grand group (n=130) and (b) separate groups (n=26 each) over endogenic (0.005-0.02 Hz),
neurogenic (0.02-0.04 Hz), and myogenic (0.04-0.2 Hz) frequency bands. p-values above each
group of bars in (b) represent results from ANOVA test. The error bars indicate the standard
error of the mean. **: p < 0.01. I values represent intraclass correlation coefficients for each
group. ............................................................................................................................................ 55
Figure 4. 1 (a) Experiment setup including two channels of bbNIRS on the lateral forehead and
EEG cap. The electrophysiological data collected by EEG is not used in this paper. (b) 2-bbNIRS
light source, spectrometer, and bundle setup illustration. (c) Protocol for this study consists of 5
visits with 7-minute eyes-closed pre-stimulation, 8 minutes of randomized tPBM (R800, L800,
or R850) or sham (RS or LS), and 7-minute post-stimulation. The bbNIRS data is collected pre-
and post-stimulation. ..................................................................................................................... 70
Figure 4. 2 Sham-subtracted tPBM-induced (a) SAHbO,SS and (b) SACCO,SS (in percent) over the
ipsilateral and contralateral sides of the prefrontal cortex for different tPBM conditions, namely,
R800, L800, R850 at endogenic (E; 0.005-0.02 Hz), neurogenic (N; 0.02-0.04 Hz), and
xvii
myogenic (M; 0.04-0.2 Hz) frequency bands. Error bars represent the standard error of the mean
(n=26). *: p<0.05, **: p<0.01 obtained from one-sample t-test. .................................................. 76
Figure 4. 3 Sham-subtracted tPBM-induced prefrontal (a) bCONHbO,SS and (b) bCONCCO,SS for
different tPBM conditions, namely, R800, L800, R850 at endogenic (E; 0.005-0.02 Hz),
neurogenic (N; 0.02-0.04 Hz), and myogenic (M; 0.04-0.2 Hz) frequency bands. Error bars
represent the standard error of the mean (n=26). *: p<0.05, **: p<0.01 obtained from one-sample
t-test............................................................................................................................................... 78
Figure 4. 4 Sham-subtracted tPBM-induced prefrontal (a) uCOPIpsi,SS and (b) bCONContra,SS for
different tPBM conditions, namely, R800, L800, R850 at endogenic (E; 0.005-0.02 Hz),
neurogenic (N; 0.02-0.04 Hz), and myogenic (M; 0.04-0.2 Hz) frequency bands. Error bars
represent the standard error of the mean (n=26). *: p<0.05, **: p<0.01 obtained from one-sample
t-test............................................................................................................................................... 79
Figure 5. 1 (a) Experiment setup including two channels of bbNIRS on the lateral forehead and
EEG cap. The electrophysiological data collected by EEG is not used in this paper. (b) 2-bbNIRS
light source, spectrometer, and bundle setup illustration. (c) Protocol for this study consists of 5
visits with 7-minute eyes-closed pre-stimulation, and in two visits followed by 8 minutes of
randomized tPBM (L800) or sham (LS), and 7-minute post-stimulation. The bbNIRS and EEG
data are collected pre- and post-stimulation. ................................................................................ 93
Figure 5. 2 A data processing flow chart, including steps for (1) EEG data analysis (blue boxes),
(2) bbNIRS data analysis (orange boxes), (3) bilateral hemodynamic and metabolic connectivity
(yellow boxes), and (4) integrated EEG-bbNIRS data analysis including epoch synchronization
and unilateral physiological network analysis (green boxes). ...................................................... 94
xviii
Figure 5. 3 Schematic illustration of the process for down-sampling EEG data. (a) EEG time
series segmented in 1.5-s epochs. (b) power spectral density (PSD) obtained from each epoch.
The area under the curve in the beta band is calculated and used to construct (c) down-sampled
EEG time series............................................................................................................................. 97
Figure 5. 4 Adjacency matrices and graphical illustration of resting state metabolic-vascular
network on the prefrontal cortex obtained from 2-bbNIRS. Three boxes represent (a) endogenic,
(b) neurogenic, and (c) myogenic components of ISO. The nodes in the network are HbOleft,
CCOleft, HbOright, and CCOright. ...................................................................................................... 99
Figure 5. 5 Directed bilateral hemodynamic and metabolic connectivity between left and right
prefrontal cortex in endogenic, neurogenic, and myogenic bands. Error bars represent the
standard error of the mean. *: p-value < 0.05. ............................................................................ 100
Figure 5. 6 Adjacency matrices and graphical illustration of resting state physiological network
on the prefrontal cortex obtained from dual-mode 2-bbNIRS and EEG dataset. Three columns
represent (a) endogenic, (b) neurogenic, and (c) myogenic components of ISO over left and right
PFC. The nodes in the network are HbO, CCO, and EEG. ........................................................ 101
Figure 5. 7 Changes in directed bilateral (a) hemodynamic and (b) metabolic connectivity
between left and right prefrontal cortex in endogenic, neurogenic, and myogenic bands in
response to left prefrontal 800-nm tPBM. Error bars represent the standard error of the mean. *:
p-value < 0.05, **: p-value < 0.01. ............................................................................................. 102
Figure 5. 8 Adjacency matrices and graphical illustration of changes in unilateral coupling in the
physiological network on the prefrontal cortex in response to left 800-nm tPBM. Three columns
represent (a) endogenic and (b) neurogenic components of ISO over left and right PFC. The
xix
nodes in the network are HbO, CCO, and EEG. Drawn links are the uCOPj,i,ss in which
uCOPj,i,tPBM and uCOPj,i,sham return a p-value less than 0.05. ................................................. 103
Figure B1 Schematic flow chart of spectral analysis for the quantification of SA and coherence.
For demonstration, two time series, Δ[HbO] and Δ[CCO], are used as separate input signals with
a time period of ‘t’. Blue and orange blocks represent frequency analysis steps operated on signal
1 (i.e., Δ[HbO]) and signal 2 (i.e., Δ[CCO]), using “ft_freqanalysis” function (outlined by black
dashed boxes). The word of “double” and “complex” indicates a real number with double
precision and a complex number, respectively. Furthermore, green blocks represent connectivity
analysis steps operated on the frequency-domain outputs of the two signals, using
“ft_connectivityanalysis” function (red dashed box). ................................................................. 131
Figure B2 (a) A 7-min time series of [HbO] derived from one channel of 2-bbNIRS of a
subject’s dataset. The three panels on the right were obtained after Butterworth band-pass
filtering of the original signal in the three predefined E/N/M bands, namely, 0.005-0.02 Hz, 0.02-
0.04 Hz, and 0.04-0.2 Hz, respectively. ...................................................................................... 132
xx
List of Tables
Table 2.1 List of references that reported EEG responses to tPBM with related measurement and
analysis parameters ......................................................................................................................... 9
Table 2.2 Parameters for 8-min tPBM and sham by a 4.2-cm diameter laser at 1064 nm over the
right eyebrow ................................................................................................................................ 13
Table 2.3 Comparisons of global graphical metrics among three studies .................................... 34
Table 2.4 Comparisons of nodal graphical metrics among three studies ..................................... 34
Table 3. 1 Grand averages of SAHbO over all measurements (n=130) on the left and right
forehead across three ISO frequency bands .................................................................................. 51
Table 3. 2 Grand averages of SACCO over all measurements (n=130) on the left and right
forehead across three ISO bands ................................................................................................... 52
Table 3. 3 Resting-state prefrontal connectivity (bCONHbO and bCONCCO) averaged over the
grand data set (n=130) at E/N/M band .......................................................................................... 54
Table 3. 4 Prefrontal uCOPHbO-CCO values on the left and right cortical regions averaged over the
grand set of measurements (n=130) at each of the E/N/M bands ................................................. 56
Table 3. 5 Measurable ISO features for characterization of the prefrontal human brain at rest .. 61
Table 3. 6 Measurable ISO features for characterization of the prefrontal human brain at rest
with similar values between male and female groups (the reported values are average of all
measurements, including male and female subjects) .................................................................... 64
Table 4. 1 Laser stimulation parameters for active tPBM and sham ........................................... 71
xxi
Table A1 Description of global and local graphical metrics significantly altered by tPBM ([65,
82]) .............................................................................................................................................. 128
1
Chapter 1
Introduction
1.1 Significance and Specific Aims
A type of photobiomodulation technique known as transcranial photobiomodulation (tPBM)
targets the human brain with near-infrared (NIR) light [1] and human performance on a variety of
cognitive tasks has been shown to improve in several studies using tPBM with different
wavelengths for laser and LED [2-6]. Our group has previously shown that tPBM enabled
significant upregulation in concentrations of oxidized cytochrome-c-oxidase ([CCO]) and
hemoglobin oxygenation ([HbO]) during and after tPBM on the right forehead using broadband
near-infrared spectroscopy (bbNIRS) [7-9]. These studies support the hypothesis that tPBM photo-
oxidizes CCO to boost the metabolism of cells [1, 10, 11].
Electroencephalogram (EEG) is commonly used to monitor and record the electrical activity of
neurons. Brain functional (topographical) connectivity, which shows the communication between
regions of the brain, can be utilized as a measure to understand the effects of stimuli, diseases, and
cognitive decline, in resting-state or task-based measurements [13-20]. Understanding the
correlation between different physiological systems, namely electrophysiological, hemodynamic,
and metabolic systems in the human brain, has been an attractive topic among researchers in recent
decades [21-25]. In this study, Neurovascular, neurometabolic, and metabolic-vascular coupling
is calculated using oscillations of EEG, [HbO], and [CCO] time series to investigate the
connectivity between different physiological systems in a specific region of the brain.
The goal of this study is to examine how tPBM can alter the topographical and physiological
connectivity in the electrophysiological, hemodynamic, and metabolic systems of the human brain
2
in resting state. As the first step, I found the regions and clusters in EEG sensor space that have
been affected the most by tPBM during and after stimulation using both cluster-based power
analysis and graph-based connectivity analysis. In the second step, I assessed the power and
connectivity between bilateral frontal regions of the human brain in resting state and in response
to different conditions of tPBM using two-channel bbNIRS system. Finally, I quantified the
modulation in metabolic-vascular and physiological network in rest and in response to left frontal
tPBM by determining the effective connectivity between electrophysiological, hemodynamic, and
metabolic oscillations of the human brain using a dual-modality (i.e., EEG and bbNIRS) dataset.
1.1.1 The need for a topographical mapping of the brain’s electrophysiological network
modulation in response to tPBM
The electrophysiological response of the human brain to tPBM is not well studied and understood.
All previous studies utilized either laser or LED sources of NIR light and recorded the
electrophysiological responses using a 19-channel (i.e., the 10-20 EEG electrode placement) or
64-channel (i.e., the 10-10 electrode placement) EEG system. Power spectral density (PSD)
analysis and graph theory analysis (GTA) are two major data analysis methods. Graph theory is a
branch of mathematics that can serve as a theoretical tool for quantifying the specific features of
network architecture (topology). The outcome of GTA can provide information on the anatomical
localization of areas responding to given stimuli or tasks (topography), or human brain functional
connectivity and networks [26].
Although the previous studies have consistently reported that tPBM enables alterations in EEG
PSDs, different tPBM protocols, such as laser versus LEDs, 19- versus 64-channel electrodes, and
PSD versus GTA analysis, have given rise to sparse and incomparable findings. For example, only
two studies reported 64-channel EEG responses to tPBM [27, 28], but they did not examine tPBM-
3
induced topological alterations in brain connectivity. On the other hand, based on 19-channel EEG,
two recent studies reported tPBM-induced modulations of global network parameters [29] and
alterations in brain connectivity between the two hemispheres [30], respectively. However, light
stimulation was delivered at different cortical regions, either locally at the right forehead or at
multiple scalp sites near the default mode network, and by LEDs.
The distinct focus of this study was to investigate tPBM-induced modulations of EEG functional
connectivity by performing GTA on a 64-channel EEG, followed by quantification of changes in
brain connectivity in global network metrics and 10 nodal/local regions. Specifically, I identified
10 cortical clusters in the sensor space that were most affected by tPBM during and after
stimulation. Next, cluster-based permutation testing after PSD analysis was performed to identify
topographical regions that were significantly modulated by tPBM. Then, global and nodal
graphical measures or metrics were obtained based on GTA, revealing respective augments of
functional connectivity, brain network, and information pathways in response to tPBM.
1.1.2 The need for a topographical mapping of the resting state prefrontal cortex
hemodynamic and metabolic network
Many studies have focused on investigating the mechanism of cerebral metabolic activity and have
found vasomotion to be a major source of metabolic and hemodynamic modulations [31-35].
Vasomotion is a spontaneous oscillation that originates from the blood vessel wall with an infra-
slow oscillation (ISO) of 0.005-0.2 Hz [36, 37]. In addition, a correlation is found between the ISO
of cerebral metabolic activities and human cognitive functions [38]. Furthermore, vasomotion
malfunction has been observed in older adults and in patients with different diseases, such as
atherosclerosis [39], cardiovascular disease [40], and Alzheimer’s disease [41]. Thus, it may be
beneficial to quantify and characterize cerebral metabolism in the ISO range, which may provide
4
better insight into neurophysiological mechanisms and discover features that differ between
healthy humans and patients with brain disorders.
In the present exploratory study [42], I hypothesized that 2-bbNIRS, along with frequency-domain
analysis, one can quantify prefrontal cortical connectivity and coupling of ISO in the resting human
brain. Specifically, the features analyzed from the 2-bbNIRS time series included (1) resting-state
spectral amplitude (SA) of bilateral cortical hemodynamic and metabolic (i.e., SAHbO_i and
SACCO_i) activities, where i represents either the left or right prefrontal region, (2) bilateral
hemodynamic connectivity (bCONHbO), (3) bilateral metabolic connectivity (bCONCCO), and (4)
coupling between cerebral hemodynamic and metabolic activities on the unilateral side (uCOPHbO-
CCO_i) of the prefrontal cortex over the three ISO frequency bands. Then the results support this
hypothesis by presenting relatively stable and consistent values for these features in healthy young
humans, revealing the translation potential of these features for future clinical applications.
