Content uploaded by Mahasweta Bhattacharya
Author content
All content in this area was uploaded by Mahasweta Bhattacharya on Feb 15, 2023
Content may be subject to copyright.
Neurocrit Care
https://doi.org/10.1007/s12028-020-00971-x
ORIGINAL WORK
Resting-State NIRS–EEG inUnresponsive
Patients withAcute Brain Injury: A
Proof-of-Concept Study
Marwan H. Othman1†, Mahasweta Bhattacharya2†, Kirsten Møller3,4, Søren Kjeldsen5, Johannes Grand5,
Jesper Kjaergaard4,5, Anirban Dutta2 and Daniel Kondziella1,4*
© 2020 Springer Science+Business Media, LLC, part of Springer Nature and Neurocritical Care Society
Abstract
Background: Neurovascular-based imaging techniques such as functional MRI (fMRI) may reveal signs of conscious-
ness in clinically unresponsive patients but are often subject to logistical challenges in the intensive care unit (ICU).
Near-infrared spectroscopy (NIRS) is another neurovascular imaging technique but low cost, can be performed serially
at the bedside, and may be combined with electroencephalography (EEG), which are important advantages com-
pared to fMRI. Combined NIRS–EEG, however, has never been evaluated for the assessment of neurovascular coupling
and consciousness in acute brain injury.
Methods: We explored resting-state oscillations in eight-channel NIRS oxyhemoglobin and eight-channel EEG band-
power signals to assess neurovascular coupling, the prerequisite for neurovascular-based imaging detection of con-
sciousness, in patients with acute brain injury in the ICU (n = 9). Conscious neurological patients from step-down units
and wards served as controls (n = 14). Unsupervised adaptive mixture-independent component analysis (AMICA) was
used to correlate NIRS–EEG data with levels of consciousness and clinical outcome.
Results: Neurovascular coupling between NIRS oxyhemoglobin (0.07–0.13 Hz) and EEG band-power (1–12 Hz) sig-
nals at frontal areas was sensitive and prognostic to changing consciousness levels. AMICA revealed a mixture of five
models from EEG data, with the relative probabilities of these models reflecting levels of consciousness over multi-
ple days, although the accuracy was less than 85%. However, when combined with two channels of bilateral frontal
neurovascular coupling, weighted k-nearest neighbor classification of AMICA probabilities distinguished unresponsive
patients from conscious controls with > 90% accuracy (positive predictive value 93%, false discovery rate 7%) and,
additionally, identified patients who subsequently failed to recover consciousness with > 99% accuracy.
Discussion: We suggest that NIRS–EEG for monitoring of acute brain injury in the ICU is worthy of further explora-
tion. Normalization of neurovascular coupling may herald recovery of consciousness after acute brain injury.
Keywords: Cardiac arrest, Coma, Consciousness, Electroencephalography, Near-infrared spectroscopy, Neurovascular
coupling, Neurovascular unit, Prognosis, Traumatic brain injury
Introduction
Neurovascular coupling refers to interactions within
the “neurovascular unit,” which consists of neurons,
astrocytes, and vascular cells, including the blood–
brain barrier (Fig. 1). Briefly, neuronal activation is
accompanied by increased cerebral blood flow and
*Correspondence: daniel_kondziella@yahoo.com
†Marwan H. Othman and Mahasweta Bhattacharya have contributed
equally to this work.
‡Anirban Dutta and Daniel Kondziella have contributed equally to this
work.
1 Department of Neurology, Rigshospitalet, Copenhagen University
Hospital, Blegdamsvej 9, 2100 Copenhagen, Denmark
Full list of author information is available at the end of the article
increased cerebral metabolic rate for oxygen, leading
to functional hyperemia and energy supply [1]. Neu-
rovascular coupling thus reflects the close temporal
and regional connection between neuronal activity
and cerebral blood flow [1–4]. Functional brain imag-
ing techniques such as functional magnetic resonance
imaging (fMRI) and near-infrared spectroscopy (NIRS)
rely on neurovascular coupling to infer changes in neu-
ronal activity; and neurovascular coupling is the basis
of the BOLD fMRI signal. Of note, some brain-injured
patients who are unresponsive at the bedside show
evidence of (partially) preserved consciousness when
examined by active and/or resting-state fMRI [5–10].
However, fMRI-based paradigms are labor inten-
sive, expensive, logistically challenging, and not read-
ily available in the intensive care unit (ICU) [9–11]. A
cheap, easy-to-apply test for consciousness assessment
is needed.
NIRS and EEG are low-cost devices that can be admin-
istered at the bedside, which is an important advantage
in the ICU compared to fMRI [12]. EEG captures neu-
ronal activities with poor spatial (centimeters) but excel-
lent temporal (milliseconds) resolution. EEG within 24h
predicts neurological outcome of comatose patients after
cardiac arrest, albeit with low sensitivity [13, 14]. Since
altered hemodynamics contributes to anoxic damage,
integrating neurovascular coupling in the assessment of
patients with anoxic brain injury might help to increase
sensitivity [15].
NIRS is a promising tool in this regard. A noninvasive
optical method, NIRS measures local changes in oxygen-
ated and deoxygenated hemoglobin in the outmost lay-
ers of the cerebral cortex, consistent with spontaneous
cerebral oscillations [16–23]. Low (≈ 0.1 Hz) and very
low (≈ 0.05 to 0.01Hz) frequency oscillations of oxy-Hb
are believed to reflect cortical cerebral autoregulation.
us, like transcranial Doppler, NIRS assesses cerebral
autoregulation with high temporal resolution, but while
Doppler measures blood flow velocity in large cerebral
vessels, NIRS detects changes in microcirculatory blood
volume and blood flow by measuring cortical oxy-Hb
and deoxy-Hb. NIRS offers distinct advantages compared
to transcranial Doppler in being operator independent,
easier to perform and assessing cortical tissue. Method-
ologies based on NIRS have been applied in a variety of
neurological conditions, including ischemic stroke [e.g.,
18,24,26–28], subarachnoid hemorrhage [27], migraine
[23], traumatic brain injury [28], post-cardiac arrest
anoxic brain injury [29], and neonatal encephalopathy
[15, 30]. For a recent review about the application of
NIRS in neurological conditions, including technical and
analytical limitations, see [31]. NIRS captures cerebral
metabolic changes with lower temporal resolution than
EEG because of the inherent hemodynamic delay, but
NIRS has better spatial resolution and is less susceptible
to electrical noise, sedation and muscle artifacts [12, 25,
28]. (Commercially available portable functional NIRS
systems have a depth sensitivity of about 1.5cm and a
spatial resolution up to 1cm [32].) Functional imaging
signals are characterized by slow fluctuations (≤ 0.1 Hz)
[33], originating from neuronal (EEG) and, as already
explained, hemodynamic (NIRS) signals [34], which have
been proposed as “central pacemaker oscillations” [35].
e vascular origin of these 0.1Hz oscillations has been
assigned to vasomotion [36] and Mayer waves [37], and
the neuronal origin to neurovascular coupling [38–40].
While EEG and fMRI have long-standing roles for the
evaluation of consciousness following brain injury [8],
the application of NIRS in this field has recently gained
momentum [4, 41–45]. Of note, NIRS–EEG hybrid sys-
tems combine the benefits from EEG and NIRS (Fig.1),
while avoiding the logistical challenges associated with
fMRI. Neurovascular coupling revealed by NIRS–EEG
has been proposed as an important feature of the rest-
ing brain, the hypothesis being that the aforementioned
fluctuations of EEG band-power corresponding with the
NIRS oxyhemoglobin oscillations around 0.1 Hz repre-
sent fluctuations in brain excitability [26, 38, 40]. Corrob-
orating this idea, cerebral excitability is more impaired in
the vegetative state/unresponsive wakefulness syndrome
(VS/UWS) than in the minimally conscious state (MCS)
[46]. However, although NIRS–EEG has been studied in
Fig. 1 Resting-state NIRS-EEG, using a wireless wearable bedside
device, allows evaluation of the neurovascular unit, consisting of
astrocytes and other glial cells, neurons, and blood vessel cells.
