1

Title: Quantifying cerebral blood flow in an adult pig

ischemia model by a depth-resolved dynamic

contrast-enhanced optical method

Authors: Jonathan T. Elliott, Ph. D. 1,2

; Mamadou Diop, Ph. D. 1,2

; Laura B.

Morrison, B. Sc. 2; Christopher D. d'Esterre, Ph. D.

1,2; Ting-Yim

Lee, Ph. D. 1,2

; Keith St. Lawrence, Ph. D. 1,2

Affiliations: 1. Department of Medical Biophysics, Western University,

London, Ontario, Canada, N6A 5C1;

2. Imaging Division, Lawson Health Research Institute, London,

Ontario, Canada, N6A 4V2;

Present Address: Jonathan T. Elliott

Thayer School of Engineering

Dartmouth College

14 Engineering Drive, Hanover, NH 03755-8000

Tel: 603 646-0775

email: jonathan.t.elliott@dartmouth.edu

Sources of Support: Ontario Neurotrauma Foundation, Canadian Institutes of Health

Research, Ontario Graduate Scholarship

Running Header: Optical measurement of cerebral hemodynamics

2

Abstract

Dynamic contrast-enhanced (DCE) near-infrared (NIR) methods have been proposed for

bedside monitoring of cerebral blood flow (CBF). These methods have primarily focused on

qualitative approaches since scalp contamination hinders quantification. In this study, we

demonstrate that accurate CBF measurements can be obtained by analyzing multi-distance time-

resolved DCE data with a combined kinetic deconvolution / optical reconstruction (KDOR)

method. Multi-distance time-resolved DCE-NIR measurements were made in adult pigs (n = 8)

during normocapnia, hypocapnia and ischemia. The KDOR method was used to calculate CBF

from the DCE-NIR measurements. For validation, CBF was measured independently by CT

under each condition. The mean CBF difference between the techniques was −1.7 mL/100 g/min

with 95% confidence intervals of −16.3 and 12.9 mL/100 g/min; group regression analysis

revealed a strong agreement between the two techniques (slope = 1.06 ± 0.08, y-intercept =

−4.37 ± 4.33 mL/100 g/min, p < 0.001). The results of an error analysis suggest that little a

priori information is needed to recover CBF, due to the robustness of the analytical method and

the ability of time-resolved NIR to directly characterize the optical properties of the extracerebral

tissue (where model mismatch is deleterious). The findings of this study suggest that the DCE-

NIR approach presented is a minimally invasive and portable means of determining absolute

hemodynamics in neurocritical care patients.

Keywords:

Cerebral blood flow, cerebral hemodynamics, neurocritical care, kinetic modelling, near-infrared

spectroscopy

3

Highlights

Cerebral blood flow is measured using a novel dynamic contrast-enhanced near-infrared

method

The method is quantitative and absolute, since it corrects for scalp contamination

An adult pig model of ischemia was used for validation, and is described

A strong correlation between the proposed method and CT perfusion is observed

The technique can be used when anatomy and tissue optical properties are unknown

1. Introduction

Following neurotrauma—including ischemic stroke, severe traumatic brain injury, and

subarachnoid hemorrhage—a principle concern of the neurointensive care unit is avoiding

complications that can cause cerebral ischemia, ultimately leading to secondary brain injury.

These complications, which include increased intracranial pressure (ICP), hemorrhage and

vasospasms, are often not clinically evident until permanent damage to the brain has occurred

(Vergouwen et al., 2010). In addition, conventional imaging modalities that are capable of

detecting ischemia, such as computed tomography (CT) and magnetic resonance imaging (MRI),

are impractical for longitudinal measurements and patients are often too unstable to be

transferred to imaging suites (Waydhas, 1999). Instead, a noninvasive bedside technique for

measuring cerebral blood flow (CBF) is desirable, as the ability to monitor CBF could improve

patient outcome by alerting hospital staff to an ischemic event before brain damage occurs. A

variety of near-infrared (NIR) techniques have been proposed for this purpose because tissue is

relatively transparent to near-infrared light. One approach is to use indocyanine green (ICG), an

FDA-approved optical contrast agent, in combination with serial NIR measurements to quantify

cerebral hemodynamics (Brown et al., 2002)—analogous to dynamic contrast-enhanced (DCE)

4

MRI and CT methods. Despite successful use in neonatal applications (Arora et al., 2013; Brown

et al., 2002; Tichauer et al., 2006), the application of DCE NIRS to the adult head is impeded by

signal contamination from extracerebral tissue (i.e., scalp, skull and CSF), resulting in substantial

underestimations of CBF (Gora et al., 2002; Keller et al., 2003).

Extracerebral layer (ECL) contamination is a difficult problem to overcome for three

reasons: First, the scalp dye concentration is highly correlated with the brain concentration,

which hinders the application of principle component or cross-correlation analysis commonly

used with functional NIRS (Franceschini et al., 2006). Second, due to the fall-off of light fluence

with the square of distance, extracerebral components will account for the majority of the

detected signal (Okada et al., 1997). Finally, and most significantly, the highly scattering nature

of light in tissue makes it impossible to determine the exact path taken by a given photon. As a

result, spatial reconstruction of the detected signal is fundamentally an ill-posed problem

dominated by model mismatch, discretization errors, stochastic noise, and other systemic errors

(Bonfert-Taylor et al., 2012). Notwithstanding these challenges, significant advances have been

made regarding the measurement of cerebral hemodynamics by DCE NIR techniques. It has been

demonstrated that the statistical moments of photon time-of-flight distributions show different

wash-in and wash-out dynamics when measured in stroke patients (Liebert et al., 2005). In

another study, relative differences in tissue dynamics were observed between normal and stroke-

affected regions of the brain (Steinkellner et al., 2011). These studies illustrate the fundamental

sensitivity of these NIR techniques to changes occurring in the brain, but they did not recover

quantitative estimates of CBF. Given that DCE-NIR measurements acquired on the surface of the

head are indeed sensitive to cerebral changes in blood flow, the challenge remains to develop a

clinical relevant technique for quantifying CBF.

5

In this paper, we present experimental validation of a novel approach for reconstruction

DCE NIRS data, referred to kinetic deconvolution optical reconstruction (KDOR) (Elliott et al.,

2012), that overcomes the problems caused by extracerebral signal contamination. The KDOR

method improves the optical reconstruction of light absorption in different tissue layers by

incorporating “dynamic priors” (i.e., mathematical constraints based on the time-dependent

behavior of a contrast agent predicted by tracer kinetic theory). For validation, CBF and cerebral

blood volume (CBV) values obtained by KDOR were compared to values obtained by CT

perfusion (Lee, 2002). To replicate the clinical scenario, experiments were conducted in pigs

since the thickness of the ECL closely approximates the adult human head. Validation was

conducted under three hemodynamic states: normocapnia (baseline), hypocapnia, and ischemia.

Concomitant CT perfusion data were acquired under each state. Regression and Bland-Altman

analysis were used to compare the CT and DCE-NIR parameters and determine the degree of

difference between the two techniques.

2. Materials and Methods

2.1 Animal protocol

Animal experiments were conducted according to the guidelines of the Canadian Council

on Animal Care (CCAC) and approved by the Animal Use Committee at Western University

(AUP #2007-050-06). Eight Duroc x Landrace crossbred pigs (Sus scrofa domesticus, weight =

15.4 ± 0.74 kg, 6 females) were obtained from a local supplier on the day of the experiment.