1.1.3 The need for a topographical mapping of the prefrontal cortex hemodynamic and
metabolic network in response to different wavelengths and stimulation sites of
tPBM
Our group recently introduced a set of hemodynamic and metabolic characteristics quantified by
frequency-domain spectral amplitude and connectivity analysis of hemodynamic and metabolic
ISO activity of prefrontal cortex, assessed by a dual-channel bbNIRS setup [43]. These metrics are
(1) bilateral hemodynamic (i.e., [HbO]) connectivity, (2) bilateral metabolic (i.e., [CCO])
connectivity, (3) unilateral hemodynamic-metabolic coupling on the left and (4) right side of the
prefrontal cortex. In addition, we have demonstrated that these constant and relatively reproducible
characteristics can be considered potential biomarkers to identify neurological disorders and
diseases [44]. Furthermore, we have shown distinct alterations in these metrics as well as Δ[HbO]
5
and Δ[CCO] ISO spectral amplitudes across all three frequency bands in response to 1064-nm
tPBM. Modulation of Δ[HbO] and Δ[CCO] in response to tPBM is also demonstrated to be
wavelength-dependent in another study [12]. As reported in [12], 800-850 nm wavelengths enable
CCO to be more stimulated with an increased concentration. On the other hand, 1064 nm laser
have demonstrated its effect on enhancement of human cognition. 1064 nm is not at the absorption
peak of CCO, but it has much less light scattering ability or a smaller light scattering coefficient
than that at 800850 nm light which leads to deeper penetration in tissue. Moreover, no behavioral
or physiological alteration is reported in response to left prefrontal tPBM. Thus, it would be
beneficial to investigate the effect of the laser’s wavelength and stimulation site on the alteration
of the proposed metrics.
In this study, young healthy human participants were at rest while [HbO] and [CCO] time
series with a sampling frequency of 0.67 Hz were acquired using a 2-bbNIRS system from two
sides of the prefrontal cortex in pre- and post-tPBM. These time series were then analyzed to
quantify the amplitude and coherence of hemodynamic and metabolic ISO over three frequency
bands. In the first set of analyses, I assessed the spectral amplitude of hemodynamic and metabolic
activity (SAHbO and SACCO) over the three ISO frequency bands. Then, four physiological metrics
were used to characterize the connectivity/coupling between each pair of signals. These measures
include (1) bilateral hemodynamic connectivity (bCONHbO), (2) bilateral metabolic connectivity
(bCONCCO), (3) coupling between cerebral hemodynamic and metabolic activities on the
ipsilateral side to the stimulation (uCOPIpsi), and (4) coupling between cerebral hemodynamic and
metabolic activities on the contralateral side to the stimulation (uCOPContra), of the prefrontal cortex
over the three ISO frequency bands. Five separate visits with different conditions were used for
each participant including 8 minutes of (1) Right prefrontal 800-nm laser, (2) Right prefrontal 850-
6
nm laser, (3) Right prefrontal sham, (4) Left prefrontal 800-nm laser, (5) Left prefrontal sham in a
randomized order.
Then results support the hypothesis that hemodynamic and metabolic ISO is significantly
modulated by tPBM. This modulation is distinct for each frequency band, can be local or
bilateral/global, and in some cases is closely related to the wavelength and stimulation location.
These observations can be beneficial for further investigation of the mechanism behind cerebral
metabolism as well as wavelength- or location-specific cognitive function improvement or
treatment of neurological disorder/disease based on the modulated metric and frequency band.
1.1.4 The need for a physiological connectivity assessment of the brain’s
electrophysiological, hemodynamic, and metabolic network modulation in response
to tPBM
Investigation of the intertwined behavior of hemodynamic and metabolic ISO activities can be
facilitated by assessment of metabolic-vascular coupling (MVC), and utilizing EEG and bbNIRS
in parallel opens the door to the assessment of neurometabolic coupling (NMC). Although these
different metrics can be assessed separately, they can be observed as a semi-complex physiological
network of a specific region in the cerebral cortex where electrophysiological (here, the beta band
of EEG), hemodynamic, and metabolic activity are the nodes, and these coupling metrics are the
links between them. Furthermore, the level by which the hemodynamic and metabolic activities
on lateral sides of prefrontal cortex are interacting can be quantified by the topographical
functional connectivity.
In this study, a dual-mode setup (EEG and bbNIRS) was utilized to assess the topographical
effective connectivity between lateral prefrontal hemodynamic and metabolic ISO to identify any
directionality between these activities in young healthy human adults at rest. In addition, local
7
MVC was quantified to determine the effective coupling between hemodynamic and metabolic
ISO. Furthermore, I constructed two local physiological networks on the lateral prefrontal cortices
consisting of a beta band of EEG, Δ[HbO], Δ[CCO] as nodes, and NVC, NMC, and MVC as links.
Finally, I quantified the tPBM-induced changes in the abovementioned networks and investigated
any possible alterations in information flow among these nodes.
Then the results support the hypothesis that resting-state interactions between
electrophysiological, hemodynamic, and metabolic ISO is mostly bi-directionally balanced and
they can be significantly altered by tPBM. This modulation can be local or bilateral/global. These
observations can be beneficial for further investigation of the mechanism behind NVC, NMC, and
MVC as well as location-specific cognitive function improvement or treatment of neurological
disorders/diseases based on the modulated metrics.
8
Chapter 2
Neuromodulation of brain topography and network
topology by prefrontal transcranial photobiomodulation
Sadra Shahdadian, Xinlong Wang, Hashini Wanniarachchi, Hanli Liu
(This chapter is a manuscript that is under revision in Journal of Neural Engineering)
2.1 Introduction
Transcranial photobiomodulation (tPBM) is a non-invasive neuromodulation technique that
delivers near-infrared (NIR) light to the human brain using lasers or light-emitting diode (LED)
clusters [11, 45-47]. Recent studies have demonstrated the promising effects of tPBM in treating
traumatic brain injuries [47-51], psychiatric or neurological disorders [52-55], and in enhancing
cognitive performance in normal humans [2-6, 56]. To better understand the underlying
mechanism of tPBM, neurophysiological measurements of the human brain were performed non-
invasively from healthy human controls using an optical spectroscopy approach before, during,
and after prefrontal tPBM. The results of these measures quantified and demonstrated tPBM-
induced increases in mitochondrial metabolism (i.e., the redox state of cytochrome-c-oxidase
([CCO])) and hemodynamic oxygenation (i.e., oxygenated haemoglobin ([HbO])) [7-9].
Moreover, it was shown that the increases in both [CCO] and [HbO] were not caused by the
thermal effects of the tPBM [57], hardware-related noise, or drift [58, 59]. All these published
reports provide strong support that tPBM facilitates the photo-oxidization of mitochondrial CCO
to boost the cellular metabolism of neurons [1, 10, 11].
However, the electrophysiological response of the human brain to tPBM is not well studied
and understood. Table 2.1 summarizes the recently published articles that reported scalp
9
electroencephalography (EEG) responses to tPBM with the respective measurement and analysis
parameters. All studies utilized either laser or LED clusters of NIR light sources and recorded the
electrophysiological responses using a 19-channel (i.e., the 10-20 EEG electrode placement) or
64-channel (i.e., the 10-10 electrode placement) EEG system. Pruitt et al. [12] have shown that the
laser and LED have different effects on the stimulated tissue in terms of changes hemodynamic
and metabolic activity which is mainly due to significantly lower irradiance of LED sources, which
is a result of broader bandwidth of LED compared to laser. The table also lists two major data
analysis methods, namely, power spectral density (PSD) analysis and graph theory analysis (GTA).
Graph theory is a branch of mathematics that can serve as a theoretical tool for quantifying the
specific features of network architecture (topology). The outcome of GTA can provide information
on the anatomical localization of areas responding to given stimuli or tasks (topography), or human
brain functional connectivity and networks [26].
Table 2.1 List of references that reported EEG responses to tPBM with related measurement and analysis parameters
Refs. Authors Source of tPBM
Location of
tPBM
# of
Subjects # of Channels
PSD analysis
Graph-theory based
connectivity analysis
[60]
Berman,
et al., 2017
Multiple LED clusters
(1070 nm) Whole head
19 19
No. It was based on
qEEG analysis No
[30] Ghaderi,
et al., 2021 1 cluster of LED
(850 nm)
Right
forehead 40 19
No.
Yes; changes in connectivity
within each of or between two
hemispheres
[61] Spera, et al.
4 clusters of LED
(830 nm)
Bilateral
frontal
10 19 Yes, with topography
No
[5] Vargas, et al.
2017 Laser
(1064 nm) Right
forehead 12 19 Yes, but no
topography No
[27]
Wang, et al.,
2021
Laser
(1064 nm) Right
forehead 46 64 Yes, with topography
No
[28] Wang, et al.
2022 Laser
(1064 nm)
Right
forehead 44
64
No. It was based on
singular value
decompensation
No
[29] Zomorrodi, et
al., 2019 3 clusters of LED
(810 nm)
3 default
mode
locations 20 19 Yes, with topography
Yes; changes in global
connectivity parameters only
Current study Laser
(1064 nm)
Right
forehead 45 64 Yes, with topography
Yes; changes of connectivity in
global network metrics and 10
nodal regions
10
PSD is the most common method for analysing EEG data and gives rise to an absolute
oscillation power spectrum of local electrophysiological signals in the frequency range of 0.5 to
70 Hz or higher. Frequency-dependent PSD values facilitate a better understanding of the effects
or impacts of external stimuli, cognitive decline, disturbances in consciousness, and certain brain
disorders in the human brain [13, 16-20]. However, statistical analysis of multi-channel EEG is
challenging when conducting multi-variable comparisons (e.g., 64 comparisons for 64 channels)
[62, 63]. Cluster-based permutation testing is an established method for minimizing type-I errors
in null hypothesis testing for these datasets. This method is based on the fact that the
electrophysiological time series of brain oscillations is highly correlated with neighbouring
channels [63]. Using cluster-based permutation testing, we identified topographical clusters of
EEG channels on the human scalp template, where frequency-specific EEG oscillation powers
were significantly altered between two different conditions (i.e., active tPBM versus sham
intervention). Consequently, the corresponding brain regions with modulated powers were
identified.
Topologically, GTA is a practical and quantitative approach for characterizing functional
connectivity and networks in the human brain [64]. In this method, a network is considered a
mathematical representation of a real-world complex system and is defined by the composition of
nodes (vertices) and links (edges) between pairs of nodes. The outcome of GTA measures mainly
represents the functional integration, segregation, and centrality of the network, all of which can
topologically characterize changes in brain functional connectivity in both global and nodal
regions [14, 15, 65]. When GTA is used to analyse EEG data, the scalp locations of the EEG
electrodes represent the network nodes, and the links among the electrodes represent the functional
connections between these nodes [14].
11
In general, both PSD and GTA-derived parameters or metrics can provide instructive
information on functional brain networks in the resting state or under external neuromodulation.
Because multi-channel EEG signals contain rich information in the temporal, spectral, and spatial
domains, these pieces of information can be grouped temporally and/or spectrally to visualize brain
activation and networks in topographical clusters and regions [66, 67].
Although the studies summarized in Table 2.1 have consistently reported that tPBM enables
alterations in EEG PSDs, different tPBM protocols, such as laser versus LEDs, 19- versus 64-
channel electrodes (i.e. instrument with low cost and low setup time versus high spatial resolution
data), and PSD versus GTA analysis, have given rise to sparse and incomparable findings. For
example, only two studies reported 64-channel EEG responses to tPBM [27, 28], but they did not
examine tPBM-induced topological alterations in brain connectivity. On the other hand, based on
19-channel EEG, two recent studies reported tPBM-induced modulations of global network
parameters [29] and alterations in brain connectivity between the two hemispheres [30],
respectively. However, light stimulation was delivered at different cortical regions, either locally
at the right forehead or at multiple scalp sites near the default mode network, and by LEDs.
As shown in Table 2.1, the distinct focus of this study was to investigate tPBM-induced
modulations of EEG functional connectivity by performing GTA on a 64-channel EEG, followed
by quantification of changes in brain connectivity in global network metrics and 10 nodal/local
regions. Specifically, we identified 10 cortical clusters in the sensor space (scalp EEG) that were
most affected by tPBM during and after stimulation. Next, cluster-based permutation testing after
PSD analysis was performed to identify topographical regions that were significantly modulated
by tPBM. Then, global and nodal graphical measures or metrics were obtained based on GTA,
12
revealing respective augments of functional connectivity, brain network, and information
pathways in response to tPBM.
2.2 Materials and methods
2.2.1 Participants
A total of 49 healthy human subjects (30 men and 19 women; 26 ± 8.8 years of age) were enrolled
from the local community of the university. The subjects were seated on a chair with closed eyes
in resting state during the measurement. Four subjects were removed from the dataset because of
self-reported or observed tiredness or sleepiness (through subjects’ movement and EEG signal)
during the measurement, resulting in 45 participants being considered for further data analysis.
The participants were instructed to refrain from consuming any caffeinated drinks for at least 3 h
before each experiment. All measurements were obtained with informed consent from each
participant.
2.2.2 Experimental setup and protocol
In this study, tPBM and sham experiments were performed using a continuous-wave laser at 1064
nm (Model CG-5000 Laser, Cell Gen. Therapeutics LLC, Dallas, TX, USA), which was cleared
by the Food and Drug Administration (FDA). The laser was delivered to each participant’s right
forehead above the eyebrow with an aperture of 4.2 cm in diameter and a period of 8 min. A sham
experiment was performed with the laser device turned on but set to 0.1 W also for 8 min, while
Figure 2.1 A crossover experimental protocol for tPBM and sham experiments (n=45) with
simultaneous EEG recording. The participants were at wakeful resting state with eyes closed.