Neuronal (EEG) and hemodynamic (NIRS) signals are characterized
by a dominant frequency at 0.1 Hz [34]. We postulated that NIRS oxy-
hemoglobin oscillations between 0.07 and 0.13 Hz would be found in
most acute brain injury patients, but that the state of the neurovas-
cular coupling would be affected by the level of consciousness, and
that recovery of neurovascular coupling would occur in parallel with
recovery of consciousness. EEG—electroencephalography; fNIRS—
functional near-infrared spectroscopy
children with perinatal hypoxic injury [47–50], it has not
yet been applied to adults with acute brain injury.
In the present study, we explored the feasibility of low-
density NIRS–EEG to assess the evolution of neurovas-
cular coupling with improving (or declining) levels of
consciousness in unresponsive adult patients with acute
brain injury. We postulated that resting-state frontal
NIRS oxyhemoglobin oscillations would be increasingly
coupled with EEG band-power oscillations [38], when
consciousness recovers after brain injury, reflecting
increased cerebral excitability.
Primary objective: To assess the feasibility of NIRS–
EEG for the assessment of neurovascular coupling and the
classification of consciousness levels in ICU patients with
brain injury
We hypothesized that neurovascular coupling in
patients from the ICU and step-down units could be
evaluated with low-density (eight-channel) resting-state
NIRS-EEG. We further hypothesized that an adaptive
mixture-independent component analysis (AMICA)
[51] classifier could be trained to learn the transiently
changing models across different levels of conscious-
ness, thereby assessing neurovascular coupling during
deterioration and recovery of consciousness [52] (Fig.2).
Such a probabilistic mixture framework enables accom-
modation of non-stationary environments and arbitrary
EEG source densities [53] which occur with shifting con-
sciousness levels in physiological [54] and unphysiologi-
cal conditions [48, 49].
Secondary objective: To assess the feasibility of NIRS–
EEG for prognostication of ICU patients with brain injury
following cardiac arrest or trauma
We finally hypothesized that neurovascular coupling
data from NIRS–EEG would enable us to distinguish
patients who recover consciousness from those who
fail to do so. Hence, we expected patients with good
prognosis to show rapid normalization of neurovascular
coupling after cessation of sedation, while this would not
be the case in patients with poor prognosis.
Methods
Target Population, Inclusion andExclusion Criteria,
andClinical Investigations
NIRS–EEG assessments were performed in a conveni-
ence sample of neurological/neurosurgical patients in
the neuro-ICU and post-cardiac arrest patients in the
cardiological ICU at a tertiary referral hospital (n = 9).
Exclusion criteria were major scalp injury and acute
life-threatening cardiovascular instability. Conscious
patients, admitted to the neurological step-down unit,
served as controls (n = 14). Levels of consciousness
were estimated by MHO under the supervision of DK,
a board-certified neurologist experienced in neuro-
critical care, following detailed neurological examina-
tion, including the Full Outline of UnResponsiveness
(FOUR) [55], and categorized into coma, VS/UWS,
minimal conscious state (MCS) minus, MCS plus, or
emerged from MCS (eMCS) [56–58]. Briefly, coma was
defined as a state of profound unawareness from which
patients cannot be aroused, while eyes are closed, and a
normal sleep–wake cycle is absent [59]. We diagnosed
VS/UWS, if patients opened their eyes but exhibited
only reflex behaviors and were considered unaware of
themselves and their surroundings [60]. Further, we
classified patients as MCS, when they showed unequiv-
ocal signs of non-reflex behaviors occurring inconsist-
ently, yet reproducibly, in response to environmental
stimuli [61]. MCS patients were further classified into
MCS plus (i.e., if they were able to obey commands) or
MCS minus (i.e., if they localized pain, exhibited visual
pursuit, or showed appropriate emotional expressions)
Fig. 2 This figure depicts methods of signal processing and statistical analysis used in the present study, which can be divided into three compo-
nents (a–c). Continuous and non-stationary (i.e., constantly fluctuating) dynamics of brain network activity, underlying human cognition and behav-
ior, are challenging to assess by statistical evaluation, but adaptive mixture-independent component analysis (AMICA) may be a suitable approach
[52] (a, blue). AMICA performs an independent component analysis decomposition on EEG data with five ICA models to model EEG dynamics
that are associated with brain state changes during the recovery of consciousness. AMICA consists of three layers of mixing, where the first layer
consists of mixture of ICA models that learn the underlying data clusters, the second layer consists of mixture of independent components (ICs)
that decompose the data cluster into statistically independent source activations, and the third layer consists of mixture of generalized Gaussian
distribution that approximates the probability distribution of the source activation. Non-stationary model probabilities of the five ICA models that
were learned by AMICA EEG decomposition were used in conjunction with the averaged wavelet coherence (b, yellow) between the NIRS oxygen-
ated hemoglobin signal in the frequency band between 0.07 Hz and 0.13 Hz and the EEG band-power (1–12 Hz) signal at the frontal areas (F3 and
F4, 10–20 system) as features for classification of unresponsive/low-responsive patients from conscious controls as well as prognostication of con-
sciousness recovery. Here, the first two principal components of the non-stationary model probabilities of the five ICA models accounted for > 90%
variance, which along with two channels of averaged wavelet coherence were used in the weighted k-nearest (k = 10, Euclidean distance metric
with squared inverse distance weight) neighbor classifier and evaluated based on receiver operating characteristic curve (c, green). AMICA—adap-
tive mixture-independent component analysis; EEG—electroencephalography; fNIRS—functional near-infrared spectroscopy; ICs—independent
components; PDFs—probability density functions (Color figure online)
(See figure on next page.)
[62]. Patients who recovered functional communication
or functional object use were considered as “emerged
from MCS” (eMCS) [61]. Hence, rather than the total
FOUR score, classification of consciousness levels into
coma, VS/UWS, MCS minus/plus, and eMCS was
based on visual and motor FOUR subscales, as well as
evidence of command following, appropriate emotional
expressions, communication, and functional object use.
Clinical data were stored in a dedicated database, fol-
lowing approval from local authorities.
Experimental Setup forNIRS–EEG Recordings
NIRS–EEG was conducted at the bedside, using the
wireless StarStim NIRS–EEG system (Artinis Medical
Systems, 6662 PW Elst, e Netherlands) [63] (online
supplemental files, FiguresS1 and S2). Technical details
of the StarStim NIRS–EEG system [63] are as follows:
for NIRS, eightsource wavelengths at nominal 760 and
850nm; two photodiodes with integrated ambient light
protection; Bluetooth connection (up to 100 meters) for
online measurements at 50Hz sampling rate; and up to
6h of recording with one interchangeable and recharge-
able battery. For EEG, 8 channels available at a sampling
rate of 500Hz; bandwidth: 0 to 125Hz (DC coupled); res-
olution: 24 bits–0.05µV resolution; noise: < 1µV RMS;
common mode rejection ratio: -115dB; input impedance:
1 GΩ; and operating time: 5h 10min when using Wi-Fi
connection. EEG data were processed using EEGLAB
toolbox [64] in MATLAB.
We recorded from F3, F4, C3, C4, P3, and P4 scalp
locations (10–20 positioning system) following Lehembre
etal. [65] and added Fp1 and Fp2 scalp locations (10–20
positioning system) for prefrontal recordings. Eight NIRS
sources were positioned at AF7, AF3, AF8, AF4, CP4,
FC4, CP3, and FC3, and the two NIRS detectors at Cz
and FPz at an optode distance of around 35mm. e sen-
sitivity analysis for the NIRS is explained in detail in the
online supplemental files (Appendix). NIRS–EEG data
[24] were recorded in 15-min segments for 45min to 1h
with the patients resting and eyes closed. Unconscious
patients were examined by resting-state NIRS–EEG from
day 1 after admission to discharge from the ICU/step-
down units.
