## Abstract

There is an increasing need for quantitative and computationally affordable models
for analyzing tissue metabolism and hemodynamics in microvascular networks. In this
work, we develop a hybrid model to solve for the time-varying oxygen
advection-diffusion equation in the vessels and tissue. To obtain a three-dimensional
temporal evolution of tissue oxygen concentration for realistic complex vessel
networks, we used a graph-based advection model combined with a finite-element based
diffusion model and an implicit time-advancing scheme. We validated this algorithm
for both static and dynamic conditions. We also applied it to a complex vascular
network obtained from a rodent somatosensory cortex. Qualitative agreement was found
with *in-vivo* experiments.

©2008 Optical Society of America

^{◊}Data sets associated with this article are available at http://hdl.handle.net/10376/1051. Links such as View 1 that appear in figure captions and elsewhere will launch custom data views if ISP software is present.

## 1. Introduction

Numerous optical methods are utilized to quantify tissue blood flow, blood volume, and oxygen delivery, ranging from the macroscopic whole tissue level to the microscopic capillary network level and sub-cellular level. Near-infrared spectroscopy (NIRS) and imaging utilize near-infrared light to quantify the concentrations of oxygenated and deoxygenated hemoglobin over cubic centimeters of tissue [1–3]. Visible light tissue reflectance imaged with a CCD camera is frequently used to image superficial changes in hemoglobin concentrations with light spectroscopy [4] and blood flow with laser speckle contrast [5, 6], particularly in the neurosciences. Confocal and multiphoton microscopy enable sub-micrometer resolution with depth penetration of up to ~100 Μm and ~500 Μm or more, respectively, and are used to image the three-dimensional (3D) vascular anatomy and red blood cell velocity within vessels [7, 8], and oxygenation of the blood [9–11]. Doppler optical coherence tomography is enabling rapid volumetric imaging of flow within the vascular network over potentially cubic millimeters in seconds [12–14]. There seems to be a unique opportunity for optics to provide a detailed characterization of oxygen delivery to the tissue by combining these measurements of blood flow, volume, and oxygenation with fluorescence-based measurements of cellular activity [15, 16] and metabolism [17, 18].

The advanced analysis of near-infrared spectroscopy data from centimeters of tissue and visible light imaging from hundreds of micrometers of tissue are yielding information about oxygen metabolism [19–23]. While these estimates of oxygen metabolism appear precise, validating the accuracy is challenged by the lack of a gold standard method. On the other hand, the microscopic measures of individual microvessels and surrounding tissue provide more direct measures of oxygen delivery, but detailed models are needed to provide an analysis framework to quantify physiological parameters like vessel wall oxygen permeability, tissue oxygen diffusion, and oxygen consumption that are not directly measurable. Further, due to the time varying nature of oxygen delivery, particularly during brain activity, there is a need for models that can handle dynamic, time-varying processes.

In this paper, we describe a detailed computational model of tissue oxygen delivery by a
microvascular network that will provide a framework for analyzing multimodal microscopic
measures to better quantify additional physiological parameters, and will enable more
detailed and realistic analyses of the accuracy of near-infrared spectroscopy and
visible light imaging. We start by describing our method for solving the time-varying
oxygen advection diffusion equation that estimates velocity within individual
microvessel segments, oxygen flux across the permeable vascular walls, and then oxygen
diffusion and consumption within the tissue. After validating our solver, we demonstrate
its application by calculating oxygen delivery by a microvascular network acquired in
rodent somatosensory cortex by two-photon microscopy over a 230×230×450
µm^{3} volume of tissue.

## 2. Methods

#### 2.1 Models and Domain Discretization

The transport and consumption of oxygen in the vessel network and tissue are characterized by the advection-diffusion equation [24], which can be expressed as the following equations:

where *C*
_{F} and *C*
_{B} are the free and bounded oxygen concentrations (µM),
respectively; *C*
_{T}=*C*
_{B}+*C*
_{F} denotes the total oxygen concentration, *v*⇀
denotes the velocity (m/s), *D*O2 is the oxygen diffusion coefficient
(m/s^{2}), and OC is the tissue oxygen consumption rate (µM/s).
Equation (2) denotes the
equilibrium between the hemoglobin bound oxygen and free oxygen where
*H* denotes hematocrit, *C _{Hb}* is the hemoglobin concentration within a red blood cell,
SO

_{2}(

*C*

_{F}) is the hemoglobin oxygen saturation and it is in equilibrium with the concentration of free oxygen,

*C*

_{F}, as described by the oxygen disassociation curve. We use an invertible form for the oxygen disassociation curve given in Ref. [25]. We note that under physiological conditions, all of these parameters can vary in space and time.