13
the laser aperture was covered with a black-colour cap. A power meter was used to confirm that
the output power in the presence of the cap was zero. Table 2.2 lists the light irradiance (W/cm2),
light fluence (J/cm2), and total energy (dose) (J) delivered by the 8-min active and sham tPBM,
respectively. The participants wore protective goggles throughout the experiment.
Table 2.2 Parameters for 8-min tPBM and sham by a 4.2-cm diameter laser at 1064 nm over the right
eyebrow
Stimulation
(W/cm2)
Fluence
(J/cm2)
Total
Dose (J)
tPBM 0.25 120 1662
sham 0 0 0
EEG data were collected during the entire experiment using a 64-channel EEG instrument
(Biosemi, Netherlands). Each subject wore an EEG cap with 64 electrodes positioned according
to the standard 10-10 EEG electrode placement [68]. Electrode gel (MFI Medical Equipment Inc.,
CA, USA) was used to increase the signal-to-noise ratio of the recorded EEG data. The recorded
EEG time series were directed to a desktop computer using electrical cables.
The stimulation protocol (figure 2.1) consisted of a 2-min baseline (pre), an 8-min stimulation
(active tPBM or sham), and a 3-min recovery (post) period. The EEG data were acquired at either
256 Hz or 512 Hz; all 512-Hz data were down-sampled to 256 Hz during data pre-processing.
tPBM was delivered near electrodes FP2 and AF8 under either sham or active conditions.
The study was conducted in a single-blind crossover design, with each subject completing both
sham and active tPBM experiments in a random order, with a minimum 5-day interval between
the two experiments. All participants were asked whether they felt drowsy during the experiment
and whether they noticed any heat sensations at the stimulation site.
14
2.2.3 Overview of data processing steps
Each of the recorded EEG datasets represented a 13-min time series of 64 channels during both
the active and sham tPBM experiments for 45 participants (after exclusion of 4 subjects). Because
the data processing and analysis procedures included multiple steps in this study, we outline a flow
chart of these steps in figure 2.2 to better guide the reader through them easily. Briefly, there were
five steps in data processing: (1) data pre-processing, (2) PSD-based power analysis to obtain
frequency-specific power topography, (3) graphical edge formation based on the “imaginary part
of coherence” analysis, (4) GTA-based analysis to quantify global connectivity metrics altered by
tPBM, and (5) GTA-based analysis to identify local or nodal graphical metrics changed by tPBM
in 10 cluster regions.
2.2.4 Data pre-processing for EEG time series
EEGLAB, an open-source software package based on MATLAB, was used to pre-process the EEG
data. First, EEGLAB's "filtfilt" function was used to band-pass filter the 64-channel raw EEG data
Figure 2. 2 A data processing flow chart, including steps for (1) data pre-processing (blue boxes),
(2) PSD-based analysis and permutation tests to form power topographies (orange boxes), (3)
graphical edge formation based on the “imaginary part of coherence” analysis, (4) GTA-based
assessment for global graphical connectivity metrics (pink boxes), and (5) GTA-based
assessment for nodal graphical connectivity metrics (gray boxes).
15
to ensure zero phase distortion of the time series, followed by a 60-Hz notch filter to eliminate line
noise. Second, each EEG time series was re-referenced with respect to the voltage average over
all the 64 channels. Next, robust PCA (rPCA) was applied to identify and remove significant signal
artifacts and/or outliers from EEG signals [69, 70]. Finally, to further remove noise and artifacts
[71, 72] such as eye movements, saccades, and jaw clenching, independent component analysis
(ICA) [73, 74] was used.
To quantify the dose-dependent responses of EEG to tPBM/sham, each artifact-free time series
was divided into four temporal sections: (1) the last minute of the 2-min baseline before the onset
of active or sham stimulation (pre), (2) the first 4-min stimulation period (Stim1), (3) the second
4-min stimulation period (Stim2), and (4) the first 2-min recovery (post). The pre-processed data
were then used for both PSD analysis and GTA-based assessment of graphical metrics during each
temporal segment.
2.2.5 EEG power spectral density and changes in power
With the use of the “Pwelch” function (with a 4-sec window and 75% overlap [75]) in EEGLAB,
a PSD curve of artifact-free time series for each EEG channel in each time section was calculated.
The frequency-specific PSD bandwidths were then selected to cover the delta (1-4 Hz), theta (4-8
Hz), alpha (8-13 Hz), beta (13-30 Hz), and gamma (30-70 Hz) bands. Next, the mean power change
at each of the five frequency bands, f,
mPowerf, during each of the three temporal segments
(Stim1, Stim2, and post) was normalized to the last minute of its respective baseline (pre), as
expressed below [27]:
=
 ×100% =×
 × ×100%
= 
 ×100%. (2 1)
16
where superscript “f” denotes the five frequency bands, subscript “irepresents the three temporal
segments (Stim1, Stim2, or post), subscript “prerepresents the baseline segment, fband denotes
the bandwidth of a chosen frequency band for PSD calculations, and PSDi and PSDpre indicate
bandwidth-averaged PSD values. Note that
mPower is a relative value or percentage change in
bandwidth-averaged power caused by tPBM or sham treatment (see the first two orange boxes in
figure 2.2). To illustrate the difference in
mPower between the two conditions, we further
calculated the sham-subtracted (ss) and tPBM-induced change in power (
mPowerss) at each
electrode for each of the five frequency bands within each of the three temporal periods:

=

. (2 2)
2.2.6 Statistical analysis for EEG power topography
Because EEG data taken at neighbouring time points and spatial channels are highly correlated, it
is necessary to perform advanced statistical analysis to remove such correlations and for multi-
variable comparisons. For this purpose, we utilized several functions (including “ft_freqstatistics”)
available in the FieldTrip toolbox [76, 77] to perform cluster-based permutation tests for statistical
comparisons of changes in the EEG power (i.e.,
mPowerss f as shown in Eq. (2-2)) among the 64
electrodes in each of the five frequency bands within each of the three temporal periods. In
principle, cluster‐based permutation tests have two components: The first is the cluster‐forming
algorithm, which converts one high-dimensional observation into a quantifiable summary of its
cluster structure. The second one creates a surrogate null distribution, against which the observed
data is compared to obtain p-values [78].
Following this method, we first grouped electrodes as clusters within a given scalp distance
(e.g., 4-5 cm), followed by identification of the EEG channels whose
mPowerssf values were
significantly different from zero based on parametric or non-parametric testing for each electrode
17
at a significance level of 0.05. Second, a statistical evaluation was performed by taking the sum of
the t-values over each cluster. Third, the summed t-value was compared to a null distribution. The
null distribution for both permutation tests and cluster-based correction was obtained by randomly
permuting
mPowerssf values 1000 times. (see the last orange box in figure 2.)
2.2.7 Amplitude and phase decomposition of EEG signal
For GTA-based connectivity quantification, we determined the edges or links of a graphical
network between all pairs of EEG electrodes. Based on the mathematical definition of the graphical
connectivity measure, the correlation between the phases or amplitudes of these EEG channels can
be interpreted as the functional connectivity between these points [13, 14]. Thus, we performed
amplitude and phase decompositions of the time series for all 64 channels. The amplitude and
phase of an EEG time point can be represented as a complex number [13, 15]. Moreover,
multitaper power spectral density estimation is a well-known method for extracting spectral
information from time series [79, 80]. In this study, we utilized multiple tapers, namely, Slepian
sequences, to taper the EEG signal in the time domain before performing the Fourier transform.
This part of the calculation was conducted using the “ft_freqanalysis” function within the FieldTrip
toolbox to generate a complex time series [76].
2.2.8 Imaginary part of coherence as EEG connectivity measure
Coherence, a widely used connectivity measure, is a frequency-domain function equivalent to the
time-domain cross-correlation function. The coherence coefficient is a normalized quantity
between 0 and 1, and is computed mathematically for the frequency of ω as follows [13]:
()=()
()()
=󰇻()()(()())󰇻
()
() , (2-3)
18
where Sxx, Syy denote the power estimates of signals x and y, and Sxy represents the averaged cross-
spectral density term of the two signals. In addition, A and
ϕ
are amplitude and phase of the signals
obtained from multitaper method.
The main drawback of using coherence in sensor space, especially in high-density EEG, is the
effect of volume conduction. In general, signals generated in one region of the brain can be
detected by several electrodes because of the high electrical conductivity of the brain, which leads
to an artificially high coherence value among these channels. To negate this effect, the numerator
of equation (2-3) is set to zero when the phase difference between x and y signals is 0 or π. This
method is called the ‘imaginary part of coherence, which explicitly removes instantaneous
interactions that are potentially spurious owing to volume conduction [13]. A study has shown that
the imaginary part of coherence allows for excellent detection of brain interaction from rhythmic
EEG data [81].
The pairwise connectivity values for all pairs of electrodes (64 in this study) can be represented
in an n×n (i.e., 64 × 64) adjacency matrix, where n is the number of nodes (i.e., 64 channels). The
FieldTrip toolbox facilitates the computation of the imaginary part of coherence for all pairs of
channels using the “ft_connectivityanalysis’ (see the first two green boxes in figure 2).
Each temporal segment was divided into 10-second epochs (i.e., 6 epochs per min), and the
adjacency matrices generated for all epochs in each frequency band were averaged for each of the
three temporal segments and five frequency bands for each subject individually. These averaged
matrices were then binarized by varying the sparsity level and used for further GTA-derived global
and nodal connectivity analysis as described in the following sections.
19
2.2.9 Global and nodal graphical metrics selected for GTA
Graph theory analysis enables researchers to explore topological changes in brain networks
through pair-wise functional connectivity between channels (nodes). A network and its regions
can be characterized based on three main measures: functional segregation, functional integration,
and centrality. A couple of previous studies have shown tPBM-induced alterations in global
connectivity and graph measures of the brain network with different setups and protocols [29, 30].
However, these studies focused only on the residual modulation in post stimulation with a lower
spatial EEG electrode placement (i.e., 19 channels).
In this study, we utilized GRETNA [82], a widely used GTA toolbox, to quantify the global
and nodal graphical metrics of the human brain network for individual subjects under both active
and sham tPBM in four temporal segments and five frequency bands. This step was repeated 19
times to assess the respective values of the chosen metrics under a sparsity range from 5% to 95%
with an increment of 5% (see the last two green boxes in figure 2.2).
To examine tPBM-induced effects, five global graphical measures were chosen for the
analysis: synchronization (S), global efficiency (GE), small-worldness (SW), rich-club, and
assortativity. However, no significant difference was observed between tPM and sham groups in
rich-club and assortativity. The group-level values (averaged over 45 subjects) for each global
measure at each sparsity level were statistically compared between the active tPBM and sham
conditions using a paired t-test (see the pink boxes in figure 2.2).
As noted in the Introduction, two publications reported tPBM-induced modulations of the
GTA-derived brain network [29, 30]. However, they showed alterations only in global network
metrics [29] and within each hemisphere or between two hemispheres [30], respectively. In this
study, we aimed to detect the ability of tPBM to neural-modulate regional or local brain
20
connectivity, based on 64-channel EEG measurements. Accordingly, our analysis gave rise to five
nodal graphical metrics, namely, nodal clustering coefficient (nCC), nodal efficiency (nE), nodal
local efficiency (nLE), betweenness centrality (BC), and degree centrality (DC), which were
significantly altered by tPBM across different clusters on the human scalp. It is worth noting that
another nodal metric, i.e., shortest path length, was evaluated and no statistically significant
difference was observed in comparison between sham and tPBM groups. The definitions or
explanations of the three global and five local graph metrics are listed in Table A1 in the Appendix.
2.10 Topographical clusters for nodal connectivity
Although GTA was performed on the 64-
channel EEG, resulting in quantitative
changes in nodal network metrics, the 64
EEG nodal locations were too dispersed to
locate or identify cortical and anatomical
regions on the human scalp. Thus, we focused
on 10 nodal/local sections according to
several key cortical areas of the human brain,
namely the prefrontal, central, temporal,
parietal, and occipital regions [83].
Therefore, 64 nodes (i.e., EEG channels) were divided into 10 clusters based on their locations in
the brain topography, with an average of 6-10 electrodes in each cluster. Figure 2.3 illustrates the
topographical clusters of EEG electrodes. The 10 medial electrodes were grouped twice in both
the left and right clusters (see the first two right gray boxes in figure 2.2).
Figure 2. 3 A layout of 10 clusters for the 64 EEG
electrodes. Circles represent electrodes; lines
separate different clusters. The 10 medial
electrodes are grouped in both left and right
regions; the overlapped regions are marked by the
two dashed lines. LF: Left frontal; RF: Right
frontal; LC: Left central; RC: Right central; LP:
Left parietal; RP: Right parietal; LT: Left
temporal; RT: Right temporal; LO: Left occipital;
RO: Right occipital.
21
At the subject level, each nodal graphical metric within each cluster area was obtained by
averaging the specific metric over all electrodes within the respective region (for each of the three
temporal segments and five frequency bands). To compare the changes induced by tPBM vs. sham,
nodal measures for each temporal segment were baseline-normalized by subtracting the
corresponding baseline (pre) values from those in each of the three subsequent time windows
(Stim1, Stim2, and post). Next, for each cluster region, group-level (n=45) and baseline-subtracted
nodal metric values were compared between the active and sham conditions using paired t tests.
To correct for multiple comparisons, false discovery rate (FDR) correction was performed for 10
regions with a corrected significance level of 0.05, as shown in the last two gray boxes in figure
2.2.