AMICA Signal Processing, Neurovascular Coupling,
andClassier Training
Figure2 provides an overview of data acquisition, signal
processing and statistical analysis. Technical details are
outlined (Appendix, online supplemental files). An inde-
pendent statistical review of the methods, required by
one of the referees of this paper, can be accessed online
(Statistical Report, online supplemental files).
EEG data were first down-sampled to 250Hz, follow-
ing application of a 1–40 Hz band-pass filter, and the
EEG data were then re-referenced to the common aver-
age. en, the AMICA signal processing was applied
(Appendix, online supplemental files). Briefly, AMICA
is an unsupervised approach that can learn the models
underlying non-stationary (i.e., constantly fluctuating)
unlabeled EEG data (Fig.2a). As outlined earlier, we pos-
tulated that AMICA models could capture the changing
brain states that correspond to transitions between dif-
ferent levels of consciousness. Restated, AMICA may
offer a promising unsupervised tool to learn the changes
in EEG patterns, including those from low-density EEG
recordings [52], during recovery of consciousness. As
we aimed to categorize levels of consciousness into
coma, VS/UWS, MCS minus, MCS plus and eMCS, five
AMICA models were used with one model for each level
of consciousness.
One of the challenges with NIRS data is a high sen-
sitivity for scalp and other extracranial hemodynam-
ics, including changes in the heartbeat and breathing
cycle, which can mask cerebral activation [66]. Differ-
ent methods have been proposed to remove extracra-
nial hemodynamics from continuous-wave NIRS signal
[67], including short-distance detector approaches [68]
and tomography [69]. us, larger changes are observed
in the extracranial oxygenated hemoglobin when com-
pared to extracranial deoxygenated hemoglobin concen-
tration. We used HOMER2 routines [70] in MATLAB
(MathWorks, Inc.) for the NIRS data analysis based on
the modified Beer–Lambert law (technical details in the
Appendix, online supplemental files). We evaluated neu-
rovascular coupling between NIRS oxygenated hemo-
globin in the frequency band between 0.07 and 0.13Hz
and the EEG band-power (1-12Hz) signals at the fron-
tal areas (F3 and F4, 10–20 system), using wavelet cross-
spectrum analysis (Fig. 2b). To determine the threshold
for neurovascular coupling, we used the statistical test
developed by Bigot and colleagues [71]. us, the wavelet
coherence data between 0.07 and 0.13Hz were divided
into 1-min epochs (N = 30 for 30-min data) to measure
statistical significance at alpha = 0.01 based on Bigot
et al. [71]. For significant neurovascular coupling, the
phase from the WCS was used to indicate the relative
lag between NIRS oxygenated hemoglobin and the EEG
band-power. A custom MATLAB script was written for
these computations.
To illustrate the separability of NIRS–EEG data using
the feature space learned by AMICA, we selected a
weighted k-nearest neighbor classifier because of the
low-dimensional data in our study (Fig.2c). Rather than
optimizing the classifier for the best performance, we
wanted to prevent overfitting and to assess the gener-
alization ability of our predictive model. So, we used a
tenfold cross-validation, one of the most widely used
data resampling methods [72], and computed confusion
matrices and receiver operating characteristics curves
(including area under the curves). e results were sum-
marized for each weighted k-nearest neighbor classifier
in the outcome measure.
Outcome Measures
Outcome measures include assessment of low-density
NIRS–EEG data for the classification of the levels of
consciousness (primary) and the prognostication of
the recovery of the unresponsive patients with acute
brain injury (secondary). ese involved development
of supervised classifier models using the Classification
Learner app in the MATLAB (MathWorks, Inc.). We
used AMICA model probabilities and neurovascular cou-
pling as features and applied weighted k-nearest neighbor
classification following principal component analysis
(component reduction for 90% variance accounted for) to
2-s data windows to classify the NIRS–EEG data into (a)
unresponsive ICU patients with acute brain injury (n = 5)
versus 14 conscious controls; (b) 5 unresponsive ICU
patients with acute brain injury into their FOUR scores—
primary outcome measure; and (c) 5 unresponsive ICU
patients with acute brain injury into those who regained
consciousness (n = 3) versus those who failed to recover
(n = 2)—secondary outcome measure. As said, due to
low-dimensional data, weighted k-nearest neighbor clas-
sifier was used to illustrate the separability of NIRS–EEG
data using the feature space learned by AMICA decom-
position and neurovascular coupling. e tenfold cross-
validation accuracy, confusion matrix, and the receiver
operating characteristics curve were computed, and
results were summarized for each weighted k-nearest
neighbor classifier.
Ethics Statement
e Ethics Committee of the Capital Region of Denmark
approved the study and waived the need for written con-
sent because risks were deemed negligible (Reference
J. No. H-19001774). e study was approved under the
quality control legislation for commercially available
medical devices, informed consent was waived, and data
were anonymized prior to statistical analysis.
Results
Clinical Data
We enrolled a convenience sample of 23 patients (11
females; median age 63years, range 19–79years). From
the ICU, we included patients who were monitored over
multiple days (n = 5) or only once (n = 4). Six patients
were in coma or VS/UWS at initial assessment. Diag-
noses included anoxic encephalopathy after cardiac
arrest, traumatic brain injury, and delirium associated
with Guillain–Barré syndrome. NIRS–EEG monitoring
and assessment of consciousness levels were performed
in unsedated patients or, whenever deemed safe by the
treating physicians, during wake-up calls (for dosages
and infusion rates of sedatives, see online supplemental
files). Table1 provides details.
Conscious neurological patients from the neurological
step-down unit served as controls (n = 14; FOUR score
16; 8 females; median age 51.5years, range 19–78years).
Diagnoses included multiple sclerosis (n = 3), ischemic
stroke (n = 3), epilepsy, central nervous system lym-
phoma, central nervous system vasculitis, non-infectious
myelitis, autoimmune encephalitis, motor neuron dis-
ease, Guillain–Barré syndrome, and Susac syndrome (all
n = 1).
Cortical Sensitivity ofthe NIRS Montage
e PHOEBE approach [73] (technical details in Appen-
dix, online supplemental files) validated optical scalp
contact for all frontal optodes (sources at AF7, AF3, AF8,
AF4, and detector at FPz) across 9 ICU subjects (one
sample from Table2 was discarded) and 14 controls from
neurological step-down unit. However, the optodes at the
central areas (sources CP4, FC4, CP3, FC3, and detec-
tor at Cz) were not consistent across most patients due
to signal interference from the subjects’ hair so were not
analyzed.
Primary Study Objective: Classication ofConsciousness
Levels
In unresponsive or low-responsive (FOUR score ≤ 7)
patients from the ICU, average EEG power at 11 Hz,
22Hz, and 34Hz (alpha and beta activity) was centered
at right frontal areas, whereas average EEG power at 2Hz
(delta activity) and 6Hz (theta activity) was centered at
left frontal areas (Fig.3a). erefore, NIRS sources posi-
tioned at AF3 and AF4 and the NIRS detector at FPz were
used for the two-channel neurovascular coupling data at
bilateral frontal areas (F3 and F4, 10–20 system) to inves-
tigate recovery of neurovascular coupling based on WCS,
as shown with an illustrative example in Figs.3b. Repre-
sentative examples of raw EEG data and raw NIRS data
are provided in FiguresS3 and S4; online supplemental
files).
Multi-model AMICA (five models) processed at the
Neuroscience Gateway [74] was completed in 15 h,
43min. e temporal dynamics learned from the EEG
data from ICU patients and conscious controls (in the
neurological ward) with 30-s smoothing by the five-
model AMICA showed that Models 1 and 2 were pri-
marily active in conscious controls (class False), while
models 2 and 3 were primarily active in unresponsive
or low-responsive ICU patients (class True) with acute
brain injury as shown in Fig. 4a. Models 4 and 5 were
largely inactive in both patient groups. Figure4b shows
the differences in the recovery of neurovascular coupling
in ICU patients based on the average wavelet coherence
between 0.07 and 0.13Hz at the frontal areas (F3 and F4,
10–20 system).