Certainly, one could solve Eq. (1) using a single numerical approach, such as a finite element (FE) or a finite difference (FD) method, by applying a full spatial discretization of the tissue volume, the exterior surface and the interior volume of vessels. Because of the small length-scale of the vessel wall, the meshing and computational expense can be prohibitive with increasing complexity of the vessel networks. Instead, we devised a hybrid model that includes a graph-based one-dimensional (1D) oxygen advection model within the vessel, a 1D oxygen flux conservation model across the vessel wall, and an FE-based 3D oxygen diffusion model within the tissue. An implicit (Crank-Nickson) [26] scheme was implemented to compute the updates at each time step. A schematic of the problem domain decomposition and our spatial discretization scheme is shown in Fig. 1. The computational space is separated into three domains to characterize three distinct physical and physiological processes:

I. vessel network: inside the vessel, the oxygen transport is represented by the advection equation subject to the equilibrium of the free and hemoglobin bound oxygen:

$$\frac{\partial {C}_{B}}{\partial t}=-\stackrel{\rightharpoonup}{v}\xb7\nabla {C}_{B},$$

$${C}_{B}=4{C}_{Hb}HS{O}_{2}\left({C}_{F}\right).$$

An oriented graph representation was created to represent the vessel network: the axial line of each vessel was extracted and discretized by nodes and oriented edges (see Fig. 1). The connection (including branches) of the vessel network can be naturally expressed by the adjacency matrix [27] of the graph. We calculated a vessel volume associated with each vessel node and assumed uniform oxygen distribution within the volume.

II. vessel wall: the oxygen flux across a thin vessel wall is a diffusion process. To simplify analysis, we only discretized the exterior of vessel walls by a triangular surface mesh. Because the diffusion profile within the vessel wall layer is not the primary interest of this study, we used an oxygen conversation relationship based on Fick’s law [28]:

where *J*
_{i} is the oxygen flux (mol/m^{2}/s) across the vessel wall at the
*i*
^{th} vessel wall node, *A*
_{i} and *V*
_{i} are the associated vessel wall area and tissue volume for this node. The
right-hand-side of Eq. (4) represents
the extracted oxygen from the vessel per unit time. The flux, *J _{i}*, is computed by

where *K _{w}* is the vessel wall permeability, w is the vessel wall thickness,

*C*and

_{F,i}*C*are the free oxygen concentrations at the exterior and interior surfaces of the vessel, respectively;

_{F,i0}*α*is the Bunsen solubility coefficient and equals 1.27×10

^{-15}µmol/(µm

^{3}mmHg) [28].

III. tissue: the free oxygen that is transferred across the vessel wall diffuses through the whole tissue volume governed by the diffusion equation

where *D*
_{O2} is the tissue oxygen diffusion coefficient and the consumption of
oxygen is governed by OC. The tissue volume is discretized by a tetrahedron mesh and
is truncated by a bounding box.

In the following subsections, we detail our numerical implementation for solving the above models for each domain.

#### 2.1.1 Vessel blood advection

Equation (3) is solved in two steps
for each time advance. We first solve the two advection equations in Eq. (3). Integrated within the associated
volume of each vessel node, the advection equations lead to the conservation
equations of free and bound oxygen. Assuming that the adjacency matrix for the vessel
graph is *K*, for the *i*
^{th} vessel node (excluding the vessel end nodes) the conservation of oxygen
implies that

where *N _{ν}* is the total number of vessel nodes;

*k*is the (

_{i,j}*i,j*)

^{th}element of

*K*with a value of 0 if nodes

*i*and

*j*are not adjacent, 1 for in-flow from node

*j*to

*i*, and -1 for out-flow from node

*i*to

*j; d*is the distance between the

_{i,j}*i*

^{th}and

*j*

^{th}nodes and

*C*

_{i}and C

_{j}are the oxygen concentrations (free or hemoglobin bound);

*ν*

_{i,j}is the velocity,

*V*

_{i}and

*V*

_{j}are the associated vessel volumes of the two nodes. An implicit finite difference scheme is used for the time-derivative term in (7). This results in a linear equation with a form of

where *θ*∈[0,1] is a constant,
Δ*t* is the time step and n is the time index. This can be
re-written in matrix form as

$${A}_{a}{\mathbf{C}}_{B}^{n+1}={B}_{a}{\mathbf{C}}_{B}^{n}.$$

The equilibrium of *C*
_{F} and *C*
_{B} is achieved by solving

subject to the conservation of *C*
_{T}.