In summary, figure 2.2 outlines the processing procedures, where the first two blue boxes show
the common pre-processing steps. The three orange blocks and arrows denote the power spectral
analysis at the subject and group levels. The four green boxes represent the connectivity analysis
and lead to two separate outputs: the global and nodal connectivity measures. All analysis steps,
except statistical analysis, were applied to individual subject datasets, followed by statistical
analysis as the last step performed at the group level.
2.3 Results
The EEG signals from 45 subjects in three temporal segments (Stim1, Stim2, and post) were
analysed for both active and sham stimulation cases. The respective results are presented in the
following three sub-sections: First, baseline-normalized, sham-subtracted topographies were
obtained, illustrating significant differences in topographical EEG powers (
mPowerss) between
the two experimental conditions in the respective frequency bands. Second, three global graphical
metrics were derived from the GTA and between the active tPBM and sham treatments compared
22
at the group level. Comparisons were made for all three temporal segments in the beta band only
because tPBM significantly affected the chosen connectivity metrics of the beta rhythm. Finally,
five nodal graphical metrics were characterized and presented using topographic maps. Thus, we
revealed how the segregation, integration, and centrality of each cluster/region were significantly
altered by tPBM at the group level. Similar to the global metrics case, it was only in the beta band
that tPBM significantly altered the local graphical metrics.
2.3.1 Topographic changes in power between tPBM and sham stimulations
As shown in figure 2.2 and described in Sections 2.2.5 and 2.2.6, the baseline-normalized values
of
mPowerf (see eq. (2-1)) for each group of tPBM and sham conditions among the three temporal
segments (Stim1, Stim2, and post) and five frequency bands were calculated. Group-level
statistical comparison to the baseline for each stimulation case was performed using a permutation
test for each of the three temporal periods and five frequency bands, resulting in respective group-
23
level topographies (n=45). However, to show clear statistical differences in
mPowerf between
the two stimulation conditions, baseline-normalized and sham-subtracted topographical maps of
mPowerssf (%) values (see Eq. (2-2)) over 64 channels were achieved, as shown in figure 2.4 for
all five frequency bands. In addition, after the cluster-based permutation testing for 64-channel
statistical comparison, the electrode sites/clusters that were significantly affected by tPBM are
superimposed on the topographies in figure 2.4 with ‘*’ denoting p<0.01 and ‘×’ denoting p<0.05.
These results illustrate a significant, dose-dependent increase in EEG rhythm powers at alpha
(8-13 Hz) and beta (13-30 Hz) during the last 4 min of tPBM (i.e., Stim2). Specifically, the increase
in alpha
mPowerss was shown as two major clusters of channels in the bilateral frontal and left
parietal-occipital regions, whereas the increase in beta
mPowerss was mainly seen as one cluster
of electrodes in the central/parietal region of the scalp. The enhanced alpha
mPowerss remained
Figure 2.4 Topographic maps of group-averaged (n=45), baseline-normalized, and sham-subtracted
changes in mPowerss (see eq. (2-2)) in delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30
Hz), and gamma (30-70 Hz) bands during the first 4 minutes of tPBM (Stim1), second 4 minutes of
tPBM (Stim2), and post tPBM period. Also, statistical results after the cluster-based permutation
testing are superimposed in each topographical map, showing significant differences in mPower
between the tPBM and sham stimulations during respective three time segments and in five frequency
bands with corrected significance levels of p < 0.05 (×) and p < 0.01 (*).
24
in the affected locations during the post-tPBM period, whereas significant increase in beta
mPowerss ceased during the recovery time. Furthermore, delta power was reduced in the frontal,
left temporal, and occipital regions during tPBM, and in the right frontal region during recovery.
2.3.2 Global graphical metrics of functional connectivity altered by tPBM
Following the steps given in Sections 2.2.7-2.2.9, adjacency matrices for all three temporal
segments and five frequency bands were generated. These matrices were further binarized for
different sparsity values, resulting in the GTA-derived graphical networks. In this study, we
identified three global network metrics (S, GE, and SW) that were significantly altered by tPBM
with respect to the sham condition and only in the beta band. As shown in figure 2.5, the three
rows illustrate the respective global metrics during Stim1, Stim2, and the post-period under both
active and sham stimulation. The grey bars in each panel of figure 2.5 highlight the sparsity values
at which the corresponding graphical metrics were altered significantly by tPBM based on paired
t-tests (p < 0.05).
These results suggest that tPBM significantly reduces the global synchronization, global
efficiency, and small-worldness of the network connectivity of the human brain. Specifically,
significant decreases in synchronization and global efficiency is observed during Stim1 and the
recovery period with more sparsity values, while a significant reduction in small-worldness
appeared in Stim2 with more sparsity units. Additionally, we confirmed that there was no
significant difference between the pre-stimulation baselines under tPBM and sham stimulation for
any of the three global network metrics.
25
2.3.3 Nodal graphical metrics of functional connectivity altered by tPBM
After performing the analysis steps given in Sections 2.2.9, cluster-averaged, baseline-
subtracted values for each of the five nodal graphical metrics (i.e., nCC, nE, nLE, BC, and DC)
were obtained for each of the 10 spatial clusters (figure 2.3) under both tPBM and sham conditions.
Next, after performing paired t-tests with FDR correction for 10 spatial clusters (i.e., p < 0.05,
FDR corrected), we identified or categorized the clusters whose nodal metric values were
significantly altered (i.e., increased or decreased) by tPBM for each of the five metrics at all three
Figure 2.5 Three GTA-derived global graphical metrics, namely, synchronization (the 1st
column), global efficiency (the 2nd column), and small-worldness (the 3rd column), of the EEG
brain network in the beta band (13-30 Hz) under both active tPBM and sham stimulation during
Stim1 (the 1st row), Stim2 (the 2nd row), and the post period (the 3rd row). In each panel, the y
axis denotes respective metric values while the x axis presents sparsity values with an increment
of 5%. The grey bars mark sparsity values at which the corresponding graphical metrics were
altered significantly by tPBM with respect to sham based on paired t-tests (p < 0.05). Error bars
represent standard error of the mean.
26
temporal periods, and only in the beta rhythm band. The topographical representations of the
results for the five nodal metrics are shown in figure 2.6.
Based on figure 2.6, we made the following observations. (1) During Stim1, significant
increases in nodal efficiency, betweenness centrality, and degree centrality occurred in the right
frontal region near the tPBM stimulation site. (2) During Stim2 and post stimulation, significant
changes occurred in the bilateral frontal regions for all five nodal metrics. More specifically,
tPBM significantly decreased the clustering coefficient and nodal local efficiency while the
stimulation significantly increased the other three nodal metrics. (3) Combined temporal and
spatial results revealed that both nodal efficiency and degree centrality were initially enhanced by
tPBM in the right frontal region during Stim1, followed by expansion of this enhancement to the
contralateral side during Stim2, which remained during the post-tPBM period. (4) In the case of
betweenness centrality, unilateral enhancement in the right frontal region remained during the
entire stimulation time (i.e., Stim1 and Stim2) and then expanded to the contralateral side in the
post. (5) On the other hand, during Stim2 and post stimulation, significant decreases occurred in
the bilateral frontal regions for the nodal clustering coefficient (nCC) and nodal local efficiency
(nLE). (6) During the same time periods for the same two metrics (nCC and nLE), decreases
occurred in the left parietal and occipital regions. (7) Moreover, the left temporal region showed
a reduction in nodal efficiency and betweenness centrality only for Stim2. (8) The only significant
modulation in the right occipital region was a decrease in betweenness centrality in Stim2.
27
2.4 Discussion
In Section 2.3, we identified and showed the clusters and regions on the scalp where tPBM
modulated EEG oscillation powers and GTA-based EEG beta network connectivity. All these
observations provided a better topographical overview of tPBM-induced electrophysiological
effects on brain functional connectivity in the resting state. In this section, we will further interpret
and discuss our observations, compare our results with previous studies, and associate the
neurophysiological changes in different brain regions with behavioural improvement by tPBM that
have been reported by others [2-6, 46, 54, 84].
Figure 2.6 Comparative topographical maps of 10-Cluster-distributed nodal network metrics at the
beta band (13-30 Hz). The comparison was made between tPBM and sham stimulation conditions
for each of the five nodal metrics, namely, the clustering coefficient (the 1st column), nodal local
efficiency (the 2nd column), nodal efficiency (the 3rd column), betweenness centrality (the 4th
column), and degree centrality (the 5th column) of the EEG brain network during Stim1 (the 1st row),
Stim2 (the 2nd row), and the post period (the 3rd row). LF: Left frontal, RF: Right frontal, LC: Left
central, RC: Right central, LP: Left parietal, RP: Right parietal, LT: Left temporal, RT: Right
temporal, LO: Left occipital, RO: Right occipital. Red color represents tPBM > sham; blue color
indicates tPBM < sham with p < 0.05 (FDR corrected).
28
As shown in figure 2.4, under the eyes-closed resting-state condition, tPBM significantly increased
the power of alpha oscillations in clusters over the bilateral frontal, left parietal, and left occipital
regions, as well as the beta power over the bilateral central and parietal regions during the second
4-min of stimulation. These observations agree with previous reports on eyes-open tPBM
experiments [85, 86]. Moreover, tPBM significantly decreased the delta power during stimulation,
followed by a residue of reduced power over a smaller region during recovery.
2.4.1 tPBM-induced alterations on EEG
mPower in clusters of electrodes in
frontoparietal network
A significant increase in alpha power over frontal-parietal regions confirmed the ability of
tPBM to neurally modulate the frontoparietal network, which is an executive network facilitating
rapid instantiation of new tasks [87]. According to previous studies, alpha rhythm is thought to be
associated with awareness [88] and cognitive functions, such as memory encoding and attention
[89-91]. The same experimental protocol with a 1064-nm laser has been used previously,
demonstrating a significant improvement in cognitive performance in human participants [2-4, 46,
84]. Moreover, the presence of stronger beta waves has been linked to better cognitive abilities, as
reported in several studies [92, 93]. Thus, the beneficial outcome in human cognition by frontal
tPBM can be, at least partially, attributed to its significant modulation of electrophysiological alpha
and beta powers in the frontoparietal network. Regarding the reduction in delta power in the
channel clusters during the stimulation period, we interpreted this observation as a result of the
mild thermal sensation of the laser on the superficial tissue [27].
2.4.2 tPBM-induced alterations in global measures of functional network in beta band
As shown in figure 2.5, tPBM significantly changed three global graph measures, namely,
synchronization, global efficiency, and small-worldness, in the beta wave only. We discuss each
of these changes as follows:
29
Our observation that brain network synchronization in the beta band was significantly reduced
during Stim1 and recovery agrees with a recent study that reported the effects of tPBM on brain
network synchronization with 850 nm LEDs [30]. In addition, a behavioural study attributed a
decrease in synchronization in the healthy human brain to awareness and cognitive processing
[94]. Thus, the outcome and effect of tPBM on desynchronizing EEG beta waves may be
associated with cognitive processing.
Similarly, the global efficiency of the brain network was significantly reduced by tPBM
compared to sham during Stim1 and post. This reduction indicates a decrease in brain network
integration, which may imply less efficient or more complex information paths in the network. In
other words, from a global point of view, tPBM may increase the energy and wiring costs of the
information flow owing to the trade-off between network efficiency, energy, and wiring costs [95,
96]. This could be an indicator of increased brain complexity, which is related to higher cognitive
function [97]. This observation indirectly supports the expected benefit of tPBM, namely, the
beneficial effects of tPBM on cognitive improvement.
It is known that small-worldness is dependent on the global integration and segregation of the
network and is calculated as the ratio of normalized integration to normalized segregation [65].
Thus, the reduction in this metric could be attributed to a significant reduction in global integration
(as reflected by a reduction in GE) and/or a significant increase in the global segregation of the
brain network caused by tPBM. These observed significant effects of tPBM on small-worldness
taking place only in Stim2 could result from the resistance of resting-state networks against
changes in network composition, as well as the dose-dependent nature of tPBM-induced effects
on neural activity [98, 99]. However, a possible explanation for the lack of significant alteration
30
in synchronization and global efficiency in Stim2 could be the high variability in the functional
topography of the frontoparietal network [87].
In summary, significant decreases in synchronization, global efficiency, and small-worldness
can be associated with potential increases in the complexity of the brain network and the
improvement or enhancement of human cognitive function [94, 97]. In addition, the observation
that tPBM altered only beta-wave oscillations in the EEG graphical network was in agreement
with other studies [30, 64]. Several publications have shown the role of the beta band in different
brain networks [100] and cognitive functions [101, 102].
2.4.3 tPBM-induced alterations in nodal graphical measures in beta band
Figure 2.6 shows how tPBM would alter nodal connectivity of the functional brain network in the
beta band in healthy participants. Specifically, this figure illustrates that nodal clustering
coefficient and nodal local efficiency, as measures of segregation of the brain network [103], were
reduced significantly in bilateral frontal regions in Stim2 and post. Furthermore, these two
segregation nodal metrics were decreased in the left parietal and left occipital regions during Stim2
and post, respectively. All these observations implied that the clusters of nodes in these regions of
the network became less segregated during the 2nd 4-min and post period of tPBM. In other words,
tPBM facilitated less separation and more integration of nodal graphical connectivity.
Consistent with the aforementioned observation, nodal efficiency, which reflects nodal
integration, increased in the right frontal region during the first 4 min of tPBM. This increase
indicates enhanced integrity of the nodes over this region in the information flow [103] and parallel
information transfer. Furthermore, during the last 4 min of tPBM, the bilateral frontal regions
showed an increase in integration into the functional network, while the integrity of the left
temporal region in the network was significantly reduced. This phenomenon implies that the tPBM
31
stimulated more network integration at the beta rhythm in the frontal regions, with the cost of
reducing network integration in the left temporal segment/cluster.