Separation of unconscious or low-responsive ICU
patients from conscious controls: Principal component
analysis of the temporal dynamics of the five-model
AMICA from all ICU patients and conscious controls
(in the neurological ward) showed that 100% variance
was accounted for by four principal components with
explained variance per component (in order): 79.6%,
15.2%, 4.0%, 1.2%, and 0%. us, the first two principal
components accounted for > 90% variance, which along
with two channels of neurovascular coupling data were
used to develop weighted k-nearest neighbor classifier
(k = 10, Euclidean distance metric with squared inverse
distance weight) to separate unresponsive or low-respon-
sive (FOUR score ≤ 14) ICU patients from conscious con-
trols. e results of the confusion matrix from the tenfold
cross-validation are tabulated in Table2. is table also
presents the area under the ROC curve to classify unre-
sponsive or low-responsive (FOUR score ≤ 14) ICU
patients. e classifier using NIRS–EEG data provided
an accuracy of 91.4%; the positive predictive values and
the false discovery rates for unresponsive or low-respon-
sive (FOUR score ≤ 14) ICU patients being 93% and 7%,
respectively, and 88% and 12% for conscious controls.
Classification of consciousness levels according to
FOUR scores: Principal component analysis of the tem-
poral dynamics of the five-model AMICA from ICU
patients only showed that 100% variance was accounted
for by four principal components with explained
variance per component (in order): 35.7%, 32.4%,
22.8%, 9.1%, and 0%. Here, the first three principal
Table 1 Patients recruited fromintensive care units
See methods for information on conscious control patients (n = 14) recruited from a neurological step-down unit
eMCS—emerged from MCS, F—female, GBS—Guillain–Barré syndrome, ICU—intensive care unit, M—male, MCS—minimal conscious state, VS/UWS—vegetative
state/unresponsive wakefulness syndrome
$ NIRS–EEG monitoring and assessment of consciousness levels were performed in unsedated patients or, whenever deemed safe by the treating physicians, during
wake-up calls. Data on dosage and infusion rates of sedatives are provided in the online supplemental les (TableS2)
& Classication of consciousness levels was based on visual and motor FOUR subscales rather than the total score
# On day 6 after admission, this patient was extubated and transferred to the neurological step-down unit
‡ Sedation was stopped during NIRS–EEG monitoring. FOUR score before wake-up call
* Sedation was stopped during NIRS–EEG monitoring. FOUR score after wake-up call
† Life-sustaining therapy was withdrawn 12–8h after the last NIRS–EEG recording
Patient ID Number
ofNIRS–EEG
assessment
Age Sex Diagnosis Days
sinceadmis-
sion
Intubated Sedation$FOUR score Level
of consciousness&
ICU 1 1. 79 M Traumatic brain
injury 4Yes No 5 (E0, M2, B2, R1) Coma
2. 5 Yes No 11 (E4, M2, B4,
R1) MCS minus
3.# 6 No No 14 (E2, M4, B4,
R4) eMCS
4. 7 No No 14 (E2, M4, B4,
R4) eMCS
ICU 2 1. 55 F Cardiac arrest 8 Yes Propofol,
remifentanil ‡ * 7 (E0, M2, B4,
R1) ‡
9 (E2, M2, B4,
R1) *
VS/UWS
2. 14 No No 16 (E4, M4, B4,
R4) Conscious
ICU 3 1. 58 M Cardiac arrest 3 Yes No 4 (E0, M0, B4, R0) Coma
2. 4 Yes No 4 (E0, M0, B4, R0) Coma
3. 7 Yes No 4 (E0, M0, B4, R0) Coma†
ICU 4 1. 60 M Cardiac arrest 1 Yes Propofol, fentanyl 4 (E0, M0, B4, R0) Coma
2. 2 Yes Propofol, fenta-
nyl* 13 (E4, M4, B4,
R1) * MCS plus
ICU 5 1. 59 F Cardiac arrest 3 Yes Remifentanil 3 (E0, M0, B2, R1) Coma
2. 4 Yes Remifentanil 5 (E0, M0, B4, R1) Coma
3. 5 Yes Remifentanil 7 (E1, M2, B4, R0) VS/UWS
ICU 6 1. 65 M Cardiac arrest 2 Yes Propofol, fentanyl 2 (E0, M0, B2, R0) Coma†
ICU 7 1. 63 M Cardiac arrest 6 No No 16 (E4, M4, B4,
R4) Conscious
ICU 8 1. 74 F GBS, delirium 2 Yes No 16 (E4, M4, B4,
R4) Conscious
ICU 9 1. 66 M Cardiac arrest 2 Yes Propofol, fenta-
nyl* 16 (E4, M4, B4,
R4) * Conscious
components (accounting for > 90% variance) from the
temporal dynamics of the five-model AMICA along
with two channels of neurovascular coupling data were
used to develop weighted k-nearest neighbor classifier
(k = 10, Euclidean distance metric with squared inverse
distance weight) to identify the FOUR scores catego-
rized as conscious, coma, VS/UWS, or MCS/eMCS of
the ICU patients. e confusion matrix from the ten-
fold cross-validation is shown in Fig.5a, and the ROC
curve to classify fully conscious (FOUR = 16), coma,
VS/UWS, and MCS/eMCS is shown in Fig. 5b. e
accuracy was found to be 87.8%; the positive predictive
values and the false discovery rates for the conscious,
coma, VS/UWS, MCS/eMCS were 87% and 13%; 85%
and 15%; 89% and 11%; and 93% and 7%, respectively.
e ROC curves showed that the positive class, coma,
VS/UWS, and MCS/eMCS could be classified with true
positive rate greater than or equal to 90%; however, the
fully conscious (FOUR = 16) state in the ICU patients
(some under anesthesia) could be classified with true
positive rate of 77% only.
Secondary Study Objective: Prognostication
ofConsciousness Recovery
To distinguish unresponsive ICU patients recovering
consciousness from those failing to do so, we devel-
oped a weighted k-nearest neighbor classifier based on
two principal components (> 90% variance accounted
for) from the temporal dynamics of the five-model
AMICA together with two channels of neurovascular
coupling data from the first NIRS–EEG assessment of
each patient. e results of the confusion matrix from
the tenfold cross-validation are shown in Table2. is
table also presents the area under the ROC curve to
identify patients who regained consciousness. e posi-
tive predictive value and the false discovery rate for
unresponsive patients regaining consciousness were
99% and 1%, respectively, and 99% and 1% for those
who failed to recover.
Discussion
Recovery of consciousness is a very important predictor
for outcome after acute brain injury, and underestima-
tion of consciousness levels in the ICU may cause pre-
mature treatment withdrawal [75]. Reliable assessment
of consciousness levels after acute brain injury is highly
desirable to better predict clinical outcome, to improve
neurorehabilitation potential, and to decrease caregiver
burden and health costs [8].
Integrity and activity of brain networks that are critical
for the recovery of consciousness can be assessed using
fMRI, but this is logistically challenging, expensive, and
incompatible with prolonged monitoring. In contrast,
NIRS, EEG, and NIRS–EEG are low-cost noninvasive
neuromonitoring devices that can be performed seri-
ally or continuously at the bedside, without the need
for in-hospital transport for imaging, which are impor-
tant advantages in the ICU [12, 25, 28, 76–78]. Moreo-
ver, after correction for scalp and other extracranial
hemodynamics [66], NIRS–EEG may provide an extra
layer of information compared with EEG alone, because
neurovascular coupling can be assessed. Neurovascular
coupling is relatively unaffected by common anesthet-
ics, including pentobarbital, isoflurane, and propofol
[79]. erefore, when compared to EEG alone, NIRS–
EEG-based measures of the neurovascular coupling may
provide a marker of the severity of brain injury with com-
paratively little influence from anesthetics [79].