#### 2.1.2 Oxygen flux across vessel wall

From Eq.(4), we can get

which can be converted into an implicit time update as

The matrix form of Eq. (12) can be written as a linear equation defined between the vessel wall and axial nodes:

#### 2.1.3 Tissue oxygen diffusion

To accommodate for the arbitrary shape of a complex vessel network, we used a tetrahedral mesh to discretize the tissue region outside the vessel. A finite element method (FEM) coupled with implicit time-stepping is used to solve for the diffusion of oxygen in the tissue.

Using the Galerkin’s method, the FEM weak form of Eq. (6) can be written as [26]

where *ϕ _{i}* and

*ϕ*are the basis functions in a tetrahedral element;

_{j}*∂*Ω is the surface of the tissue, and

*n*̂ is a unit vector pointing to the normal direction of a surface element; <∙> denotes volume integration inside an element. After applying the implicit method to discretize the time derivative in Eq. (14), we obtain

where the elements of matrices *A _{d}* and

*B*and vector

_{d}*R*are, respectively, defined by

#### 2.1.4 Lumped system

The discretized advection equation, Eq. (9), flux equation, Eq. (13) and diffusion equation, Eq. (15), are defined on the vessel axial nodes, vessel surface/axial nodes and vessel surface/tissue nodes, respectively. In order to solve the entire domain, we can assemble these equations into a lumped system by assigning a global index for the nodes located in three regions. Because of the nonlinear nature of Eq. (10), we solve for the advection separately. We combine Eq. (13) and Eq. (15) as

Then, we solve for the time evolution of the oxygen concentration (including both free and hemoglobin bound oxygen) by solving Eqs. (9), (10), and (16) sequentially.

We also note that the unconditional stability of implicit time-stepping may no longer hold due to the presence of Eq. (10). A rigorous derivation of the stability condition for this model is beyond the scope of this paper. We confirmed the stability of our solution by obtaining identical solutions with smaller time steps.

#### 2.1.5 Boundary conditions

Equations (9) and (16) must be solved with proper boundary conditions. The boundary nodes include all the nodes on the exterior surface of the domain, which are either tissue surface nodes or the in-flow/out-flow vertices of vessel axes. For the tissue surface nodes, we used a mirror boundary condition. i.e.

This effectively eliminates the surface integration term in Eq. (14) and the expression for the
elements of the vector **R** is simplified to

*r _{i}*=<OC,

*ϕ*>Δ

_{i}*t*.

For the vessel axes end nodes, the flow rates at all in-flow peripheral points are specified (type-1 boundary condition) and a sink condition is used for the out-flow peripheral points.

## 2.2 Two-photon microscopy measurements of rodent microvasculature network

Sprague Dawley rats (250–320 g) were initially anesthetized with isofluorane and ventilated with a mixture of air and oxygen during surgical procedures. Cannulas were inserted in the femoral artery and vein and heart rate, blood pressure, body temperature and blood gases were monitored. A cranial window 3 mm×3 mm in size was opened in the parietal bone and dura was removed. The window was filled with agarose and sealed with a coverslip. During the measurement isofluorane was discontinued and anesthesia was maintained with 50 mg/kg intravenous bolus of alpha-chloralose followed by continuous intravenous infusion at 40 mg/(kg h). All experimental procedures were approved by the Massachusetts General Hospital Subcommittee on Research Animal Care.

Blood plasma was labeled with Dextran-conjugated fluorescein (FD 2000S, Sigma-Aldrich;
500 nM concentration in blood). Imaging of the cortical vasculature was performed with a
two-photon microscope (Ultima, Prairie Technology Inc.) at 800 nm and with an Olympus
40X water immersion objective (*NA*=0.8). The individual 256×256
pixels frames in the *X-Y* planes perpendicular to the optical axis
spanned 230 μm × 230 μm. They were obtained as an average of four
images in order to smooth the appearance of the red blood cells (not labeled with the
dye) and to improve further image analysis. The step size between frames along the
optical axis was set to Δz=1 μm and adjustment of optical power was made
continuously with the imaging depth to address increased scattering and absorption along
the optical path-length. An image stack was acquired over a depth of 450 μm.