Following anesthetic induction with 1.75-3% isoflurane, the animals were tracheotomized and

mechanically ventilated on a mixture of oxygen and medical air. A polychloroprene probe holder

was constructed in-house and fixed on the head of the animal with tissue glue (Vetbond®

(n-

6

butyl cyanoacrylate), 3M, St. Paul, MN). Because of the high blood flow in temporalis muscle,

three incisions were made (caudial, rostral, and lateral to the probe holder) to reduce scalp blood

flow from ~25 ml·min-1

·100g-1

(Elliott et al., 2013a) to about 10 ml·min-1

·100g-1

, which is

similar to human scalp flow (Friberg et al., 1986). Following the preparation and surgical

procedures, the animals were moved to the CT imaging suite and allowed to stabilize for 1 hour

before the experiment was continued. At the end of the experiment, animals were euthanized

according to guidelines set forth by the CCAC.

2.2 Study Design

The study consisted of collecting CT and DCE-NIR measurements at three different

levels of CBF. Following stabilization, physiological parameters were recorded and a DCE-NIR

measurement was acquired over a period of about 5 minutes. The optical probes were removed

from the probe holder and a CT perfusion scan was performed. Following the CT scan, the

probes were again placed in the probe holder and a second DCE-NIR measurement was

acquired. A period of at least 15 minutes was allowed between the start of DCE-NIR

measurements to allow for clearance of the dye, and CT data was acquired 1-2 minutes after the

first DCE-NIR measurement, so that all three measurements were made within approximately 20

minutes. This process was repeated for the hypocapnia and ischemia conditions, resulting in a

total of six optical datasets and three CT perfusion datasets for each animal. For regression

analysis, an average value representing the pair of DCE-NIR measurements was compared with

the single CT perfusion measurement. A total of five animals were subjected to all three

conditions and an additional three animals were subjected to only baseline and hypocapnia

conditions. For two of these animals, the experiments were conducted before a protocol

7

modification was approved for the ischemia procedure, and endothelin-1 was unavailable for the

remaining animal.

2.3 Modifying cerebral blood flow

Hypocapnia was achieved by adjusting the ventilation rate on the mechanical ventilator to

achieve a PCO2 of ~25 mmHg. Blood gas levels were confirmed using a hemoximeter (ABL 80

FLEX CO-OX, Radiometer, Copenhagen, Denmark). Once the appropriate PCO2 level was

achieved, the experiment procedure was delayed by approximately 15 min to ensure that the

physiological parameters were stable. Ischemia was achieved by intra-cortical injection of

endothelin-1 (Sigma-Aldrich, St. Louis, MO), a potent vasoconstrictor. A burr hole was drilled

through the scalp and skull, just lateral to the probe holder, half way between the source optode

and the detector optode 40 mm away. A 30-gauge needle was carefully inserted into the cortical

tissue, angled towards midline, and a low-dose CT scan was performed to confirm the location of

the needle tip. The needle was connected to a catheter, which in turn was connected to a syringe

containing 1 μg/kg of endothelin-1 in 0.1 ml of sterile water. Using a syringe pump, the contents

of the syringe were infused into the cortical tissue over 10 minutes. Data collection proceeded 10

minutes after the endothelin-1 injection was finished.

The reproducibility of endothelin-induced CBF changes was reported as the standard

deviation of mean CBF across the DCE NIR region of interrogation (determined from light

propagation modeling). The reproducibility of infarct size was determined by tracing the infarct

perimeter free-hand and computing the corresponding area using image analysis software

(ImageJ 1.46, National Institutes of Health, Bathsheba, MD). The area was determined on each

8

of the eight CBF maps, and the ischemic volume was calculated by multiplying the sum of the

areas from each stack by the slice thickness (5 mm).

2.4 Physiological measurements

In addition to the CT and DCE-NIR measurements, physiological parameters were

monitored throughout the experiment, including arterial oxygen saturation, heart rate, respiration

rate, end-tidal CO2, mean arterial pressure, and rectal temperature. In addition, a hemoximeter

was used to measure arterial blood pH, PCO2, PO2, and glucose. The main purpose of collecting

these measurements was to ensure physiological stability in each of the three flow conditions.

The stability of the blood gases was achieved by adjusting the respiration rate on the mechanical

ventilator. Temperature and blood glucose levels were maintained by a heated blanket and

intermittent dextrose administrations, respectively.

2.5 Instrumentation

Optical measurements were collected using a time-resolved NIR system that has been

described previously (Diop et al., 2010). The light source was a picosecond diode laser emitting

at 802 nm with a repetition rate of 80 MHz. A 1.5-m long multimode fiber was used to guide the

light from the source to a single position on the animal’s head. Photons exiting the scalp were

collected by four fibre bundles configured linearly at distances of 6, 20, 30, and 40 mm from the

source. Light from each detection optode was guided to one of four photomultiplier tubes (PMC-

100, Becker & Hickl Gmbh, Berlin, Germany) after having passed through a bandpass filter to

remove fluorescence (FEB800-10, Thorlabs, NJ). The instrument response function (IRF) was

measured at the start of each experiment, by placing a thin piece of paper between the emission

9

and detection fiber (Diop et al., 2010). To perform DCE-NIR measurements, time-of-flight

histograms were collected every 0.4 seconds for a total of 89.6 s during the passage of

indocyanine green (ICG). Baseline data of 12.8 s were collected before the bolus injection.

Distributions of time-of-flight (DTOF) were measured at all four channels simultaneously. In

addition, the arterial input function (AIF) was acquired using a pulse dye densitometer

(DDM2000, Nihon-Koden, Japan).

Computed tomography anatomical images and CT perfusion maps were acquired using

the LightSpeed QXi scanner (GE Healthcare, Waukesha, WI). Anatomical images were acquired

in helical mode before and during contrast enhancement (slice thickness = 1.25 mm, current =

200 mA, energy = 140 kVp, field of view (FOV) = 140 x 140 x 155 mm). Dynamic CT data

were acquired during the bolus injection of an iodine-based contrast agent (1.0 ml/kg of

iopamidol [300-Isovue®], Bracco S.p.A., Milan, Italy) at a rate of 1 ml/s (slice thickness = 5

mm, current = 200 mA, energy = 140 kVp, FOV = 140 x 140 x 40 mm)

2.6 Data analysis

The DTOFs measured at the four source-detector distances were first denoised using a

previously described Anscombe transformation and wavelet method (Diop and St Lawrence,

2012), and then converted to the three statistical moments of attenuation, mean time-of-flight and

variance (Liebert et al., 2004). Instrument response functions were subtracted from each time

series, and a smoothing Savitzky-Golay filter (using a window size of 12 seconds and a 7th

order

polynomial) was applied to the data prior to further analysis. The KDOR method, described in

previous publications (Elliott et al., 2012; Elliott et al., 2013a) and expanded in the Appendix,

was used to analyze the DCE-NIR moments. The entire procedure is outlined in Figure 1. First,

10

CT anatomical images were manually segmented, slice-by-slice, into scalp, skull and brain (seg

in Figure 1). Second, the baseline optical properties (i.e., the absorption, μa, and reduced scatter

coefficient, μs´) of the ECL were determined from the time-resolved data acquired at a source-

detector distance of 6 mm (fit in Figure 1) (Diop et al., 2010). The ECL fitting step involved

optimizing the difference norm of time-resolved data and the solution to the diffusion

approximation for a homogenous semi-infinite medium (Kienle and Patterson, 1997). It was

assumed that at this short separation, the measured signal would be almost entirely comprised of

extracerebral components (Okada and Delpy, 2003). Next, the Jacobian, A, was calculated for

each source-detector distance using Monte Carlo eXtreme (Fang and Boas, 2009). This step

incorporated the anatomical information from the CT images and the optical properties of the

ECL (obtained using fit) and brain layers (assuming μs′ = 1.11 mm-1

and μa = 0.011 mm-1

)

(Elliott et al., 2010), where the ECL was represented as a single layer with bulk optical

properties obtained from the fit step. For the final step, the Jacobian, the DCE-NIR data and the

AIF were input into the KDOR algorithm to recover BF and BV for the ECL and brain regions.