In the present study, we showed the feasibility of apply-
ing a commercially available low-density NIRS–EEG
bedside device to capture neurovascular coupling in
unresponsive traumatic and non-traumatic patients with
acute brain injury and conscious neurological inpatients
Table 2 Correlation ofEEG, respectively NIRS–EEG, data withlevels ofconsciousness andclinical outcome
Performance of the weighted k-nearest neighbor classier to classify in the ICU, 1. unresponsive patients (class True) and 2. unresponsive patients who recovered
consciousness (class True), based on two principal components from the temporal dynamics of the ve-model AMICA (EEG) data with or without two channels of
neurovascular coupling (NIRS + EEG) data. See Methods and Results for details
ICU—intensive care unit, KNN—k-nearest neighbor classier
True posi-
tive (%) False
positive
(%)
Area
underthe
curve
1a. KNN classifier to separate unresponsive patients (class True) in the ICU from conscious controls (EEG only) 82 15 0.89
1b. KNN classifier to separate unresponsive patients (class True) in the ICU from conscious controls (NIRS + EEG) 93 12 0.97
2a. KNN classifier to identify unresponsive patients in the ICU with recovery of consciousness (class True) versus
those without (EEG only) 84 18 0.82
2b. KNN classifier to identify unresponsive patients in the ICU with recovery of consciousness (class True) versus
those without (NIRS + EEG) 99 1 1
Fig. 3 a Topographical scalp map and power spectrum of the eight-channel EEG data at 2 Hz, 6 Hz, 11 Hz, 22 Hz, and 34 Hz from unresponsive
patients in the ICU. b Neurovascular coupling was evaluated based on wavelet coherence between the NIRS O2Hb in the frequency band between
0.07 and 0.13 Hz and the EEG band-power (1–12 Hz) signals at the frontal areas (F3 and F4, 10–20 system). As an example, wavelet coherence
between the NIRS O2Hb and the EEG band-power (1–12 Hz) signals is shown below for patient ICU 1, day 1 (at F3). For significant wavelet coher-
ence, the phase lag of NIRS O2Hb with respect to EEG band-power (1–12 Hz) is shown with arrows that are spaced in time and scale. The direction
of the arrows corresponds to the phase lag on the unit circle. Examples of raw EEG and raw NIRS data are provided in the online supplemental files
(controls). We further showed the feasibility to train a
machine learning algorithm to assess neurovascular cou-
pling and potentially distinguish unresponsive patients
from conscious controls (positive predictive value 93%),
to classify the levels of consciousness after brain injury
according to the FOUR score (positive predictive values
between 85 and 93), and to differentiate unresponsive
patients with recovery of consciousness from those with-
out (positive predictive value 99%). (e accuracy was
less than 85% when using the AMICA probabilities from
EEG data only.)
It remains to be seen if NIRS–EEG performs differ-
ently in conditions of global brain injury (e.g., anoxia)
compared to more focal injury (e.g., stroke, traumatic
brain injury). From a clinical neuro-management per-
spective, the very limited spatial resolution of NIRS is
an important limitation. Admittedly, NIRS shares this
limitation with a range of other neuromonitoring devices
that measure metabolism locally (albeit invasively) such
as microdialysis [80, 81], brain temperature [82], and
brain oxygen tension [82–85]. Documentation of neu-
rovascular coupling in preserved brain tissue far remote
from a brain lesion might be less helpful for prevention
of secondary brain injury in the ICU. However, evidence
of intact neurovascular coupling in crucial brain areas
such as the frontal lobes might factor into algorithms
for outcome prediction and classification of conscious-
ness levels, as shown in the present study. In addition to
resting-state NIRS–EEG, active NIRS paradigms based
on, e.g., motor imagery to detect sign of preserved con-
sciousness [41, 44, 86–90] might add a further layer of
consciousness assessment in the ICU. Although active
and resting-state fMRI paradigms have been shown to
be feasible in the ICU [9, 10], NIRS–EEG would have an
obvious advantage given the low costs and lack of major
logistical challenges. Head-to-head studies comparing
the diagnostic precision of fMRI to that of NIRS–EEG do
not yet exist, to our knowledge.
is feasibility study has limitations that should be
acknowledged. ese include diagnostic heterogeneity,
Fig. 4 a 1–3 Model probabilities of a five-model AMICA decomposition applied to the EEG recordings from unresponsive ICU patients (class True,
green) and conscious controls (class False, red). Models 1 and 2 were primarily active in conscious controls, while models 2 and 3 were primarily
active in unresponsive ICU patients with acute brain injury. b Differences in the recovery of neurovascular coupling in ICU patients (ICU 1–5, Table 1)
were investigated based on the average wavelet coherence between 0.07 and 0.13 Hz at the frontal areas (F3 and F4, 10–20 system). Wavelet coher-
ence above 0.14 was considered statistically significant for the averaged data over 30 epochs (total 30-min data)
small numbers of subjects, and convenience sampling.
Most of our ICU patients had hypoxic-ischemic post-
cardiac injury, which might not be generalizable to other
neurocritical care settings. Moreover, some subjects were
monitored for several days and some only once, if they
were discharged or had life-supporting treatment with-
drawn. Also, consciousness assessment was performed
using neurological bedside examination, including the
FOUR score, although the gold standard currently is the
Coma Recovery Scale-Revised [91]. (Still, the FOUR is a
reasonable alternative in ICU and more discriminative
than the Glasgow Coma Scale [92, 93]). In addition, NIRS
signal interference with the subjects’ hair was noted for
optodes at the central areas, which is an important tech-
nical limitation. Future studies with carefully controlled
prospective patient enrollment and larger numbers will
be needed to determine whether recovery of neurovas-
cular coupling indeed signals recovery of consciousness,
and whether low-density (i.e., 8 channels) is equal to or
inferior to standard (e.g., 28 channels) or high-density
(e.g., 256 channels) recordings. Finally, the present study
was not powered to show that NIRS–EEG is undoubt-
edly associated with better algorithm performance than
EEG alone, and this needs to be investigated further. On
the positive side, strengths of the study include the true-
to-life setting, careful avoidance of extracerebral NIRS
signal contamination, and application of a tenfold cross-
validation to prevent overfitting of our classifier. Further,
an independent statistical review of the methods and sta-
tistics did not find cause for major concern (online sup-
plemental files).
Conclusions andFuture Directions
NIRS–EEG is a noninvasive, low-cost bedside device that
allows serial evaluation of neurovascular coupling in a
naturalistic ICU setting. e resulting data can be used to
train a classifier to potentially distinguish unresponsive
patients with good recovery from those without. Taken
together, our results suggest that NIRS–EEG may be
worth exploring in the future as an add-on to multimodal
neuromonitoring. We hypothesize that recovery of neu-
rovascular coupling after acute brain injury may herald
recovery of consciousness. In the neurocritical care set-
ting, this would be clinically meaningful information.
Electronic supplementary material
The online version of this article (https ://doi.org/10.1007/s1202 8-020-00971 -x)
contains supplementary material, which is available to authorized users.
Fig. 5 a Confusion matrix for the weighted k-nearest neighbor (KNN) classifier to identify the FOUR scores categorized as coma, VS/UWS, minimal
conscious state (MCS) minus, MCS plus, or emerged from MCS (eMCS) of unresponsive patients in the ICU (green, positive predictive values; pink,
false predictive values). b Receiver operating characteristic curve for the weighted k-nearest neighbor classifier, based on two principal components
from the temporal dynamics of the five-model AMICA together with two channels of neurovascular coupling data. AUC—area under the curve;
ROC—receiver operating characteristics (Color figure online)
Author details
1 Department of Neurology, Rigshospitalet, Copenhagen University Hospital,
Blegdamsvej 9, 2100 Copenhagen, Denmark. 2 Department of Biomedical
Engineering, University at Buffalo, State University of New York, Buffalo, NY,
USA. 3 Department of Neuroanesthesiology, Copenhagen University Hospital,
Copenhagen, Denmark. 4 Department of Clinical Medicine, Faculty of Health
and Medical Sciences, University of Copenhagen, Copenhagen, Denmark.