## 2.3 Graphing of the vessel network

Our two-photon microscopy images of the vascular network have a high contrast-to-noise
ratio enabling easy identification of the vascular networks. Rather than employing an
automatic segmentation procedure for graphing the vascular network, we manually graph
the network using custom software to accelerate the process. The software enables us to
place node points and connecting edges along the vessels. Vessel diameter is estimated
at each node by thresholding the image at a low value of approximately 2% of the maximum
image intensity, but still above the image noise, considering lines through the node
point oriented every 3 degrees in the local *X-Y* plane perpendicular to
the vessel axis, and finding the minimum distance from vessel edge to vessel edge. At
the graph end points we need to establish boundary conditions for velocity or blood
pressure. We set the blood pressure boundary condition for the major arterioles and
venules and use a zero velocity boundary condition for minor vessel branches that do not
connect back to our draining vein. The vascular resistance for each edge between 2 node
points is given by Poiseuille’s law

where *η* is the viscosity of blood
(*η*=15×10^{-6} mmHg s [29]), *l* is the length of the vessel, and
*d* is the vessel diameter. Given the resistor network and boundary
conditions on blood pressure and velocity, it is straight-forward to solve the set of
simultaneous equations for blood pressure and blood velocity at every node point. We
follow the procedure we described previously, in Ref. [30].

## 3. Results

In this section, we first present our simulation results for a simple vessel network: a
single vessel across the tissue. Using these results, we validate our model proposed in
the previous section for both steady-state and dynamic simulations. In the second half
of this section, we apply our analysis to a more complex vessel network extracted from a
two-photon microscopy scan of a rodent somatosensory cortex. These simulations further
demonstrate the accuracy and scalability of the proposed model. We note that, while our
calculations are for the concentration of oxygen, we report the partial pressure of
oxygen (PO_{2}) in our figures as it is commonly reported in experimental
papers. The PO_{2} is related to the oxygen concentration, *C*,
by PO_{2}=*C/α*.

#### 3.1 Validation

The tissue that we are simulating is 100×100×100
µm^{3} in size and contains a single blood vessel with diameter of
*d*=10 µm. The vessel is passing through the center of the
tissue and parallel to the *x*-axis. Unless otherwise stated, for this
simple simulation the velocity of blood inside the vessel is 400 µm/s,
hematocrit is set to 0.1, and PO_{2}=90 mmHg at the input boundary of the
vessel. The vessel wall permeability is *K _{w}*=1.115×10

^{-12}µmol/(µm s mmHg) and wall thickness is 1 µm. The oxygen diffusion coefficient is

*D*

_{O2}=2.4×10

^{6}µm

^{2}/s. Oxygen consumption is 41.7×10

^{-16}mmol/(µm

^{3}s), or 10% of the baseline value for the rat cortex [31]. A mesh with 25554 nodes and 120942 tetrahedral elements was created using our customized mesh generator. The meshing process took about 14 seconds on a 1.8GHz Intel Xeon (64bit) PC and building the matrices in Eqs. (9) and (16) took about 11 seconds. Subsequently, we solved Eqs. (9), (10) and (16) for about 1.3 seconds per time step.

In Fig. 2(a) we show a contour plot of the
steady state PO_{2}, obtained for the above simulation parameters in
*X-Y* plane that contains the vessel axis. We see a high
PO_{2} for the inflowing blood on the left (starting at 90 mmHg),
diminishing as the blood flows across the vessel to a minimum of 64 mmHg indicated by
the last contour on the right. In Fig. 2(b),
2(c), and 2(d), we plot the PO_{2} along the interior of the vessel for
various values of the blood velocity, hematocrit, and tissue oxygen consumption,
respectively. As we increase velocity from 0.4 mm/s to 0.8 mm/s and 1.6 mm/s, we see
that the PO_{2} is increasing as more oxygen is being delivered by the blood
per unit time, but the oxygen extraction by the tissue is constant, and thus the
oxygen extraction fraction is expected to decrease. Similarly, as we increase the
hematocrit from 0 to 0.2 we see that PO_{2} increases as more oxygen is
delivered by the blood. When tissue oxygen consumption is increased, the
PO_{2} in the blood decreases, as expected. The difference between the
oxygen flowing into the volume per unit time and the oxygen flowing out should be
equal to the tissue volume integral of the oxygen consumption. Assuming the uniform
oxygen consumption (OC, in µM/s), the oxygen conservation requires

where *v* is the velocity of blood, *C*
_{T,in} and *C*
_{T,out} are the total concentrations of oxygen at the input and output
boundaries of the blood vessel, respectively, and *V*
_{tis} is the tissue volume. The left-hand-side of Eq. (19) denotes the total oxygen
extracted from the vessel per unit time while the right-hand-side denotes the
quantity of oxygen consumed by the tissue.