Betweenness centrality represents the fraction of all shortest paths in the network that pass
through a particular central node [104]. A large value of betweenness centrality denotes a large
impact of this central node on the information flow over the entire network. It is common for nodes
at intersections of disparate parts of the network to have a high betweenness centrality [65]. It is
clear from figure 2.6 that betweenness centrality increased in the right frontal region during Stim1
and Stim2. Similar changes in the frontal and temporal regions of beta waves have been reported
during cognitive training [101]. After stimulation, the effects of tPBM on betweenness centrality
remained bilateral in the frontal regions. These observations suggest that the frontal regions,
especially the right frontal region under tPBM, became more prominent in connecting the disparate
parts of the network throughout the stimulation and recovery periods.
The last nodal measure investigated was degree centrality, which quantifies the number of links
from nodes in a specific region to other nodes in the same region or other regions [65]. An increase
in this nodal metric was observed in the right frontal region during all three temporal segments
and in the left frontal region during Stim2 and Post. The enhancement of this nodal parameter
revealed that these cortical/brain regions could be prominently stimulated by tPBM for more
information connections or links at beta wave oscillations. It has also been reported that a working
memory task during the encoding phase triggers similar increases in degree centrality over the
frontal regions of the beta band [102].
Combining all these observations, we concluded that tPBM facilitated a reduction in local
segregation, increases in nodal integration and centrality of frontal regions, and the growth of
connection links between nodes in these frontal regions compared to other regions. These results
32
are in agreement with the observed changes in the flow of information reported in a tPBM-evoked
causal connectivity study [105].
2.4.4 The role of the beta band in tPBM-induced network modulation and its relation to
enhancement of human cognition
The alpha and beta powers of the human brain, especially in the frontoparietal network, are
believed to be related to cognitive functions, such as memory encoding and attention [89-91],
especially in the frontal, temporal, and parietal regions. Our observations (as presented in Section
2.3.1) clearly demonstrated that tPBM enabled increases in the alpha and beta powers in the
frontal-central-parietal regions, indicating the underlying association between tPBM and
enhancement of human memory.
However, our results in Sections 2.3.2 and 2.3.3 showed that tPBM altered EEG graphical
network metrics only in the beta band, which was consistent with the results given in ref. [30].
Thus, an implication of the relationship between prefrontal tPBM and its effects on EEG beta-
wave network metrics was sought. Regarding the beta rhythm in the prefrontal cortex (PFC),
“increased beta appears at the end of a trial when working memory information needs to be erased.
A similar ‘clear-out’ function might apply during the stopping of action and the stopping of long-
term memory retrieval (stopping thoughts), where increased prefrontal beta is also observed. A
different apparent role for beta in the PFC occurs during the delay period of working memory
tasks: it might serve to maintain the current content and/or to prevent interference from distraction.
[100]” Accordingly, beta oscillations in the prefrontal region appear to serve as short-term memory
executors and focus enhancers during executable tasks. In addition, the beta rhythm in the temporal
lobe plays an important role in long-term memory retrieval [106]. Memory retrieval starts in the
temporal lobe, passes through different parts of the neocortex, and stops in the prefrontal cortex
[100]. There is always a balance in information flow in this order, even during the resting state.
33
Accordingly, we speculated that tPBM enables to significant neuromodulation of beta oscillations
and the corresponding network connectivity globally across the scalp and regionally in several
nodal/cluster regions. The significant modulation by tPBM on beta-wave connectivity in the
human brain may be an underlying electrophysiological mechanism and association between
tPBM and the enhancement of human cognition.
2.4.5 Comparisons to two other publications
As shown in Table 2.1, only two recent papers have reported tPBM-induced modulations of global
network metrics [29] and alterations in brain connectivity between the two hemispheres [30]. It
would be helpful to compare the altered graphical metrics provided in these two articles with those
found in this study. Tables 2.3 and 2.4 below summarize and compare the global and local
graphical metrics of the three studies, respectively.
Upon inspection of the global metrics of the network, Table 2.3 reveals that only global
efficiency was altered by tPBM in all three cases, regardless of stimulation conditions. However,
a more consistent agreement exists between Ref. [30] and this study with network alterations in
the beta band. On the other hand, Ref. [29] showed increases in GE in the alpha and gamma bands
under 40-Hz tPBM, which should result in alterations of the gamma waves (30-70 Hz).
34
Table 2.4 lists several nodal network metrics for comparison with Ref. [30] and this study,
because Ref. [29] did not provide results on tPBM-altered nodal metrics. Both studies observed
consistent tPBM-induced alterations in beta wave connectivity between the two hemispheres [30]
or in the prefrontal regions (in this study), whereas the modulated network metrics were different.
Table 2.
4 Comparisons of nodal graphical metrics among three studies
tPBM
duration
Radius of
light size
nCC nE nLE
Eigenvector
Centrality
BC DC
Ref. [30]
(850-nm LED,
CW; PFC)
2.5 min
0.67 cm - - -
Yes; (beta;
between two
hemispheres)
- -
This Study
(1064
-
nm Laser,
CW; PFC)
8 min
2.1 cm
Yes;
(beta;
frontal)
Yes;
(beta;
frontal)
Yes;
(beta;
frontal)
-
Yes;
(beta;
frontal)
Yes;
(beta;
frontal)
Such differences could be accounted for by several experimental setting parameters, including
wavelengths (850 nm vs. 1064 nm), light type (LEDs vs. laser), stimulation size on the forehead
(0.67 cm vs. 2.1 cm), stimulation duration (2.5 min versus 8 min) of the tPBM. We observed that
the right forehead tPBM using a 1064-nm laser with larger stimulation size and longer duration
Table 2.3 Comparisons of global graphical metrics among three studies
CC
Characteristi
c path length
GE LE Energy Entropy S SW
Ref. [29]
(810-nm
LED, 40 Hz;
DMN)
increase
(alpha &
gamma)
increase
(alpha &
gamma)
increase
(alpha &
gamma)
increase
(alpha &
gamma) - - - -
Ref. [30]
(850-nm
LED, CW;
PFC)
Decrease
(beta) - Decrease
(beta) Decrease
(beta)
Decrease
(beta) Increase
(beta) - -
This Study
(1064-nm
Laser, CW;
PFC)
- - Decrease
(beta) - - -
Decrease
(beta)
Decrease
(beta)
35
created significant stimulations and alterations in nodal brain connectivity metrics, particularly in
the prefrontal regions, near the stimulation site.
2.4.6 Limitations and future work
First, the international 1010 electrode placement system in this study was not strictly followed
on the human head because a clear area with 4.2 cm in diameter was needed for tPBM light
delivery on the right forehead. The EEG cap was shifted 1–2 cm backward. There was a systematic
shift in the electrode locations given in Figs. 2.4 and 2.6 with respect to the standard 64-electrode
locations. However, the precision of EEG channel locations is not affecting the final results
significantly, since we focus on the global and regional/clustered effects of tPBM. This position
precision is most important when the collected scalp EEG data is being back-projected to the brain
to evaluate power and connectivity modulation in specific regions of brain. Second, the power
spectral and connectivity analyses were performed in the sensor space. Source space analysis can
be conducted to observe specific cortical and subcortical regions in the brain affected by tPBM.
Third, the current study was based on EEG signals of the tPBM-treated human brain in the resting
state without the evaluation of any behavioural assessment. It is highly desirable to obtain
concurrent assessments of changes in brain connectivity metrics and cognitive enhancement after
tPBM. A combination of electrophysiological and behavioural measures would provide more
informative and comprehensive views of the correlation and association between functional
connectivity and behavioural effects of tPBM. Overall, there are few publications in the literature
on how tPBM affects brain connectivity and the association between tPBM-induced network
changes and cognitive improvement. It is necessary to promote and conduct more investigations
in this line of work to make tPBM a non-invasive, portable, and low-cost intervention tool for
36
healing patients with certain brain disorders as well as for healthy aging in the rapidly growing
aging population.
2.5 Conclusion
In this study, we utilized three analysis steps to identify the electrophysiological effects of tPBM
in a healthy human brain. First, power spectral analysis revealed that alterations in EEG spectral
powers were mainly present in the alpha and beta bands of the fronto-central-parietal regions.
Second, a topological approach, GTA, facilitated findings on significant modulation of the EEG
beta rhythm in the information path and enhancement of the brain network complexity at the global
network level during and after the stimulation. Finally, assessment of the nodal measures of the
network at the regional and cluster levels confirmed that tPBM had a major effect on the frontal
and parietal clusters in the beta band. The information paths were enhanced during and post tPBM
in the prefrontal regions near the stimulation site. Further studies are needed to better understand
the relationship between tPBM-induced alteration of brain networks and improvement in human
cognition if tPBM is to be developed as a useful tool for treating patients with brain disorders and
for supporting healthy aging in the aging population worldwide.
37
Chapter 3
Prefrontal Cortical Connectivity and Coupling of Infraslow
Oscillation in the Resting Human Brain:
A Two-Channel Broadband NIRS Study
Sadra Shahdadian, Xinlong Wang, Shu Kang, Caroline Carter, Akhil Chaudhari, Hanli Liu
(This chapter is a manuscript that has been accepted for publication in Cerebral Cortex
Communications)
3.1 Introduction
3.1.1 Infra-slow Oscillation of the Human Brain
The human brain plays a major role in oxygen and glucose consumption despite its relatively low
weight compared to other organs [107, 108]. The high levels of consumption are due to the high
metabolic activity of neurons, which is modulated by the oxygenated blood supply and cerebral
metabolism [109, 110]. Many studies have focused on investigating the mechanism of cerebral
metabolic activity and have found vasomotion to be a major source of metabolic and hemodynamic
modulations [31-35]. Vasomotion is a spontaneous oscillation that originates from the blood vessel
wall with an infra-slow oscillation (ISO) of 0.005-0.2 Hz [36, 37]. In addition, a correlation is
found between the ISO of cerebral metabolic activities and human cognitive functions [38].
Furthermore, vasomotion malfunction has been observed in older adults and in patients with
different diseases, such as atherosclerosis [39], cardiovascular disease [40], and Alzheimer’s
disease [41]. Thus, it may be beneficial to quantify and characterize cerebral metabolism in the
ISO range, which may provide better insight into neurophysiological mechanisms and discover
features that differ between healthy humans and patients with brain disorders.
38
Relaxation-contraction cycles of blood vessel walls have been shown to be the driving force
for the infra-slow rhythms of cerebral hemodynamic oscillations, independent of respiration and
heartbeat [32, 111-113]. Three intrinsic frequency components of infra-slow cerebral
hemodynamic rhythms have been found to correspond to the specific physiological and
biochemical activities of the vascular wall layers [114]. These frequency bands consist of (1)
endogenic (0.005-0.02 Hz), neurogenic (0.02-0.04 Hz), and (3) myogenic (0.04-0.2 Hz) [115-117]
rhythms. The endogenic band corresponds to dilation-contraction cycles in the endothelial layer
affected by the release of potent vasoactive factors, such as nitric oxide (NO), free radicals,
prostacyclin, endothelium-derived hyperpolarizing factor, and endothelin [118, 119]. Oscillation
in releasing vasoactive ions and neurotransmitters from neurons leads to modulation of the vessel
dilation-contraction cycles in the neurogenic band [120]. Rhythmic myogenic activity, on the other
hand, occurs as a result of the relaxation and contraction of vascular wall smooth muscle cells
[117]. Such hemodynamic ISO can be detected by different measurement modalities, such as
functional magnetic resonance imaging (fMRI) [121], transcranial cerebral doppler (TCD) [122],
and functional near-infrared spectroscopy (fNIRS) [38]. However, these methods are not capable
of concurrently monitoring the metabolic rhythms originating in the mitochondria. As
mitochondria play a major role in cerebral metabolism and vasomotion, detecting mitochondrial
activity and ISO is essential important [123].
3.1.2 Exploration of the Prefrontal Cortical Connectivity and Coupling of ISO
The bilateral prefrontal connectivity of the human brain with respect to certain neurophysiological
functions reflects the level at which the lateral sides of the prefrontal cortex oscillate synchronously
or coherently. Therefore, a higher level of connectivity represents a bilaterally or globally driven
oscillation while a lower level of connectivity denotes locally driven activity [124]. On the other
39
hand, unilateral hemodynamic-metabolic coupling indicates how the supply demand relationship
between local oxygenated hemodynamics and metabolism is regulated. Any impaired, abnormal,
or diminished bilateral connectivity and/or unilateral/local coupling of the prefrontal ISO could
reflect or result from neurological diseases or brain disorders. This is because prefrontal cortex
activity is closely associated with human cognition; many studies have provided evidence of
correlations between prefrontal cortex activity and human cognition [125-129]. Thus, it is
desirable to quantify prefrontal cortical connectivity and coupling of ISO in the human brain,
which may be closely associated with normal or abnormal brain states, and may be developed for
clinical applications in the near future.
3.1.3 Broadband Near-infrared Spectroscopy and Resting-State Analyses
Broadband Near-infrared spectroscopy (bbNIRS) has been investigated for more than 2 decades
[130-134] and accepted as a reliable tool to quantify changes of oxygenated and deoxygenated
hemoglobin concentrations ([HbO] and [HHb], respectively) as well as redox-state cytochrome-c-
oxidase concentration ([CCO]) based on absorption and scattering of NIR light by these
chromophores [131, 135, 136]. In particular, cytochrome-c-oxidase is the terminal enzyme in the
mitochondrial respiratory chain that catalyzes the reduction of oxygen for energy metabolism
[137-140]. Because redox CCO has a light absorption peak at ~800 nm, bbNIRS can quantify
changes in [CCO] ([CCO]) and enable us to reveal the metabolic state of a tissue [131, 135,
136]. However, since the concentration of CCO is much smaller than those of HbO and HHb in
living tissues, accurate estimation of changes in [CCO] requires a multispectral approach [135,
136, 141]. In the past several years, our group has successfully quantified altered redox [CCO] in
response to photobiomodulation using 1-channel or 2-channel bbNIRS (2-bbNIRS) taken on the
human forearm or forehead [7, 9, 12, 142].