11

Figure 1 Flow chart outlining the steps involved in recovering the hemodynamic parameters. The parallelograms

indicate the measured data from the three devices. Rectangles represent the analytical processes used: fit, diffusion

approximation (DA) fitting routine; seg, manual or automatic segmentation; and MC, Monte Carlo modeling of the

Jacobian. Intermediary parameters are represented by ovals: μ, region-specific bulk optical properties, B, binary

segmented CT volume; and A, Jacobian. Finally, the recovered hemodynamic parameters (BF, blood flow; BV,

blood volume; and MTT, mean transit time) are indicated by triangles.

For comparison to the NIRS CBF and CBV measurements, region-of-interest analysis

was performed on CT perfusion parameteric maps to determine the mean values of CBF and

CBV within the volume of tissue interrogated by the DCE-NIR light (Lee, 2002). Parametric

12

maps were computed using the clinical CT workstation (PERFUSION 4, GE Healthcare), by

deconvolution of tissue enhancement curves by the AIF in 2 x 2 pixel blocks. Partial volume

averaging of the AIF was corrected using the venous time density curve and ROI analysis was

performed using in-house developed software using a thresholding procedure to remove tissue

signal contributions from large vessels (Murphy et al., 2006), since the detected optical signal is

assumed to contain very little information from large vessels because of high absorption (Liu et

al., 1995). The region-of-interrogation was estimated from Monte Carlo generated sensitivity

maps (Fang and Boas, 2009). In addition to DCE-NIR validation, CT perfusion maps were also

used to confirm the magnitude and volume of the ischemic lesion following endothelin-1

injection.

2.7 Error analysis

An error analysis was conducted to investigate two of the principle sources of uncertainty

in the analytical framework. First, the anatomical information used in the KDOR procedure

could be influenced by errors in co-registering the position of the optical probe with the CT

images, which would result in incorrect estimates of the true ECL thickness. This error was

investigated by generating DCE-NIR forward data on a two-layer slab (ECL and brain)

according to a previously described method (Elliott et al., 2012). Following the addition of 5%

Gaussian noise to the NIR and AIF signals, CBF was recovered with KDOR using a Jacobain

modeled with four-layer slabs of varying thicknesses (10 to 20 mm).

A second potential source of error is incorrect values of the optical properties used to

define the Jacobian function. Of particular concern is the scattering coefficients for ECL and

brain since they dominate the predicted spatial distribution of light (Elliott et al., 2010). The

13

magnitude of this error was assessed by generating forward optical data with known values of μa

(0.015 and 0.012 mm-1

for ECL and brain, respectively) and μs´ (0.8 and 1.1 mm-1

for ECL and

brain, respectively). Next, CBF was estimated from the same data using the KDOR algorithm,

but with the ECL μs´ value ranging from 0.4 to 1.2 mm-1

. The procedure was then repeated with

the brain μs´ value ranging from 0.6 to 1.6 mm-1

, and finally, the ECL μa value ranging from

0.008 to 0.022 mm-1

.

2.8 Statistical analysis

All data are presented as mean ± SEM unless otherwise indicated. An analysis of

variance (ANOVA) was used to uncover any conditional effects (i.e. flow state) on the

hemodynamic parameters, which were further investigated with a Tukey's post hoc test when

appropriate.

Linear regression analysis was performed on the CBF and CBV measurements to

examine correlations between the values obtained from the two techniques (DCE-NIR and CT

perfusion). Since datasets included multiple measurements from the same animal, independence

could not be assumed and therefore a variation of the generalized estimating equation was

utilized. First, a linear fit was applied to the data from each animal separately. A t-test was used

to compare the slope and intercept values obtained from each animal to a null hypothesis (i.e., a

slope of zero). Upon rejection of the null hypothesis, the distribution of slopes was compared

with a slope of unity to determine if the mean regression line was significantly different from the

line of identity.

For the CBF regression analysis, the 95% confidence bounds on the regression line-of-

best fit (i.e., the region containing the 95% of the distribution of lines-of-best fit assuming a

14

Gaussian random effect) were calculated. Finally, the degree of similarity between CBF

measurements acquired with the two techniques was evaluated using a Bland-Altman plot.

Differences with p < 0.05 were considered significant.

3. Results

3.1 Physiological parameters

Table 1 summarizes the physiological parameters of the eight pigs averaged for each flow

condition. Respiration rate, which was increased by adjusting the mechanical ventilator setting,

was significantly higher during hypocapnia and ischemia conditions, resulting in the intended

decrease in pCO2. No significant differences were observed in SaO2, HR, MAP, pO2, glucose

and temperature between the three flow conditions. The mean ECL thickness as measured from

the CT images was 12.0 ± 0.6 mm (ranging from 10.5 to 14 mm).

Table 1 Physiological parameters at each flow condition

SaO2

(%)

HR

(min-1

)

RR

(min-

1)

MAP

(mmHg)

pCO2

(torr)

pO2

(torr)

glucose

(mM)

temp

(°C)

Baseline 100 ±

0 108 ± 2 29 ± 1 49± 2

37.3 ±

0.7

168.6 ±

29.4

5.3 ±

1.0

37.3 ±

0.3

Hypocapnia 99 ± 1 109 ± 4 46 ±

1* 46 ± 2

24.2 ±

1.2*

149.4 ±

10

5.4 ±

1.2

37.4 ±

0.2

Ischemia 99 ± 0 124 ±

17

41 ±

2* 47 ± 4

26.0 ±

2.4*

138.4 ±

15.2

5.2 ±

0.8

37.3 ±

0.3 SaO2, arterial oxygen saturation; HR, heart rate; RR, respiration rate; MAP, mean arterial pressure; pCO2, partial

pressure of carbon dioxide; pO2, partial pressure of oxygen. * p < 0.01 compared to baseline

15

3.2 Effect of hypocapnia and endothelin-1 injection on cerebral blood flow

Analysis of variance with flow condition (i.e., baseline, hypocapnia, or ischemia) as a

within-subject factor and modality (i.e., CT perfusion or DCE-NIR) as the between-subject

factor revealed a significant condition effect on CBF measurements acquired with the two

modalities (F2,18 = 10.27, p < 0.01), but no significant condition-by-modality interaction.

Compared with baseline, hypocapnia resulted in a significant reduction in CBF measured with

CT perfusion (43.0 ± 4.2 mL/100 g/min during hypocapnia vs. 61.7 ± 3.1 mL/100 g/min at

baseline; p < 0.05) and with DCE-NIR (42.6 ± 3.7 mL/100 g/min during hypocapnia vs. 62.2 ±

4.7 mL/100 g/min at baseline; p < 0.05). A further reduction in blood flow was observed

following intracortical injection of endothelin-1, resulting in an average CBF value of 39.9 ± 3.1

and 32.7 ± 3.6 mL/100 g/min for CT perfusion and DCE-NIR, respectively.