To validate the steady-state solution obtained by our finite element model, we calculated the PO2 distributions under various velocities, hematocrit rates and OC values, and the solutions are shown in Fig. 2. The expected conservation relationship, i.e. Eq. (19), holds for all the steady-state simulations.

The second step is to validate the model for dynamic processes. In these simulations, we separately explored the dynamics of oxygen advection within the vessel, flux across the vessel membrane and diffusion in the tissue and compared the calculations with simple analytic solutions. These results are shown in Fig. 3 and detailed below.

Oxygen advection within the vessel was isolated from oxygen diffusion by setting the
vessel wall permeability *K _{w}*=0. We started with no oxygen in the vessel and introduced 90 mmHg of oxygen
pressure at the input of the vessel ramping down to 0 mmHg at 40 µm into the
vessel. Fig. 3(a) shows the oxygen pressure
profiles along the vessel at 20 ms time intervals. We see that the oxygen wavefront
is advancing almost 8 µm every 20 ms, consistent with the velocity of 0.4
mm/s. It is known that a two-level implicit advection solver produces solutions that
underestimate the true velocity at high spatial frequencies and accurately estimate
the velocity at low spatial frequencies [26].
This is evident in Fig. 3(a) with smoothing
of the edges at 90 mmHg and 0 mmHg as the oxygen advects along the vessel. We note
that this is an intrinsic limitation of the discretization scheme, and although we
have implemented the most accurate solver, it places limits on the accuracy of our
solution for oxygen content varying with high spatial frequency along a vessel. As
noted in Ref. [26], this limitation can be
reduced by using a refined time step or smaller grid spacing.

Flux of oxygen across the vessel wall was validated by fixing the vessel
PO_{2} at 90 mmHg, initiating the tissue PO_{2} at 0 mmHg, setting
the tissue oxygen consumption to 0, and increasing the oxygen diffusion coefficient
within the tissue by a factor of 1000 such that the system dynamics would be limited
by the vessel wall permeability *K _{w}*. In Fig. 3(b), we plot the tissue
temporal evolution of PO

_{2}, which is increasing to 90 mmHg due to oxygen flux across the permeable vessel wall. When we reduce

*K*by a factor of 2, we see that the rate of increase for the tissue PO2 is reduced. The rate of change in the total oxygen content of the tissue volume is given by Fick’s law [28] as

_{w}where *N*
_{O2} is the number of moles of oxygen in the vessel or tissue,
*d* is the diameter of the vessel, *l* is the vessel
length, *w* is the vessel wall thickness, and we assume that any
oxygen that crosses the vessel wall distributes uniformly throughout the tissue
volume effectively immediately (i.e., the diffusion time constant is much shorter
than the vessel wall flux time constant). Under these conditions, the solution of
Eq. (20) is a simple exponential.
Note that

where x indicates either tissue (*tis*) or vessel
(*ves*). The solution of Eq. (21) is shown in Fig. 3(b) and
agrees well with the solution of our finite element model.

Oxygen diffusion was validated by setting oxygen consumption to 0 and initiating the
tissue PO_{2} to 0 mmHg except for a single volume element far from the
boundaries of the tissue volume that was given a unit quantity of oxygen. This oxygen
was then diffused with our finite element model. The early responses of diffusion
after 20 (blue), 25 (green), and 30 ms (red) are plotted in Fig. 3(c) and they roughly agree with the analytic solution of
the diffusion equation for a homogeneous infinite medium. We also computed the finite
element solution reducing the diffusion coefficient by a factor of 2 and plot the
results at 40 (blue), 50 (green), and 60 ms (red) and observed the expected
*D*
_{O2}
*t* scaling of the diffusion equation. That is, if we half the
diffusion coefficient and double the time, the analytic solution [32] predicts the same solution.

Tissue oxygen consumption was validated by setting PO_{2,tis} to 90 mmHg
throughout the tissue volume, inhibiting oxygen flux across the vessel wall by
setting *K _{w}*=0, and increasing oxygen consumption. Our simulation result (not shown)
demonstrates a linear decrease in oxygen concentration in the tissue with time, as we
expected.

#### 3.2 Application to a microvessel network

After validating the steady state and dynamic solutions of our oxygen
advection-diffusion solver, we proceed in solving for the delivery of oxygen to
tissue by a realistic microvessel network. We obtain a volumetric image of the
microvessel network from the somatosensory cortex of a rat using two-photon
microscopy as described in the methods. The volumetric image spanning
230×230×450 µm^{3} is shown in Fig. 4(a). In this image, we identify a 25 µm in
diameter surface pial arteriole in the upper right delivering oxygenated blood to the
tissue and a draining ascending venule in the upper left of the image, connected by
the intervening capillary network.