40
However, most studies in the field of either fNIRS or bb-NIRS are based on time-domain
analyses and are often performed under task-evoked brain states [143-145]. Numerous articles on
fNIRS-derived resting-state connectivity have been based only on hemodynamic (HbO)
oscillations [146-148]. Little or no report could be found on the frequency-domain analysis of bb-
NIRS measurements in the resting human brain. It is also unknown whether 2-bbNIRS can
facilitate characterization of prefrontal connectivity and coupling in the brain.
In the present exploratory study [42], we hypothesized that 2-bbNIRS, along with frequency-
domain analysis, enables us to quantify prefrontal cortical connectivity and coupling of ISO in the
resting human brain. Specifically, the features analyzed from the 2-bbNIRS time series included
(1) resting-state spectral amplitude (SA) of bilateral cortical hemodynamic and metabolic (i.e.,
SAHbO_i and SACCO_i) activities, where i represents either the left or right prefrontal region, (2)
bilateral hemodynamic connectivity (bCONHbO), (3) bilateral metabolic connectivity (bCONCCO),
and (4) coupling between cerebral hemodynamic and metabolic activities on the unilateral side
(uCOPHbO-CCO_i) of the prefrontal cortex over the three ISO frequency bands. By the end of this
exploratory study, we would support this hypothesis by presenting relatively stable and consistent
values for these features in healthy young humans, revealing the translation potential of these
features for future clinical applications.
3.2 Materials and Methods
3.2.1 Participants
31 healthy human subjects were recruited from the local community at the University of Texas,
Arlington. They were screened using the same inclusion criteria as those used in the previous
studies [7, 9]. In summary, the inclusion criteria included: either sex, any ethnic background and
in an age range of 18–40 years old. The exclusion criteria included: (1) diagnosed with a
41
psychiatric disorder, (2) history of a neurological condition, or severe brain injury, or violent
behavior, (3) have ever been institutionalized/imprisoned, (4) current intake of any medicine or
drug, or (5) currently pregnant. Each participant had five visits, separated by at least 7 days.
Because of the high sensitivity of bbNIRS to motion artifacts, five subjects with excessive motion
during one or more of the five experiments were excluded from the data. After exclusion, a total
of 26 young and healthy humans (14 males and 12 females, mean ± SD age = 22.4 ± 2.3 years)
participated in the 5-visit experiments. The study protocol complied with all applicable federal
guidelines and was approved by the Institutional Review Board (IRB) of the University of Texas
at Arlington. Informed consent was obtained from all participants.
3.2.2 Experiment Setup and Protocol
The data analyzed in this study were obtained from single-mode, resting-state, bilateral
measurements with 2-channel bbNIRS, which is one of the dual-mode (i.e., bbNIRS and EEG)
modalities. Specifically, a 2-channel bbNIRS probe (Figure 1(a)) was placed bilaterally on the
forehead of the participants to acquire prefrontal ISO signals of Δ[HbO] and Δ[CCO] at rest. The
2-channel system consisted of two branches of a broadband white light source (Model 3900e,
Illumination Technologies, NY, USA) and two CCD array spectrometers (QEPRO, Ocean Optics
Inc., Orlando, FL, USA) as light detectors (Figure 3.1(b)). The two bbNIRS recording channels
were positioned symmetrically on the subject’s forehead (visual judgement). Each channel
consisted of one fiber bundle for light delivery to the forehead and another for backscattered light
collection from the brain tissue, with a source-detector separation of 3 cm. A 2-channel probe
holder was designed and 3D printed with a flexible material to ensure comfortable and firm
attachment of the fiber bundles to the forehead skin, accommodating each participant’s forehead
curvature. The probe holder was fastened to each participant's forehead with hook-and-loop
42
fasteners, and adhesive medical tape was applied to the probe-skin interface to hold the probe on
the forehead more steadily (without tightening the fastener too much), thus reducing motion
artifacts.
Regarding the measurement protocol, after the consent form was signed, each participant was
instructed to sit comfortably on a chair, followed by a dual-mode probe placement on the
participant’s head firmly. Then the 2-channel bbNIRS (and EEG) started to record data at a rate of
1.5 sec per temporal point (i.e., 0.67 Hz) during the 7-min resting state while the participant kept
their eyes closed without falling asleep.
Figure 3. 1 (a) Dual-mode (bbNIRS and EEG) head probe setup, showing two separate channels
with two sets of fiber bundles that were connected to (b) the 2-channel bbNIRS. While an EEG
cap on the head is observable, the EEG data are not the topic/subject of this paper. The bbNIRS
datasets used for this study were taken during 7-min eyes-closed conditions with the setup shown
above.
3.2.2 Data Analysis
After 2-bbNIRS data acquisition, the data processing steps included both time- and frequency-
domain analyses, as outlined in Figure 3.2, in five steps. Step 1 (blue boxes in Figure 3.2) was to
obtain the [HbO] and [CCO] time series after converting the raw data to [HbO] and [CCO]
at each time point. Step 2 (the yellow box in Figure 3.2) involved performing frequency-domain
analysis using the multi-taper method that facilitated the following two steps to investigate the
cerebral hemodynamic and metabolic ISO of the human prefrontal cortex in the resting state. Step
(a)
(b)
43
3 (the orange box) was to quantify spectral amplitudes of Δ[HbO] and Δ[CCO] (i.e., SAHbO_i and
SACCO_i) in the endogenic, neurogenic, and myogenic (E/N/M) frequency bands measured on each
lateral prefrontal site, where the subscript of “i” labels either “L” or “R” for the left or right
forehead. Step 4 (green box) was used to perform coherence analysis and to determine (i) bilateral
connectivity for Δ[HbO] and Δ[CCO] (i.e., bCONHbO and bCONCCO) of the human forehead and
(ii) unilateral cerebral hemodynamic-metabolic coupling (uCOPHbO-CCO_i) for each lateral
prefrontal cortex. Steps 1 4 were repeated for each of the 26 participants and then for five sets of
measurements. Step 5: One-way ANOVA was performed to demonstrate no significant difference
among the five measurements for each of the bilateral SA, bilateral connectivity, or unilateral
coupling parameters (i.e., SAHbO_i, SACCO_i, bCONHbO, bCONCCO, uCOPHbO-CCO_i) in each of the
E/N/M bands.
Figure 3. 2 A data processing flow chart with five steps. Step 1: [HbO] and [CCO] quantification at
each time point and time series (blue boxes); Step 2: amplitude and phase decomposition using multi-taper
method (yellow box); Step 3: quantification of spectral amplitudes (SA) for endogenic, neurogenic, and
myogenic (E/N/M) frequency bands (orange box); Step 4: determination of four types of coherences for
each E/N/M bands (green box). Steps 1 to 4 were repeated for each of 26 participants (outlined by the
dotted box) and then for 5 sets of the measurements (outlined by the solid box). The bottom dashed box
44
marks Step 5, showing several statistical analyses, including one-way ANOVA, paired t-tests, and two one-
sided tests (TOST) used to identify group-level features for SA and respective coherence parameters (gray
box).
3.2.2.1 Step 1: Quantification of [HbO] and [CCO] time series
As mentioned in the Introduction, bbNIRS has been studied for more than two decades [130-134]
and is well accepted as a reliable tool for quantifying cortical [HbO], [HHb], and [CCO] based on
the modified Beer-Lambert law (MBL) [131, 135, 136]. Following the same approach, we selected
the spectral range of 780-900 nm from the recorded optical spectrum at each time point, and
quantified prefrontal Δ[HbO] and Δ[CCO] based on MBL and multiple linear regression analysis
with a low-pass filter at 0.2 Hz [143]. Detailed derivations and steps can be found in Refs. [7, 142].
After repeating the concentration quantification at all recorded time points, we obtained a time
series of Δ[HbO] and Δ[CCO] for the 7-min resting-state period at a sampling frequency of 0.67
Hz. The spectral range of 780-900 nm estimates the chromophore concentration with a low level
of error propagated from noise [149].
3.2.2.2 Step 2: Multi-taper method for spectral analysis of [HbO] and [CCO]
The multi-taper method (MTM) [79, 80] is a well-known time-frequency analysis for a time series.
Specifically, multiple tapers, mainly Slepian sequences, are used to taper the recorded signal in the
time domain before performing the Fourier transform to provide a frequency-domain spectrum
[79, 80]. This method maintains a reasonably high spectral resolution while reducing spectral
noise. In this study, the MTM enabled us to decompose the amplitude and phase of the Δ[HbO]
and Δ[CCO] time series obtained from both bbNIRS channels. Specifically, we utilized several
functions (including “ft_freqanalysis” and “ft_connectivityanalysis”) available in the FieldTrip
toolbox [76, 77] to perform MTM operations. Section B.1 in the Appendix explains the two
45
functions of “ft_freqanalysis” and “ft_connectivityanalysis” and presents a detailed flow chart
(Figure B1) to illustrate the calculations for SA.
3.2.2.3 Step 3: Quantification of SA in E/N/M Bands
One of the outputs of MTM is the power spectral density (PSD) smoothed over a given frequency
range. In this study, smoothed PSDs of Δ[HbO] and Δ[CCO] over a 7-min resting period were
obtained across E/N/M frequency bands. For a 7-min measurement duration, the spectral (or
frequency) resolution was 1/(7 min)= 1/(7 × 60 sec)= 0.0024 Hz. Accordingly, the signal spectral
power at each PSD frequency was obtained by multiplying the PSD value by the spectral resolution
at the respective frequency. Next, by taking the square root of the spectral power, we were able to
attain a spectrum of ISO amplitude versus frequency between 0.005 and 0.2 Hz (as selected in
Step 1). Finally, we obtained the mean spectral amplitudes (SA) over each ISO band for Δ[HbO]
and Δ[CCO]. The methodological steps are expressed as follows:
amplitude(f) =() = ()×
(3-1)
SAHbO_i = mean amplitude(f) of [HbO] over the ith band, (3-2)
SACCO_i = mean amplitude(f) of [CCO] over the ith band, (3-3)
where PSD(f), power(f), and amplitude(f) are the frequency-dependent spectra of PSD, power, and
amplitude, respectively; i represents the ith band for E/N/M frequencies (i.e., i = E, N, M) on each
side of the participant’s forehead.
3.2.2.4 Step 4: Hemodynamic and Metabolic Connectivity and Coupling by Coherence
In theory, brain connectivity measures rely on the amplitude and/or phase of the signal recorded
from each channel to quantify the level at which each pair of signals interact with each other. Based
on the mathematical definition of the connectivity measure, the correlation between the phases
and/or amplitudes of two time series (recorded by two respective channels) can be interpreted as
46
the functional connectivity/coupling of these time series [13, 14]. In contrast, the counterpart of
the time-domain cross-correlation calculation is coherence in the frequency domain, which can be
used to facilitate or quantify the cerebral connectivity in this study. The coherence coefficient is a
normalized number between 0 and 1 without any unit, and is expressed as a function of frequency,
ω, as follows [13]:
( ) ()
( ) ( )
xy
xy xx yy
S
coh SS
ω
ωωω
=
, (3-4)
where Sxx and Syy indicate the power estimates of the signals x and y, respectively, and Sxy represents
the averaged cross-spectral density term of the two signals. These terms can be calculated using
the complex values obtained from the MTM method [80, 150].
In the next step of spectral analysis, we quantified four pairs of spectral coherence for the
resting-state human forehead: (1) bilateral coherence of Δ[HbO] to represent bilateral
hemodynamic connectivity (bCONHbO), (2) bilateral coherence of Δ[CCO] to represent bilateral
metabolic connectivity of (bCONCCO), (3) unilateral coherence between Δ[HbO] and Δ[CCO] on
the left, and (4) right side of the forehead to designate hemodynamic-metabolic coupling on the
respective prefrontal cortex (i.e., uCOPHbO-CCO_L and uCOPHbO-CCO_R). In practice, the function of
“ft_connectivityanalysis” available in the FieldTrip toolbox [76, 77] was used to facilitate these
coherence spectra, followed by band averaging within each of the three (E/N/M) frequency bands
of the ISO. The flow chart (Figure B1) in the Supplementary Material offers graphical steps for
calculating coherence.
Step 5: Statistical Analyses for ISO features
47
The aforementioned steps were repeated for each of the two bbNIRS channels for all subjects
during each of the five visits. Three stages of statistical analyses were performed for SAHbO (or
SACCO):
(1) ANOVA was performed to prove that there was no significant difference in SAHbO (or
SACCO) among the five measurements. This set of ANOVA tests was performed for each
of the Δ[HbO] (or Δ[CCO]) metrics on the bilateral channels.
(2) A set of paired t-tests was performed to compare the bilateral values of grand-averaged
SAHbO (or SACCO) over five repeated measurements from all 26 participants at all three
E/N/M bands.
(3) The two one-sided tests (TOST) analysis was utilized to evaluate the equivalence of the
features that did not show a significant difference between bilateral values for SAHbO (or
SACCO). Details of the equivalence test using TOST can be found in Section B.2 of the
Appendix.
After these stages of analyses, all bilaterally equivalent values of SAHbO (or SACCO) at each
E/N/M band were reported as features for the prefrontal hemodynamic (or metabolic) spectral
amplitudes.
Similar statistical analyses of the three stages were performed on four coherence metrics that
were band-averaged over all the subjects for each of the five visits. In ANOVA tests, after
obtaining the bilateral connectivity indices for bCONHbO (or bCONCCO) for all three E/N/M bands,
a one-way ANOVA was performed to assess the similarity among the three bands, followed by
Tukey’s post hoc test or TOST to detect statistically different or equivalent bCONHbO (or
bCONCCO) indices, respectively, across the three bands. Finally, grand-averaged uCOPHbO-CCO
indices on the two lateral (left and right) sides were compared using a set of paired t-tests for all
48
three E/N/M bands. In case of no significant difference between the left and right uCOPHbO-CCO in
any of the three bands, TOST was performed to test and prove the equivalence of the bilateral
values of uCOPHbO-CCO. Then, the averaged value of uCOPHbO-CCO was reported as a feature for
prefrontal, resting-state hemodynamic-metabolic coupling.