16

Figure 2 (A) CT perfusion blood flow map with visible large lesion caused by endothelin-1 injection. (B) Contour

plot showing relative regional sensitivity of optical signal. (C) CT perfusion map showing regions identified as

infarct (I; red), penumbra (P; orange), benign oligemia (BO; green) based on thresholding. (D) The relative

contribution of these tissue states to the overall optical region-of-interrogation. I, < 10 mL/100 g/min; P, 10-20

mL/100 g/min; BO, 20-35 mL/100 g/min; N, normal.

A closer examination of the CT perfusion data acquired during ischemia identified the

presence of four tissue states based on CBF thresholds: infarct (< 10 mL/100 g/min), penumbra

(10-20 mL/100 g/min), benign oligemia (20-40 mL/100 g/min), and normal tissue (> 40 mL/100

g/min). Figure 2A-C shows a representative example of a CT perfusion map acquired during

ischemia, the corresponding Monte Carlo sensitivity map, and a threshold map of the CBF

regions. The mean volume of the ischemic lesion, as defined by the sum of the infarct, penumbra

and oligemia, was 27.6 ± 4.6 ml. On average, the three abnormal tissue states represented 49% of

the optical ROI. In Fig. 2D, a bar graph shows the mean contribution of each of the four tissue

17

states to the ROI. The mean CBF decrease measured by CT perfusion was 36.5 ± 3.25 % in the

region interrogated by the NIR light.

3.3 Dynamic contrast-enhanced near-infrared measurements

Figure 3 shows representative examples of the change in attenuation , mean time-of-flight

and variance measured at two source-detector distances (6 and 40 mm) following the injection of

ICG. Dynamic data sets are shown for the three flow conditions. The tissue concentration curves

recovered from these data for the scalp and brain regions are also shown, along with the

corresponding arterial input functions. The CBF estimates determined using the KDOR method

were 45 mL/100 g/min at baseline, 32.5 mL/100 g/min during hypocapnia, and 22.6 mL/100

g/min during ischemia.

18

Figure 3 The attenuation (1st row), mean time-of-flight, (2nd row), and variance (3rd row) signals measured at 40 mm

(black line) and 6 mm (gray line) source-detector distances during baseline, hypocapnia and ischemia (left-to-right).

The recovered tissue concentration curves for the brain (solid black line) and ECL (solid gray line), and the

corresponding arterial input function (dashed line) for the same data are presented in the bottom row.

3.4 Comparison of cerebral blood flow and blood volume measurements recovered with

dynamic contrast-enhanced near-infrared and CT perfusion

Figure 4 shows the regression plot of the CBF data; the line-of-best fit averaged across

the individual regression analyses had a slope of 1.06 ± 0.08, and a y-intercept of −4.37 ± 4.33

mL/100 g/min. The average slope was significantly different from the null (p < 0.001) but not

from the line of identity (p = 0.81). The average y-intercept was not significantly different from

19

zero (p = 0.76). The average r value, calculated from the individual regression analyses was 0.86

± 0.06. Figure 5 shows the Bland-Altman plot comparing the CBF measurements from the two

techniques. The mean difference between the two techniques was −1.7 mL/100 g/min, which was

bound by a 95% confidence interval of −16.3 to 12.9 mL/100 g/min. The average difference

between the two techniques was not significantly different from the null (p = 0.31).

Figure 4 Regression plot comparing CBF values calculated from DCE-NIR data and concomitant CT perfusion

values of CBF. Data are grouped by animal (n=8) and represented with a unique symbol. Regression analysis was

performed on each group individually, and the average line-of-best fit is indicated by the dashed line (slope = 1.06 ±

0.08, y-intercept = -4.37 ± 4.33 ml min 100g).

The regression analysis of CBV measured by the two techniques yielded an average line-

of-best fit with a slope of 2.49 ± 0.34 and a y-intercept of 0.14 ± 1.60 mL/100 g. The average

slope was significantly different from the null (p < 0.01) and from the line of identity (p < 0.05).

The y-intercept was not significantly different from zero (p = 0.90). The average r value for the

20

CBV regression analyses was 0.56 ± 0.13. From the Bland-Altman analysis, the mean difference

between BV calculated with the two techniques was 6.53 mL/100 g and the data was bound by a

95% confidence interval of 3.12 – 9.95 mL/100 g. Finally, the average difference between the

two techniques was significantly different from the null (p < 0.001).

Figure 5 Bland-Altman plot comparing the CBF measurements obtained with the DCE-NIR method and the CT

perfusion method. Data are grouped by animal using the same symbols as in Fig. 4. The mean difference between

the two methods is indicated by the solid line, and the 95% confidence interval is demarcated by the dashed line. No

statistically significant CBF magnitude effect was detected.

3.5 Error analysis of the dynamic contrast-enhanced near-infrared method

Figure 6 presents the results of error analyses investigating the error in the recovered

CBF due to uncertainties in the ECL thickness and in μs´ for the ECL and brain. First, sensitivity

functions were calculated based on assumed ECL thicknesses varied from 10 mm to 20 mm

(with the true ECL thickness set to 15 mm). The error in recovered CBF due to varying the ECL

thickness from 10 to 20 mm (true value was 15 mm) is presented in Fig. 6A. The CBF error

21

exhibited a parabolic relationship with the ECL error (y = −1.25 x2, R

2 = 0.94, p < 0.001).

Second, sensitivity functions were generated by varying μs´ for the ECL (μs´ECL) from 0.4 to 1.2

mm-1

(with a true value of 0.8 mm-1

). The error in recovered CBF associated with μs′ECL is shown

in Fig. 6B. Error in CBF ranged from −11.5 to 31.3% and a significant correlation between the

error in μs′ECL and error in recovered CBF was observed (slope = −0.39, y-intercept = 4%, r =

−0.96, p < 0.001). The error in recovered CBF due to variations in μa,ECL were also determined

(using a true value of 0.015 mm-1

). Equivalent variations in μa,ECL had about 25% the effect on

CBF as μs′ECL, with errors ranging from 7.6 to -3.5%. Similarly, sensitivity functions were

calculated based on μs´ for brain (μs′brain) ranging from 0.6 to 1.6 mm-1

(with a true value of 1.1

mm-1

). The error in recovered CBF associated with μs′brain mismatch is shown in Fig. 6C. The

error in recovered CBF ranged from −4.4 to 3.9% and a significant correlation between the error

in μs′brain and error in recovered CBF was also noted (slope = −0.09, y-intercept = 1%, r = −0.76,

p < 0.01).

Figure 6 (A) The relationship between error in assumed ECL thickness and error in the recovered CBF, simulated

for a range of ECL thicknesses between 10 and 20 mm. The nonlinear regression curve-of-best-fit (y = −1.25 x2) is

shown (dashed line). (B) The relationship between error in the ECL reduced scattering coefficient, μs'ECL, and the

22

error in recovered CBF. The regression line-of-best fit (slope = −0.39, y-intercept = 4%) is shown (dashed line). (C)

The relationship between error in the brain reduced scattering coefficient, μs'ECL, and the error in recovered CBF. The

regression line-of-best-fit (slope = −0.09, y-intercept = 1%) is shown (dashed line).