We manually graphed this vessel network in under 2 hours using custom software to accelerate the process, by placing node points and connecting edges along the vessels. The diameter of the vessel at each node point, the blood pressure, and velocity of blood was calculated as described in the methods. The velocity at each node in the vascular graph was used in solving the oxygen advection diffusion equation to obtain the partial pressure of oxygen at every node point in the vessels and throughout the tissue. The vessel in-flow and out-flow points (i.e., the boundary points as mentioned in Section 2.1) were manually labeled and used for the subsequent advection modeling. The generated 3D mesh for this case contains 138685 nodes and 607406 elements. The total time for mesh generation, calculating the matrices and solving for each time step are 136, 62 and 6 seconds, respectively.

In Fig. 4(b) we plot the graphed vascular network identifying the arterioles (red), capillaries (green), and venules (blue). We see that the surface pial arteriole branches twice. The left-hand branch propagates along the surface to the lower left of the image and then descends 450 µm to the bottom of the image without branching. The right-hand branch descends into the tissue at the lower right. This descending arteriole quickly branches a few times leading to a capillary network that connects with an ascending venule on the upper left of the image. The velocity of blood along the vascular network is shown in Fig. 4(c). The velocity has a maximum of 10 mm/s in the pial arteriole and reaches a minimum of ~1 mm/s in the capillaries and draining venules. The partial pressure of oxygen along the network is graphed in Fig. 4(d), showing the expected decrease from feeding arteriole to draining venule.

The median velocity, blood pressure, and partial pressure of oxygen versus vessel diameter are shown in Fig. 5. Our pressure distribution compares well with a survey of experimental data [33]. Our velocity and partial pressure of oxygen, however, drop below experimental data [33, 34] in the capillaries and venules. This is likely a result of our incomplete vascular graph not accounting for the surrounding microvessels that are providing more blood into the draining vein and delivering more oxygen to the tissue. We will correct this in the future by considering a vascular graph over a larger volume of tissue in order to obtain a more completely connected graph in the central regions.

The partial pressure of oxygen within the tissue is shown in Fig. 6 in slices at different depths and in a movie of the
volumetric data. From these data, we see the high tissue PO_{2} around the
arterioles dropping off rapidly with distance, while the lower PO_{2} around
the capillaries is much more spatially uniform. In Fig. 7 we plot the drop in PO_{2} versus distance from selected
arterioles, capillaries, and venules. These results qualitatively agree with
experimental *in-vivo* measurements [34].

## 4. Summary

We have described a hybrid numerical algorithm and a general work-flow to model the 3D
time-varying oxygen advection diffusion process in an anatomical microvascular network.
This framework provides a quantitative and computationally feasible approach for
modeling the dynamic processes in the microvessels in the tissue, which is becoming
increasingly important in studies of cerebro-vascular physiology and tumor physiology.
The proposed algorithm is structurally generic and readily applicable to vessel networks
with varying degrees of complexity. The graph representations of the vessel network and
finite-element modeling also provide a flexible spatial discretization scheme, which is
important to balance computational expense and accuracy. Solving the 3D dynamic
diffusion-advection process not only makes this framework useful for modeling tissue
oxygen delivery and consumption, but is also extendable to include other molecules such
as NO and CO_{2}, or to model the pharmacokinetic processes in drug delivery.
Future efforts for this framework include automatic graphing of the vessel network from
3D microscopic images and an in-depth understanding of the stability condition. We will
also apply this method to full-sized vessel networks to characterize the microscopic
details of oxygen delivery during brain activation and to validate macroscopic NIRS and
functional magnetic resonance imaging (fMRI) measurements of oxygen delivery.

The authors would like to acknowledge funding support from NIH R01NS057476, P01NS055104
and R01NS051188, and from the Air Force Office of Scientific Research, Medical Free
Electron Laser Program Contract FA9550-07-1-0101. The authors would also like to
acknowledge the assistance of Shuai Yuan, Mark Shalinsky, and Weicheng Wu for *in
vivo* experiments, and the help from Wes Turner (Kitware Inc.) and Scott
Dineen (OSA) for preparing the interactive datasets.