3.3 Results
The hypothesis of this study was that bilateral hemodynamic and metabolic connectivity and
unilateral coupling of the ISO in the resting human forehead can be quantified using 2-bbNIRS
and may serve as measurable features reflecting the prefrontal brain state. To prove or support this,
we took 7-min, resting-state, 2-bbNIRS measurements from the forehead of 26 young and healthy
participants (after exclusion of five subjects). The analyzed results focused on (1) SAs, (2) bilateral
coherence, and (3) unilateral coherence among four time series of [HbO] and [CCO] signals
obtained from the prefrontal cortex.
3.3.1 Time Series of [HbO] and [CCO] versus Their Spectral Analysis
After fitting the MBL with the spectral data of 2-bbNIRS (Step 1), we obtained a 7-min time series
of [HbO] and [CCO] from each lateral side of the forehead of each participant. As an example,
Figures 3.3(a) and 3.3(b) show time profiles of [HbO] and [CCO] derived from one channel of
one subject’s dataset; their time series fluctuated around 0 between ±0.3 µM and ±0.04 µM,
respectively. After performing spectral analysis (Step 2) and quantification of SA (Step 3), we
obtained SA values for both [HbO] and [CCO], as shown in Figures 3.3(c) and (d), respectively,
where the three frequency bands (E/N/M) are color-shaded. In addition, Section B.3 in the
Appendix shows an example of the [HbO] time series from one channel of 2-bbNIRS of a
49
subject’s dataset. This figure illustrates how different ISO waveforms in the three E/N/M bands
contribute to the composition of the wideband (0.005–0.2 Hz) original signal.
Figure 3. 3 (a) and (b) illustrate an example of time-domain representation of [HbO] and [CCO] signals,
respectively, with a frequency band of 0.005-0.2 Hz over a period of 7 min. This set of time series was
derived after processing Step 1 from one channel of the subject’s dataset. (c) and (d) show the frequency-
domain spectral amplitudes for [HbO] and [CCO], respectively, quantified using Steps 2 and 3. Blue,
green, and red indicate endogenic, neurogenic, and myogenic bands, respectively.
3.3.2 ISO Spectral Amplitudes of Prefrontal Δ[HbO] and Δ[CCO] in the Resting Brain
Figures 3.4(a) and 4(c) show the SAHbO and SACCO values in the E band, which are dominant over
those in the other two bands. Furthermore, the paired t-test results demonstrated no significant
difference in SAHbO between the two prefrontal regions across all three E/N/M bands. These
observations are in good agreement with those of a recent and independent study by our group
[43], which utilized a completely different bbNIRS system and setup from a different cohort of
participants. In addition, the data processing algorithms used to obtain the SAHbO differed between
the two studies. Similarly, SACCO values from both prefrontal cortices were statistically equivalent
50
in the M band. However, the SACCO values in the left prefrontal cortex were significantly higher
than those in the right prefrontal cortex in both the E- and N-bands.
Figure 3. 4 Resting-state prefrontal SAHbO (in µM) of the left and right forehead averaged over (a) a
combined set of grand/total measurements (n=130) and (b) each individual set of five measurements (n=26
per set) at endogenic (E; 0.005-0.02 Hz), neurogenic (N; 0.02-0.04 Hz), and myogenic (M; 0.04-0.2 Hz)
frequency bands. Similarly, resting-state prefrontal SACCO (in µM) of the left and right forehead averaged
over (c) the combined set of measurements (n=130) and (d) each individual set of five measurements (n=26
per set) at E/N/M bands. p-values shown for each group of bars in (b) and (d) represent ANOVA results.
All error bars are based on the standard error of the mean. *: p<0.05. I values represent intraclass
correlation coefficients for each group.
To evaluate the consistency and stability of the 2-bbNIRS measurements, five sets of
derived/quantified SAHbO and SACCO values were determined and plotted in Figures 3.4(b) and
3.4(d), respectively. One-way ANOVA was performed to assess significant differences among the
five measurements. The analysis outcomes showed no statistically significant differences among
the five datasets for each of the three frequency bands. In addition, intraclass correlation coefficient
51
(ICC) was calculated and reported in this figure for the 5 repeated measurements to evaluate the
reproducibility of the feature in the subject level.
Specifically, the second and third columns from the left of Table 3.1 represent the grand
averages of SAHbO values (as shown in Figure 3.4) over all experiments (n=130) taken from the
left and right prefrontal cortices of the 26 participants across the three ISO frequency bands. The
fourth and fifth columns list the p-values and pTOST obtained from the paired t-tests and TOST
analysis, respectively, between the left and right SAHbO values. This table indicates that left and
right SAHbO values were statistically equivalent in each E/N/M band; thus, the bilateral average
was calculated and is presented in the last column from the left. In addition, Table 3.2 shows the
results of SACCO in the three E/N/M bands using the data presentation similar to Table 3.1. It is
clear that the myogenic band was the only one with equivalent left and right prefrontal SACCO.
Overall, four bilaterally equivalent SA values were found as ISO features to characterize prefrontal
ISO of the resting human brain.
Table 3. 1 Grand averages of SAHbO over all measurements (n=130) on the left and right forehead across
three ISO frequency bands
Frequency
band SAHbO, left
(mean ±
s.d.)
SAHbO, right
(mean
±
s.d.)
Left vs right
t
-test (p-value)
Left vs right
TOST (pTOST) Bilateral average
SAHbO (mean
±
s.d.)
Endogenic 0.16 ± 0.08 0.15 ± 0.07 0.52
< 0.01;
bilaterally equivalent 0.16
±
0.07
Neurogenic
0.09 ± 0.04
0.09 ± 0.04
0.76
< 0.01;
bilaterally equivalent
0.09 ± 0.04
Myogenic
0.05 ± 0.02
0.05 ± 0.02
0.99
< 0.01;
bilaterally equivalent
0.05 ± 0.02
Note: p-values and pTOST were obtained using paired t-tests and TOST, respectively. See Section
B.2 in the Supplementary Material for details on TOST.
52
Table 3. 2 Grand averages of SACCO over all measurements (n=130) on the left and right forehead
across three ISO bands
Frequency
band SACCO, left
(mean
±
s.d.) SACCO, right
(mean
±
s.d.) Left vs right
t-test (p-value)
Left vs right
TOST (pTOST) Bilateral average
SACCO (mean
±
s.d.)
Endogenic 0.013±0.005 0.011±0.005 < 0.02 0.11 Left > right
Neurogenic 0.010±0.004 0.009±0.003 < 0.04 0.08 Left > right
Myogenic 0.007±0.002 0.007±0.002 0.72
< 0.001;
bilaterally equivalent
0.007
±
0.002
Note: p-values and pTOST were obtained using paired t-tests and TOST, respectively. See Section
B.2 in the Supplementary Material for details on TOST.
3.3.3 ISO Coherence of Prefrontal Δ[HbO] and Δ[CCO] in the Resting Human Brain
Figure 3.5(a) shows the comparisons between bilateral cerebral hemodynamic connectivity and
bilateral metabolic connectivity over the three ISO bands. Paired t-tests confirmed that bCONHbO
was significantly stronger than bCONCCO in all the E/N/M bands. To evaluate the significant
differences in these values, both bCONHbO and bCONCCO for bilateral connectivity were calculated
for each set of five measurements and are plotted separately in Figure 3.5(b). After performing a
one-way ANOVA on these five datasets, we confirmed that no statistically significant difference
in bilateral connectivity existed among the five datasets at all three frequency bands for both
Δ[HbO] and Δ[CCO], as evidenced by the p-values in Figure 3.5(b). In addition, intraclass
correlation coefficient (ICC) was calculated and reported in this figure for the 5 repeated
measurements to evaluate the reproducibility of the feature in the subject level.
53
Figure 3. 5 Resting-state prefrontal bCONHbO and bCONCCO averaged over (a) the combined set of
measurements (n=130) and (b) each separate set of five measurements (n=26 per set) over E (0.005-0.02
Hz), N (0.02-0.04 Hz), and M (0.04-0.2 Hz) bands. p-values shown on top of each group of the bars in (b)
represent one-way ANOVA results. All error bars indicate the standard error of the mean. ***: p<0.001. I
values represent intraclass correlation coefficients for each group.
Specifically, the bCONHbO and bCONCCO values averaged over the grand group (n=130) for
all three E/N/M bands are listed in Table 3.3, with p-values obtained from one-way ANOVA (the
fifth column for the left) and Tukey’s post hoc test (the sixth column). A significant difference in
bCONHbO (or bCONCCO) was observed among the three frequency bands. Next, we identified that
bCONHbO values at the E and N bands were not significantly different using Tukey’s post hoc test
and were statistically equivalent based on TOST. Therefore, these two indices were pooled to
achieve an averaged connectivity value. The same statistical analysis and spectral average over the
E and N bands were achieved for the bCONCCO values too, as listed in the rightmost column of
Table 3.3. In this case, we found two more ISO features (i.e., bilateral hemodynamic and metabolic
connectivity) that may be characteristic in the resting-state prefrontal human cortices.
54
Table 3. 3 Resting-state prefrontal connectivity (bCONHbO and bCONCCO) averaged over the grand data
set (n=130) at E/N/M band
Bilateral
Connectivity
Endogenic
(mean
±
s.d.)
Neurogenic
(mean
±
s.d.) Myogenic
(mean
±
s.d.)
ANOVA over
three bands
(p-value)
E vs N
Tukey’s
(p-value)
E vs N
TOST
(pTOST)
E, N average
(mean
±
s.d.)
bCONHbO 0.75 ± 0.20 0.78 ± 0.16 0.71 ± 0.10 < 0.003 0.35 < 0.001;
laterally
equivalent
0.77
±
0.17
bCONCCO 0.31 ± 0.21 0.30 ± 0.21 0.14 ± 0.06 < 0.001 0.89 <0.04;
laterally
equivalent
0.31
±
0.21
Note: The p-values in the 5th column from the left were obtained from one-way ANOVA to
compare those at the E/N/M bands. The p-values in the 6th column from the left were
obtained from Tukey’s post hoc test to compare bCON values averaged over the E and N
bands. The pTOST values were obtained from TOST.
Another set of coherence analyses was performed to determine the cerebral hemodynamic-
metabolic coupling on each prefrontal side. Unilateral coupling between Δ[HbO] and Δ[CCO]
indicates the level at which hemodynamic and metabolic infra-slow oscillations are synchronized
and coupled. Figure 3.6(a) shows the uCOPHbO-CCO values derived from the right and left channels
in each E/N/M band. Paired t-tests revealed that the uCOPHbO-CCO values between the left and right
prefrontal cortices were statistically identical in the E and M bands. Figure 3.6(b) illustrates the
unilateral coupling averaged over each measurement group for the five repeated measurements. A
one-way ANOVA over the five readings showed no significant difference for each coupling pair
on each lateral side, as evidenced by the p-values given at the top of Figure 3.6(b). In addition,
intraclass correlation coefficient (ICC) was calculated and reported in the figure 3.6(b) for the 5
repeated measurements to evaluate the reproducibility of the feature in the subject level. In the
neurogenic band, the uCOPHbO-CCO value in the left prefrontal region was significantly higher than
that on the right side, indicating an intrinsic lateral difference in neurogenic oscillation in resting-
state hemodynamic-metabolic coupling.
55
Figure 3. 6 Left and right resting-state prefrontal uCOPHbO-CCO obtained from (a) combined grand group
(n=130) and (b) separate groups (n=26 each) over endogenic (0.005-0.02 Hz), neurogenic (0.02-0.04 Hz),
and myogenic (0.04-0.2 Hz) frequency bands. p-values above each group of bars in (b) represent results
from ANOVA test. The error bars indicate the standard error of the mean. **: p < 0.01. I values represent
intraclass correlation coefficients for each group.
Table 3.4 lists the uCOPHbO-CCO values over the left and right prefrontal cortices across the
three ISO frequency bands averaged over the grand set of experiments (n=130). P-values obtained
from the paired t-test for uCOPHbO-CCO, left vs. uCOPHbO-CCO, right are reported in the fourth column
from the left of Table 3.4. In the case of no significant difference in uCOPHbO-CCO between the left
and right channels in the endogenic and myogenic bands, TOST for equivalence tests were
perfomed with the pTOST values reported in Table 3.4. Accordingly, the bilateral average of
uCOPHbO-CCO was calculated and represented in the rightmost column. These bilaterally averaged
uCOPHbO-CCO values at the E and M bands are the 7th and 8th features that we identified in this study
as potential biomarkers for characterizing brain disorders in the future.
56
Table 3. 4 Prefrontal uCOPHbO-CCO values on the left and right cortical regions averaged over the grand set
of measurements (n=130) at each of the E/N/M bands
Frequency
band uCOPHbO-CCO,left
(mean
±
s.d.) uCOPHbO-
CCO,right
(mean
±
s.d.) Left vs right
t-test (p-value) Left vs right
TOST (pTOST)
Bilateral average
(mean
±
s.d.)
Endogenic 0.33 ± 0.22 0.29 ± 0.20 0.11 0.02;
laterally
equivalent
0.31
±
0.21
Neurogenic
0.31 ± 0.20
0.24 ± 0.17
< 0.01
0.7
Left > right
Myogenic 0.20 ± 0.10 0.18 ± 0.08 0.18 0.04;
laterally
equivalent 0.19
±
0.09
Note: p-values and pTOST were obtained from paired t-tests and TOST, respectively, between
the left and right uCOPHbO-CCO.