4. Discussion

Dynamic contrast-enhanced optical measurements of CBF have been proposed for over

25 years as a suitable approach for bedside neuro-monitoring, especially within the context of

critical care. In particular, because DII is a leading contributor to poor outcome in patients that

have sustained severe traumatic brain injury (Bouma et al., 1991) and subarachnoid hemorrhage

(Hijdra et al., 1988), the ability to measure CBF at the patient's bedside would be of great value.

Currently, only surrogate markers of CBF (i.e., intracranial pressure and cerebral perfusion

pressure), or invasive CBF measurements (i.e., thermal diffusion flowmetry) are available to

assess neurological recovery. Despite the significant clinical potential of bedside CBF

measurements, it is only recently that optical techniques have shown promise in achieving the

stated goal.

An early study by Owen-Reece and colleagues identified the problem of extracerebral

signal contamination as a confounding factor in optical measurements of CBF, resulting in a

large underestimation of CBF (Owen-Reece et al., 1996). To overcome this problem, they

suggested the use of a correction factor to calibrate NIRS CBF measurements against

measurements obtained with an alternate technique. This was, in part, based on the assumption

that the influence of the ECL was primarily a partial volume error, and that this effect would

show low intra-subject variability (Owen-Reece et al., 1996). The notion of a direct correlation

between NIRS-CBF measurements and CBF measured by an alternate method was both

supported (Keller et al., 2003) and challenged (Schytz et al., 2009) in subsequent studies. In

23

particular, Schytz and colleagues reported an absence of any correlation between CBF measured

concomitantly with NIRS and SPECT in healthy subjects before and after administration of

acetazolamide. This study highlighted the lack of a universal correction factor for NIRS

measurements of CBF, underlining the importance of subject-specific modeling using anatomical

priors (Elliott et al., 2010).

The advent of more sophisticated optical methods—specifically time-resolved and

frequency-domain NIR techniques—has resulted in increased sensitivity to deeper tissue

structures. Time-resolved methods increase the sensitivity to the brain since late-arriving photons

are statistically more likely to have interrogated deeper tissue (Liebert et al., 2004; Steinbrink et

al., 2001). Several papers have been published demonstrating qualitative curve analysis (Liebert

et al., 2012) and within subject quantification (Steinkellner et al., 2011) of ICG bolus

measurements acquired in neurocritical care patients with time-resolved methods. While this

work has established the sensitivity of DCE-NIR methods to CBF changes, only relative blood

flow indices such as time-to-peak (TTP) were calculated. A key requirement for extending this

approach to quantitative measurements is characterizing the arterial input function, which can be

accomplished non-invasively by dye densitometry (Brown et al., 2002). Similar to pulse

oximetry, a probe is clipped to either a finger or the nose of an adult patient, or either the hand or

foot of a neonate (Arora et al., 2013). Arrival time differences of ICG at the measuring site and

in the brain region interrogated by NIRS are implicitly accounted for in the KDOR analysis. A

further advantage to measuring the arterial input function is it removes potential within-subject

and between-subject variability in bolus administration (Elliott et al., 2013b).

It is likely that absolute CBF measurements will be more clinically useful than relative

blood flow indices, since well-defined CBF thresholds for cell dysfunction and irreversible cell

24

damage are known (Jones et al., 1981). Therefore, to provide absolute measurements of CBF,

time-resolved instrumentation was combined with subject-specific light propagation modeling,

and subsequently incorporated into the KDOR framework (Elliott et al., 2012). The main

objective of this study was to confirm the previous theoretical work by validating DCE-NIR CBF

measurements against CT perfusion measurements acquired concomitantly in adult pigs. The

adult pig model used in this work provides an excellent approximation to the adult human. First,

since the overall thickness of the ECL is similar to that of an adult human, the optical region of

integration contains the same overall brain contribution—the largest source of error in DCE-NIR

measurements in the adult. In these experiments, the relative brain contribution was of the order

of 30%, which is similar to previous estimates for the human adult (Okada and Delpy, 2003).

Second, following the scalp flow modification described in the methods section, the

extracerebral blood flow contribution is similar to that of a human: the average scalp blood flow

in these experiments was 11.7 ± 1.5 mL/100 g/min (data not shown). Finally, the baseline values

of CBF in the adult pig are similar to those observed in the human.

The representative example of the optical signals measured in one animal under the three

conditions (Fig. 3) illustrates the flow and depth dependency features of the DCE-NIR data. As

expected from previous studies (Elliott et al., 2013b; Liebert et al., 2004; Steinbrink et al., 2001),

important differences between attenuation and variance signals were observed. Baseline

differences suggest increased sensitivity of variance to faster cerebral tracer kinetics. The

variance signal also exhibited greater sensitivity than the attenuation signal to changes in CBF

caused by hypocapnia and ischemia. Improvements in sensitivity provided by the variance signal

facilitated the recovery of tissue concentration curves that appeared to reflect the higher blood

flow in the brain, both in terms of a higher peak concentration and faster temporal dynamics. An

25

unexpected feature of the tissue concentration curves recovered under the ischemia condition

was the apparent increase in ICG concentration during the recirculation period. This feature is

likely an artifact of the reconstruction process, as opposed to any real kinetic effect such as

extravasation. It is important to note that the constraints implemented by the KDOR inversion

have the greatest effect on the first pass of the dye (i.e., the initial rise and peak concentration of

the tissue concentration curve). Consequently, discrepancies in the shape of the reconstructed

tissue curve during the recirculation phase did not have any significant error in recovered CBF.

For all eight animals, values of CBF and CBV recovered with the DCE-NIR method were

compared to CT perfusion values—the most commonly used clinical method (Mayer et al.,

2000). In general, an excellent agreement was observed between CBF measurements acquired

with the two modalities with a precision comparable to previous studies published by our group

for neonatal and juvenile pigs (Brown et al., 2002; Elliott et al., 2010). Improved instrumentation

(time-resolved vs. continuous-wave) and the development of an optical reconstruction method

specifically for DCE data (KDOR) contributed to the strong correlation with CT perfusion

measurements, despite the use of animals with thicker ECLs and reduced scalp flow. These

findings confirm earlier numerical studies that suggested an improvement in precision of CBF

measurements by using the KDOR method compared with a traditional two-step reconstruction

and kinetic analysis approach (Elliott et al., 2012).

While CBV values measured with the DCE-NIR method showed a strong correlation

with CT perfusion measurements, the optical technique reported values that were on average 2.5

times larger (baseline mean values of 4.7 mL/100 g and 11.39 mL/100 g determined with CT

perfusion and DCE-NIR, respectively). One explanation for this discrepancy is that the two

methods utilize different kinetic modeling approaches to calculate hemodynamic parameters

26

from DCE. The CT Perfusion 4 program utilizes a plug flow model (Bisdas et al., 2008), which

has been demonstrated to underestimate the mean transit time in certain situations (Schabel,

2012; St Lawrence et al., 2013). Consequently, CBV will also be underestimated since it is

directly proportional to MTT by the central volume principle. In contrast, the DCE-NIR method

employs a nonparametric deconvolution to recover the impulse residue function, R(t). However,

it is possible that residual scalp contamination could cause the CBV values obtained by NIR to

be higher than expected, but further study is needed to confirm this. Nevertheless, it is important

to recognize that there was a consistent bias across all eight animals, suggesting that a simple

calibration curve could be used to relate the CBV values from the two modalities.