5 Department of Cardiology, Rigshospitalet, Copenhagen University Hospital,
Copenhagen, Denmark.
Author Contributions
AD, DK contributed to the study concept and design. MHO, KM, SK, JG, JK, AD,
DK contributed to data acquisition. MB, AD, DK conducted data analysis and
interpretation. AD, DK involved in writing of manuscript. MHO, MB, KM, SK, JG,
JK, AD, DK provided important intellectual content.
Source of support
This work was supported by the Lundbeck Foundation (MW, DK), Offerfonden
(DK), Savværksejer Jeppe Juhl og Hustru Ovita Juhls Mindelegat (DK), Region
Hovedstadens Forskningsfond til Sundhedsforskning (DK), Jaschafonden
(DK), Rigshospitalets Forskningspulje (DK), and NovoNordisk Foundation
NNF17OC0028706 (JK, JG, and SK). AD and MB were supported by the Univer-
sity at Buffalo, USA. Figures 1 and 2 are created with biorender.com.
Conflict of interests
The authors declare no conflict of interests.
Ethical Approval/Informed Consent
The Ethics Committee of the Capital Region of Denmark approved the study
and waived the need for written consent because risks were deemed negligi-
ble (Reference J. No. H-19001774).
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.
References
1. Iadecola C. The neurovascular unit coming of age: a journey through
neurovascular coupling in health and disease. Neuron. 2017;96:17–42.
2. Phillips AA, Chan FH, Zheng MMZ, Krassioukov AV, Ainslie PN. Neurovas-
cular coupling in humans: physiology, methodological advances and
clinical implications. J Cereb Blood Flow Metab. 2016;36:647–64.
3. Muoio V, Persson PB, Sendeski MM. The neurovascular unit—concept
review. Acta Physiol. 2014;210:790–8.
4. Rupawala M, Dehghani H, Lucas SJE, Tino P, Cruse D. Shining a light on
awareness: a review of functional near-infrared spectroscopy for pro-
longed disorders of consciousness. Front Neurol. 2018;9:350.
5. Owen AM, Coleman MR, Boly M, Davis MH, Laureys S, Pickard JD. Detect-
ing awareness in the vegetative state. Science. 2006;313:1402.
6. Monti MM, Vanhaudenhuyse A, Coleman MR, Boly M, Pickard JD, Tshiba-
nda L, et al. Willful modulation of brain activity in disorders of conscious-
ness. N Engl J Med. 2010;362:579–89.
7. Stender J, Gosseries O, Bruno MA, Charland-Verville V, Vanhaudenhuyse
A, Demertzi A, et al. Diagnostic precision of PET imaging and functional
MRI in disorders of consciousness: a clinical validation study. Lancet.
2014;384:514–22.
8. Kondziella D, Friberg CK, Frokjaer VG, Fabricius M, Møller K. Preserved
consciousness in vegetative and minimal conscious states: systematic
review and meta-analysis. J Neurol Neurosurg Psychiatry. 2016;87:485–92.
9. Kondziella D, Fisher PM, Larsen VA, Hauerberg J, Fabricius M, Møller K,
et al. Functional MRI for assessment of the default mode network in acute
brain injury. Neurocrit Care. 2017;27:1–6.
10. Edlow BL, Chatelle C, Spencer CA, Chu CJ, Bodien YG, O’Connor KL, et al.
Early detection of consciousness in patients with acute severe traumatic
brain injury. Brain. 2017;140:2399–414.
11. Weijer C, Bruni T, Gofton T, Young GB, Norton L, Peterson A, et al. Ethical
considerations in functional magnetic resonance imaging research
in acutely comatose patients. Brain. 2016;139(Pt 1):292–9. https ://doi.
org/10.1093/brain /awv27 2.
12. Vinciguerra L, Bösel J. Noninvasive neuromonitoring: current utility in
subarachnoid hemorrhage, traumatic brain injury, and stroke. Neurocrit
Care. 2017;27:122–40.
13. Sondag L, Ruijter BJ, Tjepkema-Cloostermans MC, Beishuizen A, Bosch
FH, van Til JA, et al. Early EEG for outcome prediction of postanoxic coma:
Prospective cohort study with cost-minimization analysis. Crit Care.
2017;21(1):111.
14. Hofmeijer J, Beernink TMJ, Bosch FH, Beishuizen A, Tjepkema-Cloost-
ermans MC, Van Putten MJAM. Early EEG contributes to multimodal
outcome prediction of postanoxic coma. Neurology. 2015;85:137–43.
15. Chalak LF, Tian F, Adams-Huet B, Vasil D, Laptook A, Tarumi T, et al.
Novel wavelet real time analysis of neurovascular coupling in neonatal
encephalopathy. Sci Rep. 2017;7:45958.
16. Phillip D, Iversen HK, Schytz HW, Selb J, Boas DA, Ashina M. Altered low
frequency oscillations of cortical vessels in patients with cerebrovascular
occlusive disease—a NIRS study. Front Neurol. 2013;4:204.
17. Schytz HW, Guo S, Jensen LT, Kamar M, Nini A, Gress DR, et al. A new tech-
nology for detecting cerebral blood flow: a comparative study of ultra-
sound tagged NIRS and 133Xe-SPECT. Neurocrit Care. 2012;17:139–45.
18. Obrig H, Neufang M, Wenzel R, Kohl M, Steinbrink J, Einhäupl K, et al.
Spontaneous low frequency oscillations of cerebral hemodynamics and
metabolism in human adults. Neuroimage. 2000;12:623–39.
19. Haubrich C, Klemm A, Diehl RR, Möller-Hartmann W, Klötzsch C. M-wave
analysis and passive tilt in patients with different degrees of carotid artery
disease. Acta Neurol Scand. 2004;109:210–6.
20. Reinhard M, Wehrle-Wieland E, Grabiak D, Roth M, Guschlbauer B, Timmer
J, et al. Oscillatory cerebral hemodynamics-the macro-versus microvascu-
lar level. J Neurol Sci. 2006;250:103–9.
21. Reinhard M, Schumacher FK, Rutsch S, Oeinck M, Timmer J, Mader I, et al.
Spatial mapping of dynamic cerebral autoregulation by multichannel
near-infrared spectroscopy in high-grade carotid artery disease. J Biomed
Opt. 2014;19:097005.
22. Vernieri F, Tibuzzi F, Pasqualetti P, Rosato N, Passarelli F, Rossini PM, et al.
Transcranial doppler and near-infrared spectroscopy can evaluate the
hemodynamic effect of carotid artery occlusion. Stroke. 2004;35:64–70.
23. Schytz HW, Ciftçi K, Akin A, Ashina M, Bolay H. Intact neurovascular
coupling during executive function in migraine without aura: interictal
near-infrared spectroscopy study. Cephalalgia. 2010;30:457–66.
24. Guhathakurta D, Dutta A. Computational pipeline for NIRS–EEG joint
imaging of tDCS-evoked cerebral responses-an application in ischemic
stroke. Front Neurosci. 2016;10:261.
25. Hametner C, Stanarcevic P, Stampfl S, Rohde S, Veltkamp R, Bösel J.
Noninvasive cerebral oximetry during endovascular therapy for acute
ischemic stroke: an observational study. J Cereb Blood Flow Metab.
2015;35:1722–8.
26. Jindal U, Sood M, Chowdhury SR, Das A, Kondziella D, Dutta A. Corti-
cospinal excitability changes to anodal tDCS elucidated with NIRS–EEG
joint-imaging: an ischemic stroke study. In: 2015 37th Annual Interna-
tional Conference IEEE Engineering Medical Biology Society. IEEE; 2015. p.