3.4 Discussion
NIRS-based methods have been demonstrated and reported as well-known, non-invasive
approaches to monitor the metabolic and hemodynamic activity of the human brain, and thus
having great potential for clinical applications [136, 151, 152]. Because NIRS quantifies only
changes in cerebral hemodynamics, it is not applicable in clinical practice for disease diagnosis or
monitoring when the human brain is in a resting state. For instance, diffuse correlation
spectroscopy and functional NIRS only detect the relative blood flow index and relative changes
in hemoglobin concentration, respectively [153].
To address this weakness of NIRS, we developed a frequency-domain analysis to determine
the spectral amplitudes (SA) and coherence indices for each ISO time series on both sides of the
prefrontal cortex. While both the [HbO] and [CCO] time series in the resting brain expressed
changes with respect to a baseline point, each SA of the oscillation would be an absolute value, in
µM, and signified the respective oscillation amplitude. Because the analysis was performed in the
frequency domain, SA values in the E/N/M bands denoted oscillation magnitudes of HbO and
57
CCO at the three respective rhythms. In addition, coherence indices are absolute values in the
range of 0 to 1, regardless of the unit. It represents the degree of oscillatory similarity between the
two neurophysiological rhythms. Accordingly, we developed and demonstrated a low-cost,
portable, 2-channel bbNIRS system to record cerebral hemodynamic and metabolic ISO activity
over the prefrontal cortex of healthy young humans with a relatively large sample size ( n=26).
The recorded signals were analyzed using a frequency-domain approach to quantify the spectral
amplitude and connectivity/coupling of ISO in the resting human brain. As discussed below, this
study enabled us to prove and support our hypothesis by achieving absolute quantification of ISO-
resolved hemodynamics and metabolism in the resting-state prefrontal human cortices. The
quantified metrics were shown to be relatively stable and thus may have great potential to be
developed as biomarkers for the characterization, diagnosis, and monitoring of certain brain
disorders.
3.4.1 ISO Spectral Amplitudes of Prefrontal [HbO] and [CCO] as Brain-state Features
The ISO (0.005-0.2 Hz) consists of three distinct frequency components: endogenic (0.005-0.02
Hz), neurogenic (0.02-0.04 Hz), and myogenic (0.04-0.2 Hz). Each frequency band is associated
with a specific neurophysiological activity in the healthy human brain [115-117, 154]. Thus,
abnormal brain activity and neurological disorders in the human brain are associated with impaired
or irregular patterns of cerebral hemodynamic and metabolic ISOs. Several studies have reported
a relationship between ISO impairment and cardiovascular disease, Alzheimer’s disease,
hypertension, and stroke [40, 41, 119].
In this study, we quantified ISO-resolved spectral amplitudes of prefrontal oxygenated
hemoglobin (SAHbO) and oxidized cytochrome c oxidase (SACCO) in healthy young humans. Next,
we demonstrated that the SAHbO and SACCO values averaged over 26 participants with five different
58
repeated measurements were relatively stable and consistent, as evidenced by the ANOVA results
shown in Figure 3.4(b). Furthermore, as shown in Figure 3.4(a) and Table 3.1, the average SAHbO
values over the two lateral sides of the prefrontal cortex were statistically equivalent, indicating
similar levels of hemodynamic oscillation in the bilateral prefrontal cortices in all vasomotion-
derived ISO (i.e., E/N/M) bands. Thus, a set of standard prefrontal SAHbO values for each of the
three ISO bands can be established and further examined as features that characterize abnormal
brain functions. As shown in Figure 3.4(c) and Table 3.2, identical or equivalent levels of grand-
averaged SACCO between the two prefrontal cortices in the myogenic band are unambiguous,
implying that this metric can be considered another potential prefrontal feature.
It is worth noting that the grand-averaged SACCO indices were significantly larger on the left
than on the right prefrontal side in both the E- and N-bands. This observation may imply higher
metabolic activity in endogenic and neurogenic rhythms on the left side than on the right side of
the resting prefrontal cortex. As reported by [155], the dominant source of resting-state
lateralization in human brain activity is the default mode network (DMN), which is associated with
internal thoughts. The study [155] provided evidence that the DMN is predominantly active in the
left prefrontal cortex in the resting state, especially in right-handed participants. The implication
of our observation matched well with the results of Liu et al. (2009) because most of our
participants were right-handed and would give rise to higher anterior default-mode activity in the
left prefrontal cortex [155, 156]. Furthermore, as illustrated in Figures 3.4(a) and 3.4(c), the
endogenic band had higher values for SAHbO and SACCO than those of the other two bands. This
phenomenon has also been reported in other studies that have used different NIRS systems and
analysis algorithms as well [35, 124, 157].
59
3.4.2 Cerebral Hemodynamic and Metabolic ISO Connectivity/Coupling as Features
As shown in Figure 3.5(a) and Table 3.3, robust bilateral connectivity of prefrontal hemodynamics
(bCONHbO) was identified over the E/N frequency bands in 26 young healthy human subjects at
rest, over five repeated measurements. This high level or index of connectivity may imply
synchronized hemodynamic activity mediated by endothelial cells and inter-neurons on lateral
prefrontal regions [158-160]. In contrast, bCONCCO showed a significantly lower level of bilateral
functional connectivity than bCONHbO. A lower level of bCONCCO can be expected because
unilateral [CCO] activity is locally driven by oxygen consumption and/or mitochondrial
metabolism within neurons, has specific functions distinct from those of the other lateral prefrontal
cortex, and has less need to link to the other side. Similar to bCONHbO, bCONCCO was statistically
identical over the E and N frequency bands. These observations are in good agreement with those
of a recent study by our group, which had a smaller sample size and utilized a different 2-bbNIRS
setup and analysis methods [43]. Furthermore, Figure 3.5(b) illustrates the ANOVA-driven results
with non-significant differences among the bCONHbO (or bCONCCO) values over 130
measurements. These high stabilities suggest and support the possibility of using these metrics as
new neurophysiological features to characterize the human brain state.
As the final metric, the unilateral hemodynamic-metabolic coupling (uCOPHbO-CCO) in the
right and left channels is plotted in Figure 3.6(a). As shown in this figure and reported in Table
3.4, no significant difference in uCOPHbO-CCO between the two lateral sides existed at both the E
and M bands; thus, bilaterally pooled coupling indices at both E/M bands could be achieved.
Similar to the other metrics shown above, Figure 3.6(b) illustrates the ANOVA-driven results of
the statistically non-significant uCOPHbO-CCO values/indices bilaterally over 130 measurements in
60
the E and M bands. Thus, prefrontal uCOPHbO-CCO in the E and M bands can be included and tested
as resting-state features in future clinical applications.
In contrast, the neurogenic component of uCOPHbO-CCO was significantly higher in the left
prefrontal cortex than in the right prefrontal cortex. This observation may imply a higher vascular-
metabolic interactivity on the left prefrontal cortex than on its contralateral side. Given that most
of our participants were right-handed, higher hemodynamic-metabolic coupling in the left
prefrontal cortex would be expected [155, 156].
As mentioned in the Introduction, relaxation-contraction cycles of blood vessel walls are
expected to be the driving force for the ISO rhythms of cerebral hemodynamics. However, the
driving force of the ISO for CCO is unclear. A recent study using fMRI and PET demonstrated a
strong correlation between slow oscillations (0.01-0.1 Hz) of hemodynamics and metabolism in
the brain [161]. Specifically, the authors concluded that metabolic demand for glucose and oxygen
regulates low-frequency hemodynamic fluctuations. Because of the strong correlation and thus
close coupling between HbO and CCO constituents, we speculated that the three E/N/M
oscillations originating from slow vasomotion may be passed or translated to mitochondrial (CCO)
oscillations at the E/N/M rhythms.
3.4.3 Eight Measurable Features of Prefrontal ISO
In Section 3.3, we confirmed our hypothesis that prefrontal cortical connectivity and coupling
of the ISO can be quantified using 2-bbNIRS as features that reflect the brain state. Specifically,
through the aforementioned content, we demonstrated several stable or consistent metrics based
on prefrontal bilateral connectivity and unilateral coupling of ISO. Based on the analyses and
discussions given in Sections 3.4.1 and 3.4.2, we list eight metrics as measurable features in Table
3.5. These features can be further studied and validated using a larger sample size of both healthy
61
human participants and patients with certain brain disorders. In addition, the ICC values for
different features demonstrated that the subject-level reproducibility needs to be improved and
more robust methodology must be obtained by modifying the current method.
Table 3. 5 Measurable ISO features for characterization of the prefrontal human brain at rest
ISO features (frequency band)
Average over two lateral sides (µM)
SAHbO (E) 0.16±0.07
SA
HbO
(N)
0.09±0.04
SA
HbO
(M)
0.05±0.02
SA
CCO
(M)
0.007±0.002
ISO features (frequency band) connectivity between two lateral sides
bCONHbO (E/N) 0.77±0.20
bCON
CCO
(E/N)
0.31±0.18
ISO features (frequency band) Average over two lateral sides
uCOPHbO-CCO (E) 0.31±0.21
uCOP
HbO-CCO
(M)
0.19 ± 0.09
3.4.4 Limitations
First, the relatively low sampling frequency and short data collection duration (i.e., 7-min)
prevented us from achieving high-frequency resolution, which may have led to low accuracy in
spectral amplitude and coherence calculations in the low-frequency range, especially in the
endogenic band. It is suggested to have a longer measurement duration, for example, 10 min or
longer. Second, our bbNIRS system was sensitive to motion; the eyes-closed resting-state protocol
may have caused sleepiness in the participants during the measurements. Finally, our quantified
results or metrics may be contaminated by the extracranial layers of the human head. It is known
that fNIRS signals obtained over the scalp of human participants are contaminated by extracranial
layers, namely, the human scalp and skull. To minimize this potential confounding factor,
additional optical channels of fNIRS with a short source-detector (S-D) separation (commonly
~0.8-1.2 cm) have been used for systemic noise removal in task-evoked hemodynamic studies
62
[162-166], where a cortical region was activated by stimulating tasks. However, most fNIRS-based
studies for quantifying resting-state functional connectivity (RSFC) have not developed an
appropriate methodology to remove this confounding effect [147, 148]. It is reported only recently
that RSFC can be quantified more accurately with a short S-D reading correction than without
correction [167].
3.4.5 Future work
In future work, to enable a longer-period and less-artifact recording from the human brain,
modifications or improvements are needed in the bbNIRS setup, measurement protocol, and
computational methods to reduce movement artifacts and systemic/physiological noises. In
addition, it is necessary to consider the implementation of short-distance channels in bbNIRS to
remove the possible contamination of extracranial layers from the determined/interpreted results.
The current study included only healthy controls without any disease-related patients; thus, it
was an exploratory study [42]. While we believe that the identified ISO features are good
neurological representations of the human brain, proof-of-principle or confirmatory research must
be conducted for these features to become biomarkers of neurological diseases. Such studies
include two parts. First, the features need to be stable, reliable, and with known or tested
dependence on age, sex, and brain state. All of these quantifications need to be obtained using a
statistically large sample size of healthy controls. Second, the features must be efficient in
significantly classifying controls and patients with selected neurological disorders. Third, since the
final verified biomarkers will be used to identify physiological disorders in individual subjects,
more robust methodology needs to be developed to obtain consistent biomarkers with high
reproducibility for each subject.
63
In addition, the observed differences between left and right prefrontal cortices in features such
as SACCO and uCOPHbO-CCO needs to be further investigated to understand the physiological and
functional mechanism behind these features in different locations of PFC.
3.5. Conclusion
In this study, we hypothesized that 2-bbNIRS, along with frequency-domain analysis, enables the
quantification of prefrontal cortical connectivity and coupling of ISO in the resting human brain.
To test this hypothesis, we implemented 2-channel bbNIRS and performed bilateral, prefrontal, 7-
min measurements in an eyes-closed resting state in vivo from 26 young and healthy participants,
repeated 5 times over 5 weeks. The measured time series were analyzed using a frequency-domain
approach to detect cerebral hemodynamic and metabolic ISO in three endogenic, neurogenic, and
myogenic frequency bands at rest. Specifically, coherence analysis facilitated the quantification of
bilateral connectivity and unilateral hemodynamic-metabolic coupling in the human prefrontal
regions. Accordingly, we identified eight stable resting-state ISO-specific metrics or features,
including bilaterally averaged SAHbO in all three bands, bilaterally averaged SACCO in the M band
only, and bilaterally connected network metrics for both bCONHbO and bCONCCO, each of which
were statistically identical in the E and N frequency bands, respectively. The last two features
were the bilaterally averaged coupling indices of uCOPHbO-CCO over the E- and M-bands, given
that the coupling indices were statistically equivalent for both bands. All eight metrics as features
showed a statistically stable level for 130 measurements. In short, this exploratory study developed
a quick, low-cost, and effective methodology for exploring several prefrontal cortical connectivity
and coupling features in the resting, healthy, and young human brains. The framework reported in
this paper has demonstrated the potential of ISO features to be translatable for future clinical
64
applications, while further confirmatory studies are needed before these features become effective
biomarkers to identify certain neurological disorders.
3.6. Effect of Gender on Measurable Features of Prefrontal ISO
After separating the female and male groups, the proposed measurable features were compared
between these groups using paired t-test and TOST. The features with similar value between male
and female groups are represented in table 3.6.
Table 3. 6 Measurable ISO features for characterization of the prefrontal human brain at rest with similar
values between male and female groups (the reported values are average of all measurements, including
male and female subjects)
ISO features (frequency band) Average over two lateral sides (
µ
M)
SAHbO (E) 0.16±0.07
ISO features (frequency band) connectivity between two lateral sides
bCONHbO (E/N) 0.77±0.20
bCON
CCO
(E/N)
0.31±0.18
ISO features (frequency band)
Average over two lateral sides
uCOPHbO-CCO (E) 0.31±0.21
uCOP
HbO-CCO
(M)
0.19 ± 0.09