## References and links

**1. **M. Cope and D. T. Delpy, “System for long-term measurement of cerebral
blood flow and tissue oxygenation on newborn infants by infra-red
transillumination,” Med. Biol. Eng.
Comput. **26**, 289–294
(1988). [CrossRef] [PubMed]

**2. **A. Villringer and B. Chance, “Non-invasive optical spectroscopy and
imaging of human brain function,” Trends
Neurosci. **20**, 435–442
(1997). [CrossRef] [PubMed]

**3. **A. P. Gibson, J. C. Hebden, and S. R. Arridge, “Recent advances in diffuse optical
imaging,” Phys. Med. Biol. **50**, R1–R43
(2005). [CrossRef] [PubMed]

**4. **A. Grinvald, E. Lieke, R. D. Frostig, C. D. Gilbert, and T. N. Wiesel, “Functional architecture of cortex revealed
by optical imaging of intrinsic signals,”
Nature **324**, 361–364
(1986). [CrossRef] [PubMed]

**5. **J. D. Briers, “Laser Doppler, speckle and related
techniques for blood perfusion mapping and imaging,”
Physiol. Meas. **22**, R35–R66
(2001). [CrossRef]

**6. **A. K. Dunn, H. Bolay, M. A. Moskowitz, and D. A. Boas, “Dynamic imaging of cerebral blood flow using
laser speckle,” J. Cereb. Blood Flow
Metab. **21**, 195–201
(2001). [CrossRef] [PubMed]

**7. **A. Villringer, A. Them, U. Lindauer, K. Einhaupl, and U. Dirnagl, “Capillary perfusion of the rat brain
cortex,” Circ. Res. **75**, 55–62
(1994). [PubMed]

**8. **D. Kleinfeld, P. P. Mitra, F. Helmchen, and W. Denk, “Fluctuations and stimulus-induced changes in
blood flow observed in individual capillaries in layers 2 through 4 of rat
neocortex,” Proc. Natl. Acad. Sci. USA **95**, 15741–15746
(1998). [CrossRef] [PubMed]

**9. **R. L. Plant and D. H. Burns, “Quantitative, depth-resolved imaging of
oxygen concentration by phosphorescence lifetime
measurement,” Appl. Spectrosc. **47**, 1594–1599
(1993). [CrossRef]

**10. **I. Itzkan, L. Qiu, H. Fang, M. M. Fang, E. Zaman, I. C. Vitkin, S. Ghiran, M. Salahuddin, C. Modell, L. M. Andersson, P. B. Kimerer, K. H. Cipolloni, S. D. Lim, I. Freedman, B. P. Bigio, E. B. Sachs, L. T. Hanlon, and Perelman, “Confocal light absorption and scattering
spectroscopic microscopy monitors organelles in live cells with no exogenous
labels,” Proc. Natl. Acad. Sci. USA **104**, 17255–17260
(2007). [CrossRef] [PubMed]

**11. **A. D. Estrada, A. Ponticorvo, T. N. Ford, and A. K. Dunn, “Microvascular oxygen quantification using
two-photon microscopy,” Opt. Lett. **33**, 1038–1040
(2008). [CrossRef] [PubMed]

**12. **Y. Wang, B. A. Bower, J. A. Izatt, O. Tan, and D. Huang, “In vivo total retinal blood flow measurement
by Fourier domain Doppler optical coherence tomography,”
J. Biomed. Opt. **12**, 041215 (2007). [CrossRef] [PubMed]

**13. **B. Cense, T. C. Chen, N. Nassif, M. C. Pierce, S. H. Yun, B. H. Park, B. E. Bouma, G. J. Tearney, and J. F. de Boer, “Ultra-high speed and ultra-high resolution
spectral-domain optical coherence tomography and optical Doppler tomography in
ophthalmology,” Bull. Soc. Belg.
Ophtalmol. 123–132 (2006).

**14. **R. K. Wang, S. Jacques, Z. Ma, S. Hurst, S. Hanson, and A. Gruber, “ Three dimensional optical
angiography,” Opt. Express **15**, 4083–4097
(2007). [CrossRef] [PubMed]

**15. **H. S. Orbach, L. B. Cohen, and A. Grinvald, “Optical mapping of electrical activity in
rat somatosensory and visual cortex,” J.
Neurosci. **5**, 1886–1895
(1985). [PubMed]

**16. **B. J. Baker, E. K. Kosmidis, D. Vucinic, C. X. Falk, L. B. Cohen, M. Djurisic, and D. Zecevic, “Imaging brain activity with voltage- and
calcium-sensitive dyes,” Cell. Mol.
Neurobiol. **25**, 245–282
(2005). [CrossRef] [PubMed]