3399–402. Available from: http://www.ncbi.nlm.nih.gov/pubme d/26737
022.
27. Keller E, Froehlich J, Baumann D, Böcklin C, Sikorski C, Oberle M, et al.
Detection of del ayed cerebral ischemia (DCI) in subarachnoid haemor-
rhage applying near-infrared spectroscopy: elimination of the extracer-
ebral signal by transcutaneous and intraparenchymatous measurements
in parallel. Acta Neurochir Suppl. 2015;120:243–7.
28. Zweifel C, Castellani G, Czosnyka M, Helmy A, Manktelow A, Carrera
E, et al. Noninvasive monitoring of cerebrovascular reactivity with
near infrared spectroscopy in head-injured patients. J Neurotrauma.
2010;27:1951–8.
29. Storm C, Leithner C, Krannich A, Wutzler A, Ploner CJ, Trenkmann L, et al.
Regional cerebral oxygen saturation after cardiac arrest in 60 patients—a
prospective outcome study. Resuscitation. 2014;85:1037–41.
30. Pichler G, Avian A, Binder C, Zotter H, Schmölzer GM, Morris N, et al. EEG
and NIRS during transition and resuscitation after birth: Promising addi-
tional tools; an observational study. Resuscitation. 2013;84:974–8.
31. Andersen AV, Simonsen SA, Schytz HW, Iversen HK. Assessing low-
frequency oscillations in cerebrovascular diseases and related conditions
with near-infrared spectroscopy: a plausible method for evaluating
cerebral autoregulation? Neurophotonics. 2018;5:1.
32. Ferrari M, Quaresima V. A brief review on the history of human functional
near-infrared spectroscopy (fNIRS) development and fields of application.
Neuroimage. 2012;63(2):921–35.
33. Mantini D, Perrucci MG, Del Gratta C, Romani GL, Corbetta M. Electro-
physiological signatures of resting state networks in the human brain.
Proc Natl Acad Sci USA. 2007;104:13170–5.
34. Pfurtscheller G, Schwerdtfeger A, Brunner C, Aigner C, Fink D, Brito J, et al.
Distinction between neural and vascular BOLD oscillations and inter-
twined heart rate oscillations at 0.1 Hz in the resting state and during
movement. PLoS One. 2017;12:0168097.
35. Pfurtscheller G, Schwerdtfeger AR, Seither-Preisler A, Brunner C, Stefan
Aigner C, Brito J, et al. Brain–heart communication: evidence for “central
pacemaker” oscillations with a dominant frequency at 0.1 Hz in the
cingulum. Clin Neurophysiol. 2017;128:183–93.
36. Rayshubskiy A, Wojtasiewicz TJ, Mikell CB, Bouchard MB, Timerman D,
Youngerman BE, et al. Direct, intraoperative observation of ~ 0.1 Hz
hemodynamic oscillations in awake human cortex: implications for fMRI.
Neuroimage. 2014;87:323–31.
37. Diehl RR, Linden D, Lücke D, Berlit P. Phase relationship between cerebral
blood flow velocity and blood pressure. A clinical test of autoregulation.
Stroke. 1995;26:1801–4.
38. Pfurtscheller G, Daly I, Bauernfeind G, Müller-Putz GR. Coupling between
intrinsic prefrontal HbO2 and central EEG beta power oscillations in the
resting brain. PLoS ONE. 2012;7:e43640.
39. Foster BL, Parvizi J. Resting oscillations and cross-frequency coupling in
the human posteromedial cortex. Neuroimage. 2012;60:384–91.
40. Sood M, Besson P, Muthalib M, Jindal U, Perrey S, Dutta A, et al. NIRS–EEG
joint imaging during transcranial direct current stimulation: Online
parameter estimation with an autoregressive model. J Neurosci Methods.
2016;274:71–80.
41. Coyle SM, Ward TE, Markham CM. Brain-computer interface using a
simplified functional near-infrared spectroscopy system. J Neural Eng.
2007;4:219–26.
42. Naseer N, Hong K-S. fNIRS-based brain-computer interfaces: a review.
Front Hum Neurosci. 2015;9:3.
43. Holper L, Wolf M. Single-trial classification of motor imagery differing in
task complexity: a functional near-infrared spectroscopy study. J Neuro-
eng Rehabil. 2011;8:34.
44. Koo B, Lee H-G, Nam Y, Kang H, Koh CS, Shin H-C, et al. A hybrid NIRS–
EEG system for self-paced brain computer interface with online motor
imagery. J Neurosci Methods. 2015;244:26–32.
45. Fu Y, Xiong X, Jiang C, Xu B, Li Y, Li H. Imagined hand clenching force and
speed modulate brain activity and are classified by NIRS combined with
EEG. IEEE Trans Neural Syst Rehabil Eng. 2017;25:1641–52.
46. Bai Y, Xia X, Kang J, Yang Y, He J, Li X. TDCS modulates cortical excit-
ability in patients with disorders of consciousness. NeuroImage Clin.
2017;15:702–9.
47. Plomgaard AM, van Oeveren W, Petersen TH, Alderliesten T, Austin T, van
Bel F, et al. The safeboosC II randomized trial: treatment guided by near-
infrared spectroscopy reduces cerebral hypoxia without changing early
biomarkers of brain injury. Pediatr Res. 2016;79:528–35.
48. Ancora G, Maranella E, Locatelli C, Pierantoni L, Faldella G. Changes in
cerebral hemodynamics and amplitude integrated EEG in an asphyxiated
newborn during and after cool cap treatment. Brain Dev. 2009;31:442–4.
49. Ancora G, Maranella E, Grandi S, Sbravati F, Coccolini E, Savini S, et al. Early
predictors of short term neurodevelopmental outcome in asphyxiated
cooled infants A combined brain amplitude integrated electroencepha-
lography and near infrared spectroscopy study. Brain Dev. 2013;35:26–31.
50. Goeral K, Urlesberger B, Giordano V, Kasprian G, Wagner M, Schmidt
L, et al. Prediction of outcome in neonates with hypoxic-ischemic
encephalopathy II: role of amplitude-integrated electroencephalography
and cerebral oxygen saturation measured by near-infrared spectroscopy
Neonatology. 2017;112:193–202.
51. Palmer JA, Makeig S, Kreutz-Delgado K, Rao BD. Newton method for the
ICA mixture model. In: IEEE international conference acoustics speech
signal process. 2008. pp. 1805–8.
52. Hsu S-H, Pion-Tonachini L, Palmer J, Miyakoshi M, Makeig S, Jung T-P.
Modeling brain dynamic state changes with adaptive mixture independ-
ent component analysis. Neuroimage. 2018;183:47–61.
53. Palmer JA, Kreutz-Delgado K, Makeig S. AMICA : an adaptive mixture of
independent component analyzers with shared components. 2011.
54. Wu CW, Tsai P, Chen SC, Li C, Hsu A, Wu H, et al. Indication of dynamic
neurovascular coupling from inconsistency between EEG and fMRI
indices across sleep–wake states. Sleep Biol Rhythms. 2019;17:423–31.
55. Wijdicks EFM, Bamlet WR, Maramattom BV, Manno EM, McClelland
RL. Validation of a new coma scale: The FOUR score. Ann Neurol.
2005;58:585–93.
56. Seel RT, Sherer M, Whyte J, Katz DI, Giacino JT, Rosenbaum AM, et al.
Assessment scales for disorders of consciousness: Evidence-based rec-
ommendations for clinical practice and research. Arch Phys Med Rehabil.
2010;91(2):1795–813.
57. Giacino JT, Katz DI, Schiff ND, Whyte J, Ashman EJ, Ashwal S, et al. Com-
prehensive systematic review update summary: disorders of conscious-
ness. Neurology. 2018. https ://doi.org/10.1212/wnl.00000 00000 00592 8.