Given the strong correlation between the DCE-NIR method and CT perfusion, a closer

investigation of the analytical framework is pertinent. In particular, it is worthwhile investigating

the relative importance of the priors incorporated into the workflow before the KDOR routine is

executed. First, while the acquisition of anatomical imaging (i.e., CT or MRI) is ubiquitous in

almost all neurotrauma centers, the segmentation and incorporation of these data into a subject-

specific model may not be trivial within the context of acute care. In particular, it may be

difficult to provide an accurate method of coregistering the anatomical data with the location of

the optical probes. In these experiments this procedure was straightforward since the

polychloroprene probe holder was visible on the CT images. While the use of feducal markers

has been explored, it is likely that a landmark system, such as the 10-20 international system

(Custo et al., 2009) used for electroencephalogram placement, would be more clinically

compatible. To this end, the effect of errors in optical probe corregistration was investigated by

varying the ECL thickness. A significant effect on the recovered blood flow values was only

observed when the ECL thickness mismatch was extreme (> 5 mm), indicating that the DCE-

27

NIR method is relatively insensitive to small variations in ECL thickness that would be expected

in a typical patient. For example, we measured the variation in the ECL thickness, as determined

from T2-weighted MR images of five healthy subjects, was about ± 2 mm for positional changes

of about 15 mm in any direction (data not presented). This variation would result in an error of

less than 5% in CBF.

The error analysis of uncertainties in the optical properties demonstrated a significant

correlation between errors in μs′ECL and CBF. For example, an error of 40% in μs'ECL resulted in

approximately a 20% error in CBF. Given the range of reported μs'ECL values of the human head -

1.0 mm-1

(Boas et al., 2002), 1.2 mm-1

(Choi et al., 2004), and 1.7 mm-1

(Okada and Delpy,

2003)—assuming a known value could introduce a large amount of uncertainty in the derived

CBF values. By comparison, average values of μa,ECL and μs'ECL measured from the animals in

this study were 0.016 ± 0.002 mm-1

and 0.87 ± 0.12 mm-1

, respectively. In this study, this

potential source of error was minimized by acquiring baseline time-resolved data at a relatively

short source-detector distance (6 mm). In contrast, using an assumed value for the μs' in the brain

did not seem to introduce a significant error, as demonstrated in Fig 6C. In this case, an error in

μs′brain of 40% only resulted in an error in CBF of about 2-3%. These results indicate that the

biggest obstacle to measuring CBF accurately is properly modeling the amount of detected light

that reaches the cortical tissue, which explains the greater sensitivity to μs'ECL than μs′brain.

However, they also highlight a potential limitation in this study—that the accuracy of CBF

measurements is limited by the uncertainty in the ECL fitting routine. We compared optical

properties measured at 6 mm source-detector distance to those measured at 30 mm in a hard resin

phantom (μs′ = 0.82 mm-1

; INO, Québec City, Canada) and a soft polyvinyl alcohol phantom (μs′

= 1.09 mm-1

; ART Inc., Montréal, Canada). The uncertainty in the recovered scattering

28

coefficient calculated measurements obtained on these two phantoms were approximately 11%

and 9.2%, respectively (data not shown). While this represents a 7-8% uncertainty in CBF, it is

likely still an improvement in using population averaged values, and could be further reduced by

using hybrid PMT detectors which have IRF pulse-widths less than 150 ps. The absence of

significant bias in the CBF measurements, as shown in the Bland-Altman plot (Fig. 5), further

suggests that any systematic error in recovered scattering coefficient from 6 mm measurements

has a minimal impact on recovered CBF.

In this study of cerebral ischemia, intracortical injection of endothelin-1 was used to

produce a large lesion containing infarct and penumbra tissue states, which persisted long

enough to acquire the three measurements needed for this study. A limitation with this model is

that only a portion of blood flow in the region-of-interest was reduced to ischemic levels (14% of

total volume), as defined by CBF less than 10 mL/100 g/min; with penumbra (10 - 20 mL/100

g/min) and benign oligemia (20 - 35 mL/100 g/min) regions contributing to 18 and 19%,

respectively. Because the NIR region-of-interrogation encapsulated both the lesion and normal

surrounding tissue, the recovered CBF values by DCE-NIR did not reach ischemic levels. This

particular model was selected because of the challenges of causing cerebral ischemia in this

species. The substantial collateral flow through the vertebral arteries precludes the use a carotid

occlusion model (Tichauer et al., 2006). Furthermore, the presence of the rete mirabile, a plexus

of vessels located just distal to the carotid bifurcation, prevents the use of a middle cerebral

artery occlusion model (Burbridge et al., 2004). Despite the regional heterogeneity of this

ischemic model, the lack of any difference between the two techniques at any blood flow level

(Fig. 5), suggests that DCE-NIR should be able to measure ischemic CBF levels.

29

5. Conclusions

The present study validated a DCE-NIR method of measuring CBF that combines multi-

distance time-resolved measurements, arterial input function characterization, and subject-

specific light propagation modeling using the KDOR analytical framework. The good agreement

between CBF measurements obtained with this technique and CT perfusion in pigs—an animal

model that presents similar challenges to isolating the brain signal contribution expected in adult

humans—provides a convincing argument for using this robust optical method to measure CBF

in the neurocritical care unit. The ability to provide quantitative CBF measurements at the

bedside could help guide patient management and intervention. Patient studies could also be used

to provide further validation considering the increasing use of CT perfusion in the management

of neurological emergencies (Eilaghi et al., 2013; Kunze et al., 2012).

Disclosure/Conflict of Interest

JTE, MD, T-YL, and KS are inventors on PCT Application No. CA2013/000202 submitted

March 4, 2013 describing the kinetic deconvolution optical reconstruction method.

Acknowledgements

This study was supported through grants from the Ontario Neurotrauma Foundation and the

Canadian Institutes of Health Research, and graduate stipend support was provided by an Ontario

Graduate Scholarship. The authors thank Jennifer Hadway and Lise Desjardins for their help in

designing and conducting the animal experiments, Tae Sun Yoo for assistance in measuring

typical human head anatomy, and Lynn Keenliside for prototyping and system engineering.

30

Appendix

Derivation of Kinetic Deconvolution Optical Reconstruction Method

The analytical approach combines optical reconstruction and tracer kinetic modeling into

a single linear system in order to improve the accuracy of recovered hemodynamic parameters

(Elliott et al., 2012; Elliott et al., 2013a). Briefly, optical reconstruction is governed by the

forward problem:

j

jjii CAS , (A.1)

where Aij is the transformation between Cj, the concentration of dye in the jth

layer, and ΔSi, the

change in signal in the ith

detector caused by the absorption of light by the dye. The

transformation, Aij, also called the "Jacobian", is determined by light propagation modeling and

Cj is typically characterized using a direct or iterative solving method (Bonfert-Taylor et al.,

2012).

In the traditional DCE approach, kinetic analysis is performed on a series of

concentration maps obtained by solving Eq. 1 for each complete set of data collected as a

function of time. The time-dependent concentration, C(t), of a specific region-of-interest is given

by the following convolution (Meier and Zierler, 1954):

)()()( tFRtCtC a (A.2)

where Ca(t) is the arterial input function (the time-dependent concentration of dye in the arterial

system), F is blood flow, and R(t) is the impulse residue function. A flow-scaled R(t) is

determined by deconvoling Ca(t) from C(t). Blood volume is equal to the area under the curve of

31

FR(t) and the mean transit time (MTT) is equal to the ratio of blood volume and blood flow by

the central volume principle (Meier and Zierler, 1954).