**17. **B. Chance, N. Graham, and D. Mayer, “A time sharing fluorometer for the readout
of intracellular oxidation-reduction states of NADH and
flavoprotein,” Rev. Sci. Instrum. **42**, 951–957
(1971). [CrossRef] [PubMed]

**18. **K. A. Kasischke, H. D. Vishwasrao, P. J. Fisher, W. R. Zipfel, and W. W. Webb, “Neural activity triggers neuronal oxidative
metabolism followed by astrocytic glycolysis,”
Science **305**, 99–103
(2004). [CrossRef] [PubMed]

**19. **C. E. Elwell, J. R. Henty, T. S. Leung, T. Austin, J. H. Meek, D. T. Delpy, and J. S. Wyatt, “Measurement of CMRO2 in neonates undergoing
intensive care using near infrared spectroscopy,”
Adv. Exp. Med. Biol. **566**, 263–268
(2005). [CrossRef]

**20. **Y. Zheng, D. Johnston, J. Berwick, D. Chen, S. Billings, and J. Mayhew, “A three-compartment model of the hemodynamic
response and oxygen delivery to brain,”
NeuroImage **28**, 925–939
(2005). [CrossRef] [PubMed]

**21. **T. J. Huppert, M. S. Allen, H. Benav, P. B. Jones, and D. A. Boas, “A multicompartment vascular model for
inferring baseline and functional changes in cerebral oxygen metabolism and
arterial dilation,” J. Cereb. Blood Flow
Metab. **27**, 1262–1279
(2007). [CrossRef] [PubMed]

**22. **T. J. Huppert, M. S. Allen, G. S. Diamond, and D. A. Boas, “Inferring cerebral oxygen metabolism from
fMRI with a dynamic multi-compartment Windkessel model,”
Human Brain Mapping (to be published).

**23. **M. A. Franceschini, S. Thaker, G. Themelis, K. K. Krishnamoorthy, H. Bortfeld, S. G. Diamond, D. A. Boas, K. Arvin, and P. E. Grant, “Assessment of infant brain development with
frequency-domain near-infrared spectroscopy,”
Pediatr. Res. **61**, 546–551
(2007). [PubMed]

**24. **D. A. Beard and J. B. Bassingthwaighte, “Modeling advection and diffusion of oxygen
in complex vascular networks,” Ann. Biomed.
Eng.. **29**, 298–310
(2001). [CrossRef] [PubMed]

**25. **D. D. Lobdell, “An invertible simple equation for
computation of blood O2 dissociation relations,”
J. Appl. Physiol. **50**, 971–973
(1981). [PubMed]

**26. **D. R. Lynch, *Computational Partial Differential Equations for
Environmental Scientists and Engineers — A First Practical Course*
(Springer, 2005). [PubMed]

**27. **C. Godsil and G. Royle, *Algebraic Graph Theory*
(Springer-Verlag, 2001). [CrossRef]

**28. **A. S. Popel, R. N. Pittman, and M. L. Ellsworth, “Rate of oxygen loss from arterioles is an
order of magnitude higher than expected.” Am. J.
Physiol. Heart Circ. Physiol. **256**, H921–H924
(1989).

**29. **M. A. Haidekker, A. G. Tsai, T. Brady, H. Y. Stevens, J. A. Frangos, E. Theodorakis, and M. Intaglietta, “A novel approach to blood plasma viscosity
measurement using fluorescent molecular rotors,”
Am. J. Physiol. Heart Circ. Physiol. **282**, H1609–H1614.
(2002). [PubMed]

**30. **D. A. Boas, S. R. Jones, A. Devor, T. J. Huppert, and A. M. Dale, “A vascular anatomical network model of the
spatio-temporal response to brain activation,”
NeuroImage **40**, 1116–1129
(2008). [CrossRef] [PubMed]

**31. **P. Herman, H. K. Trubel, and F. Hyder, “A multiparametric assessment of oxygen
efflux from the brain,” J. Cereb. Blood Flow
Metab. **26**, 79–91
(2006). [CrossRef]

**32. **G. Arfken, *Mathematical Methods for Physicists*, 3rd
ed. (Academic,
1985).

**33. **H. H. Lipowsky, “Microvascular rheology and
hemodynamics,” Microcirculation **12**, 5–15
(2005). [CrossRef] [PubMed]

**34. **E. Vovenko, “Distribution of oxygen tension on the
surface of arterioles, capillaries and venules of brain cortex and in tissue in
normoxia: an experimental study on rats,” Pflugers
Arch. **437**, 617–623
(1999). [CrossRef] [PubMed]