58. Trojano L, Moretta P, Masotta O, Loreto V, Estraneo A. Visual pursuit of
one’s own face in disorders of consciousness: a quantitative analysis. Brain
Inj. 2018;32(2):1549–999.
59. Posner J, Plum F, Saper C. Plum and posner’s diagnosis of stupor and
coma. New York: Oxford University Press; 2007.
60. Laureys S, Celesia GG, Cohadon F, Lavrijsen J, León-Carrión J, Sannita WG,
et al. Unresponsive wakefulness syndrome: a new name for the vegeta-
tive state or apallic syndrome. BMC Med. 2010;8:68.
61. Giacino JT, Ashwal S, Childs N, Cranford R, Jennett B, Katz DI, et al. The
minimally conscious state: definition and diagnostic criteria. Neurology.
2002;58:349–53.
62. Bruno M-A, Vanhaudenhuyse A, Thibaut A, Moonen G, Laureys S. From
unresponsive wakefulness to minimally conscious PLUS and functional
locked-in syndromes: recent advances in our understanding of disorders
of consciousness. J Neurol. 2011;258:1373–84.
63. Products/STARSTIM - Neuroelectrics.
64. Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of
single-trial EEG dynamics including independent component analysis. J
Neurosci Methods. 2004;134:9–21.
65. Lehembre R, Marie-Aurélie B, Vanhaudenhuyse A, Chatelle C, Cologan V,
Leclercq Y, et al. Resting-state EEG study of comatose patients: a connec-
tivity and frequency analysis to find differences between vegetative and
minimally conscious states. Funct Neurol. 2012;27:41–7.
66. Tachtsidis I, Scholkmann F. False positives and false negatives in
functional near-infrared spectroscopy: issues, challenges, and the way
forward. Neurophotonics. 2016;3(3):031405.
67. Scholkmann F, Kleiser S, Metz AJ, Zimmermann R, Mata Pavia J, Wolf U,
et al. A review on continuous wave functional near-infrared spectroscopy
and imaging instrumentation and methodology. Neuroimage. 2014;85(Pt
1):6–27.
68. Goodwin JR, Gaudet CR, Berger AJ. Short-channel functional near-infra-
red spectroscopy regressions improve when source-detector separation
is reduced. Neurophotonics. 2014;1(1):015002.
69. Gregg NM, White BR, Zeff BW, Berger AJ, Culver JP. Brain specificity of
diffuse optical imaging: improvements from superficial signal regression
and tomography. Front Neuroenergetics. 2010;2:14.
70. Huppert TJ, Diamond SG, Franceschini MA, Boas DA. HomER: a review of
time-series analysis methods for near-infrared spectroscopy of the brain.
Appl Opt. 2009;48:D280–98.
71. Bigot J, Longcamp M, Dal Maso F, Amarantini D. A new statistical test
based on the wavelet cross-spectrum to detect time-frequency depend-
ence between non-stationary signals: application to the analysis of
cortico-muscular interactions. Neuroimage. 2011;55:1504–18.
72. Encyclopedia of Bioinformatics and Computational
Biology|ScienceDirect.
73. Pollonini L, Bortfeld H, Oghalai JS. PHOEBE: a method for real time map-
ping of optodes-scalp coupling in functional near-infrared spectroscopy.
Biomed Opt Exp. 2016;7:5104–19.
74. Carnevale T, Majumdar A, Sivagnanam S, Yoshimoto K, Astakhov V,
Bandrowski A, et al. The neuroscience gateway portal: high performance
computing made easy. BMC Neurosci. 2014;15:P101.
75. Turgeon AF, Lauzier F, Simard J-F, Scales DC, Burns KEA, Moore L, et al.
Mortality associated with withdrawal of life-sustaining therapy for
patients with severe traumatic brain injury: a Canadian multicentre
cohort study. CMAJ. 2011;183:1581–8.
76. Chung DY, Claassen J, Agarwal S, Schmidt JM, Mayer SA. Assessment of
noninvasive regional brain oximetry in posterior reversible encepha-
lopathy syndrome and reversible cerebral vasoconstriction syndrome. J
Intensive Care Med. 2016;31:415–9.
77. Kondziella D, Friberg CK, Wellwood I, Reiffurth C, Fabricius M, Dreier JP.
Continuous EEG monitoring in aneurysmal subarachnoid hemorrhage: a
systematic review. Neurocrit Care. 2015;22:450–61.
78. Zweifel C, Dias C, Smielewski P, Czosnyka M. Continuous time-domain
monitoring of cerebral autoregulation in neurocritical care. Med Eng
Phys. 2014;36:638–45.
79. Franceschini MA, Radhakrishnan H, Thakur K, Wu W, Ruvinskaya S, Carp
S, et al. The effect of different anesthetics on neurovascular coupling.
Neuroimage. 2010;51:1367–77.
80. Carpenter KLH, Young AMH, Hutchinson PJ. Advanced monitoring in
traumatic brain injury: microdialysis. Curr Opin Crit Care. 2017;23:103–9.
81. Oddo M, Hutchinson PJ. Understanding and monitoring brain injury: the
role of cerebral microdialysis. Intensive Care Med. 2018;44:1945–8.
82. Huschak G, Hoell T, Hohaus C, Kern C, Minkus Y, Meisel HJ. Clinical evalu-
ation of a new multiparameter neuromonitoring device: measurement
of brain tissue oxygen, brain temperature, and intracranial pressure. J
Neurosurg Anesthesiol. 2009;21:155–60.
83. Sinha S, Hudgins E, Schuster J, Balu R. Unraveling the complexities of
invasive multimodality neuromonitoring. Neurosurg Focus. 2017;43:E4.
84. Elmer J, Flickinger KL, Anderson MW, Koller AC, Sundermann ML,
Dezfulian C, et al. Effect of neuromonitor-guided titrated care on brain
tissue hypoxia after opioid overdose cardiac arrest. Resuscitation.
2018;129:121–6.
85. Lazaridis C, Robertson CS. The Role of Multimodal Invasive Monitoring in
Acute Traumatic Brain Injury. Neurosurg. Clin. N. Am. W.B. Saunders; 2016.
27(4): 509–17.
86. Kempny AM, James L, Yelden K, Duport S, Farmer S, Playford ED, et al.
Functional near infrared spectroscopy as a probe of brain function in
people with prolonged disorders of consciousness. NeuroImage Clin.
2016;12:312–9.
87. Banville H, Gupta R, Falk TH. Mental task evaluation for hybrid NIRS–EEG
brain–computer interfaces. Comput Intell Neurosci. 2017;2017:3524208.
https ://doi.org/10.1155/2017/35242 08.
88. Shin J, Kwon J, Im C-H. A ternary hybrid EEG–NIRS brain–computer
interface for the classification of brain activation patterns during mental
arithmetic, motor imagery, and idle state. Front Neuroinform. 2018;12:5.
89. Iso N, Moriuchi T, Sagari A, Kitajima E, Iso F, Tanaka K, et al. Monitoring
local regional hemodynamic signal changes during motor execution
and motor imagery using near-infrared spectroscopy. Front Physiol.
2016;6:416.
90. Abdalmalak A, Milej D, Diop M, Shokouhi M, Naci L, Owen AM, et al. Can
time-resolved NIRS provide the sensitivity to detect brain activity during
motor imagery consistently? Biomed Opt Exp. 2017;8:2162.
91. Giacino JT, Kalmar K, Whyte J. The JFK Coma Recovery Scale-Revised:
measurement characteristics and diagnostic utility. Arch Phys Med Reha-
bil. 2004;85:2020–9.
92. Almojuela A, Hasen M, Zeiler FA. The full outline of unresponsiveness
(FOUR) score and its use in outcome prediction: a scoping systematic
review of the adult literature. Neurocrit Care. 2019;31:162–75.
93. Foo CC, Loan JJM, Brennan PM. The relationship of the Four
score to patient outcome: a systematic review. J Neurotrauma.
2019;36(17):2469–83.