Combining these two steps into a single linear system results in a problem that is better

posed and can be solved using inequality and equality constraints (Elliott et al., 2012). The

combination of Eq. 1 and 2, which forms the basis of KDOR, is given by the matrix equation:

TARCS FA (A.3)

where the nth column of the signal matrix, S, corresponds to ΔS measured at the n

th timepoint, CA

is the lower triangular matrix representation of Ca(t) (arising from the discretization of the

convolution in Eq. 2), the jth

column of RF corresponds to the FR(t) function for the jth

region in

the system, and AT is the transpose of the Jacobian, A. Quantification of blood flow, F, is

achieved by recovering RF through the optimization:

0 and 0 :subject to minarg2

2 FFFA

R

HRGRARCSF

T

Since the maximum of R(t) is, by definition, unity, CBF is calculated from the maximum of

FR(t) for the region corresponding to the brain. Equality and inequality constraints G and H are

constructed on the basis of physiological assumptions, and a complete derivation of these can be

found elsewhere (Elliott, 2013). Briefly, there are four assumptions that are used: (i) tracer

requires a minimum time to arrive at the region-of-interest; (ii) since R(t) represents the residue

function of a Dirac-delta function, the entire bolus of dye arrives at a single instant in time; (iii)

there is a minimum transit time before the first dye molecules leave the region of interrogation;

(iv) no retrograde flow can occur (i.e., negative monotonicity is enforced). A constrained non-

32

negative least squares inversion, nested within an exhaustive search of minimum transit time and

arrival time, was used to solve RF―additional details of this procedure have been described

elsewhere (Elliott, 2013; Elliott et al., 2013a; Elliott et al., 2013c).

33

References

Arora, R., Ridha, M., Lee, D.S., Elliott, J., Rosenberg, H.C., Diop, M., Lee, T.Y., St Lawrence,

K., 2013. Preservation of the metabolic rate of oxygen in preterm infants during

indomethacin therapy for closure of the ductus arteriosus. Pediatr Res 73, 713-718.

Bisdas, S., Konstantinou, G., Surlan-Popovic, K., Khoshneviszadeh, A., Baghi, M., Vogl, T.J.,

Koh, T.S., Mack, M.G., 2008. Dynamic contrast-enhanced CT of head and neck tumors:

comparison of first-pass and permeability perfusion measurements using two different

commercially available tracer kinetics models. Acad Radiol 15, 1580-1589.

Boas, D., Culver, J., Stott, J., Dunn, A., 2002. Three dimensional Monte Carlo code for photon

migration through complex heterogeneous media including the adult human head. Opt

Express 10, 159-170.

Bonfert-Taylor, P., Leblond, F., Holt, R.W., Tichauer, K., Pogue, B.W., Taylor, E.C., 2012.

Information loss and reconstruction in diffuse fluorescence tomography. J Opt Soc Am A

Opt Image Sci Vis 29, 321-330.

Bouma, G.J., Muizelaar, J.P., Choi, S.C., Newlon, P.G., Young, H.F., 1991. Cerebral circulation

and metabolism after severe traumatic brain injury: the elusive role of ischemia. J

Neurosurg 75, 685-693.

Brown, D.W., Picot, P.A., Naeini, J.G., Springett, R., Delpy, D.T., Lee, T.Y., 2002. Quantitative

near infrared spectroscopy measurement of cerebral hemodynamics in newborn piglets.

Pediatr Res 51, 564-570.

Burbridge, B., Matte, G., Remedios, A., 2004. Complex intracranial arterial anatomy in swine is

unsuitable for cerebral infarction projects. Can Assoc Radiol J 55, 326-329.

Choi, J., Wolf, M., Toronov, V., Wolf, U., Polzonetti, C., Hueber, D., Safonova, L.P., Gupta, R.,

Michalos, A., Mantulin, W., Gratton, E., 2004. Noninvasive determination of the optical

properties of adult brain: near-infrared spectroscopy approach. J Biomed Opt 9, 221-229.

Custo, A., Boas, D.A., Tsuzuki, D., Dan, I., Mesquita, R., Fischl, B., Grimson, W.E., Wells, W.,

3rd, 2009. Anatomical atlas-guided diffuse optical tomography of brain activation.

Neuroimage 49, 561-567.

Diop, M., St Lawrence, K., 2012. Deconvolution method for recovering the photon time-of-flight

distribution from time-resolved measurements. Opt Lett 37, 2358-2360.

Diop, M., Tichauer, K.M., Elliott, J.T., Migueis, M., Lee, T.Y., St Lawrence, K., 2010.

Comparison of time-resolved and continuous-wave near-infrared techniques for measuring

cerebral blood flow in piglets. J Biomed Opt 15, 057004.

Eilaghi, A., Brooks, J., d'Esterre, C., Zhang, L., Swartz, R.H., Lee, T.Y., Aviv, R.I., 2013.

Reperfusion is a stronger predictor of good clinical outcome than recanalization in

ischemic stroke. Radiology 269, 240-248.

Elliott, J.T., 2013. On the development of a dynamic contrast-enhanced near-infrared technique

to measure cerebral blood flow in the neurocritical care unit. Department of Medical

Biophysics. The University of Western Ontario, London, ON.

34

Elliott, J.T., Diop, M., Lee, T.Y., Lawrence, K.S., 2012. Model-independent dynamic constraint

to improve the optical reconstruction of regional kinetic parameters. Opt Lett 37, 2571-

2573.

Elliott, J.T., Diop, M., Morrison, L.B., Lee, T.Y., St Lawrence, K., 2013a. Reconstruction of

cerebral hemodynamics with dynamic contrast-enhanced time-resolved near-infrared

measurements before and during ischemia. Proc SPIE, pp. 8578-8510.

Elliott, J.T., Diop, M., Tichauer, K.M., Lee, T.Y., St Lawrence, K., 2010. Quantitative

measurement of cerebral blood flow in a juvenile porcine model by depth-resolved near-

infrared spectroscopy. J Biomed Opt 15, 037014.

Elliott, J.T., Milej, D., Gerega, A., Weigl, W., Diop, M., Morrison, L.B., Lee, T.Y., Liebert, A.,

St Lawrence, K., 2013b. Variance of time-of-flight distribution is sensitive to cerebral

blood flow as demonstrated by ICG bolus-tracking measurements in adult pigs. Biomed

Opt Express 4, 206-218.

Elliott, J.T., St Lawrence, K., Diop, M., Lee, T.Y., Tichauer, K.M., 2013c. Kinetic deconvolution

optical reconstuction method. PCT patent application PCT/CA2013/000202. International

Publication Number. WO 2013/127003 A1, 4 March 2013.

Fang, Q., Boas, D.A., 2009. Monte Carlo simulation of photon migration in 3D turbid media

accelerated by graphics processing units. Opt Express 17, 20178-20190.

Franceschini, M.A., Joseph, D.K., Huppert, T.J., Diamond, S.G., Boas, D.A., 2006. Diffuse

optical imaging of the whole head. J Biomed Opt 11, 054007.

Friberg, L., Kastrup, J., Hansen, M., Bulow, J., 1986. Cerebral effects of scalp cooling and

extracerebral contribution to calculated blood flow values using the intravenous 133Xe

technique. Scand J Clin Lab Invest 46, 375-379.

Gora, F., Shinde, S., Elwell, C.E., Goldstone, J.C., Cope, M., Delpy, D.T., Smith, M., 2002.

Noninvasive measurement of cerebral blood flow in adults using near-infrared

spectroscopy and indocyanine green: a pilot study. J Neurosurg Anesthesiol 14, 218-222.

Hijdra, A., van Gijn, J., Nagelkerke, N.J., Vermeulen, M., van Crevel, H., 1988. Prediction of

delayed cerebral ischemia, rebleeding, and outcome after aneurysmal subarachnoid

hemorrhage. Stroke 19, 1250-1256.

Jones, T.H., Morawetz, R.B., Crowell, R.M., Marcoux, F.W., FitzGibbon, S.J., DeGirolami, U.,

Ojemann, R.G., 1981. Thresholds of focal cerebral ischemia in awake monkeys. J

Neurosurg 54, 773-782.

Keller, E., Nadler, A., Alkadhi, H., Kollias, S.S., Yonekawa, Y., Niederer, P., 2003. Noninvasive

measurement of regional cerebral blood flow and regional cerebral blood volume by near-

infrared spectroscopy and indocyanine green dye dilution. Neuroimage 20, 828-839.

Kienle, A., Patterson, M.S., 1997. Improved solutions of the steady-state and the time-resolved

diffusion equations for reflectance from a semi-infinite turbid medium. J Opt Soc Am A

Opt Image Sci Vis 14, 246-254.

Kunze, E., Pham, M., Raslan, F., Stetter, C., Lee, J.Y., Solymosi, L., Ernestus, R.I., Vince, G.H.,

Westermaier, T., 2012. Value of Perfusion CT, Transcranial Doppler Sonography, and

35

Neurological Examination to Detect Delayed Vasospasm after Aneurysmal Subarachnoid

Hemorrhage. Radiol Res Pract 2012, 231206.

Lee, T.-Y., 2002. Functional CT: physiological models. Trends in biotechnology 20, S3-S10.

Liebert, A., Milej, D., Weigl, W., Gerega, A., Kacprzak, M., Mayzner-Zawadzka, E.,

Maniewski, R., 2012. Assessment of brain perfusion disorders by ICG bolus tracking with

time-resolved fluorescence monitoring. BIOMED. Optical Society of America, Miami, FL,

p. BTu3A.

Liebert, A., Wabnitz, H., Steinbrink, J., Moller, M., Macdonald, R., Rinneberg, H., Villringer,

A., Obrig, H., 2005. Bed-side assessment of cerebral perfusion in stroke patients based on

optical monitoring of a dye bolus by time-resolved diffuse reflectance. Neuroimage 24,

426-435.

Liebert, A., Wabnitz, H., Steinbrink, J., Obrig, H., Moller, M., Macdonald, R., Villringer, A.,

Rinneberg, H., 2004. Time-resolved multidistance near-infrared spectroscopy of the adult

head: intracerebral and extracerebral absorption changes from moments of distribution of

times of flight of photons. Appl Opt 43, 3037-3047.

Liu, H., Chance, B., Hielscher, A.H., Jacques, S.L., Tittel, F.K., 1995. Influence of blood vessels

on the measurement of hemoglobin oxygenation as determined by time-resolved

reflectance spectroscopy. Med Phys 22, 1209-1217.

Mayer, T.E., Hamann, G.F., Baranczyk, J., Rosengarten, B., Klotz, E., Wiesmann, M., Missler,

U., Schulte-Altedorneburg, G., Brueckmann, H.J., 2000. Dynamic CT perfusion imaging

of acute stroke. AJNR Am J Neuroradiol 21, 1441-1449.

Meier, P., Zierler, K.L., 1954. On the theory of the indicator-dilution method for measurement of

blood flow and volume. J Appl Physiol 6, 731-744.

Murphy, B.D., Fox, A.J., Lee, D.H., Sahlas, D.J., Black, S.E., Hogan, M.J., Coutts, S.B.,

Demchuk, A.M., Goyal, M., Aviv, R.I., Symons, S., Gulka, I.B., Beletsky, V., Pelz, D.,

Hachinski, V., Chan, R., Lee, T.Y., 2006. Identification of penumbra and infarct in acute

ischemic stroke using computed tomography perfusion-derived blood flow and blood

volume measurements. Stroke 37, 1771-1777.

Okada, E., Delpy, D.T., 2003. Near-infrared light propagation in an adult head model. I.

Modeling of low-level scattering in the cerebrospinal fluid layer. Appl Opt 42, 2906-2914.

Okada, E., Firbank, M., Schweiger, M., Arridge, S.R., Cope, M., Delpy, D.T., 1997. Theoretical

and experimental investigation of near-infrared light propagation in a model of the adult

head. Appl Opt 36, 21-31.

Owen-Reece, H., Elwell, C.E., Harkness, W., Goldstone, J., Delpy, D.T., Wyatt, J.S., Smith, M.,

1996. Use of near infrared spectroscopy to estimate cerebral blood flow in conscious and

anaesthetized adult subjects. Br J Anaesth 76, 43-48.

Schabel, M.C., 2012. A unified impulse response model for DCE-MRI. Magn Reson Med 68,

1632-1646.

36

Schytz, H.W., Wienecke, T., Jensen, L.T., Selb, J., Boas, D.A., Ashina, M., 2009. Changes in

cerebral blood flow after acetazolamide: an experimental study comparing near-infrared

spectroscopy and SPECT. Eur J Neurol 16, 461-467.

St Lawrence, K., Verdecchia, K., Elliott, J., Tichauer, K., Diop, M., Hoffman, L., Lee, T.Y.,

2013. Kinetic model optimization for characterizing tumour physiology by dynamic

contrast-enhanced near-infrared spectroscopy. Phys Med Biol 58, 1591-1604.

Steinbrink, J., Wabnitz, H., Obrig, H., Villringer, A., Rinneberg, H., 2001. Determining changes

in NIR absorption using a layered model of the human head. Phys Med Biol 46, 879-896.

Steinkellner, O., Gruber, C., Wabnitz, H., Jelzow, A., Steinbrink, J., Fiebach, J.B., Macdonald,

R., Obrig, H., 2011. Optical bedside monitoring of cerebral perfusion: technological and

methodological advances applied in a study on acute ischemic stroke. J Biomed Opt 15,

061708.

Tichauer, K.M., Brown, D.W., Hadway, J., Lee, T.Y., St Lawrence, K., 2006. Near-infrared

spectroscopy measurements of cerebral blood flow and oxygen consumption following

hypoxia-ischemia in newborn piglets. J Appl Physiol 100, 850-857.

Vergouwen, M.D., Vermeulen, M., van Gijn, J., Rinkel, G.J., Wijdicks, E.F., Muizelaar, J.P.,

Mendelow, A.D., Juvela, S., Yonas, H., Terbrugge, K.G., Macdonald, R.L., Diringer,

M.N., Broderick, J.P., Dreier, J.P., Roos, Y.B., 2010. Definition of delayed cerebral

ischemia after aneurysmal subarachnoid hemorrhage as an outcome event in clinical trials

and observational studies: proposal of a multidisciplinary research group. Stroke 41, 2391-

2395.

Waydhas, C., 1999. Intrahospital transport of critically ill patients. Crit Care 3, R83-89.