In biophotonics, the light absorption in a tissue is usually modeled by the Helmholtz equation with two constant parameters, the scattering coefficient and the absorption coefficient. This classic approximation of “haemoglobin diluted everywhere” (constant absorption coefficient) corresponds to the classical homogenization approach. The paper discusses the limitations of this approach. The scattering coefficient is supposed to be constant (equal to one) while the absorption coefficient is equal to zero everywhere except for a periodic set of thin parallel strips simulating the blood vessels, where it is a large parameter The problem contains two other parameters which are small: , the ratio of the distance between the axes of vessels to the characteristic macroscopic size, and , the ratio of the thickness of thin vessels and the period. We construct asymptotic expansion in two cases: and and prove that in the first case the classical homogenization (averaging) of the differential equation is true while in the second case it is wrong. This result may be applied in the biomedical optics, for instance, in the modeling of the skin and cosmetics.
Citation: Mottin S, Panasenko G, Ganesh SS (2010) Multiscale Modeling of Light Absorption in Tissues: Limitations of Classical Homogenization Approach. PLoS ONE 5(12): e14350. https://doi.org/10.1371/journal.pone.0014350
Editor: Giuseppe Chirico, University of Milano-Bicocca, Italy
Received: December 2, 2009; Accepted: October 4, 2010; Published: December 31, 2010
Copyright: © 2010 Mottin et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The results were supported by the Program Emergence of the Region Rhone-Alpes. The theoretical part was supported by the CNRS postdoctoral grant and by the grant of the Russian Federal Agency on Research and Innovations Contract No 02.740.11.5091 “Multiscale Models in Physics, Biology and Technologies: Asymptotic and Numerical Analysis”. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Physical background of the problem
In the present paper we consider the Helmholtz equation with rapidly oscillating large potential with a periodic support having a small measure of its intersection with a period. This absorption coefficient of the potential (more precisely, it is the ratio between the absorption coefficient and the diffusion coefficient) depends on three small parameters: is a standard homogenization parameter that is the ratio of the period of the potential and the characteristic macroscopic size; is the ratio between the measure of the intersection of the support of and the period; is a small parameter standing for the inverse of the ratio of the maximal value of the coefficient and the diffusion coefficient multiplied by the square of the characteristic macroscopic size of the problem (we will consider the case when takes only two values: and ).
The Helmholtz equation(1)is considered below as a model of the light absorption in tissues under hypothesis that this absorption takes place only in the set of parallel thin blood vessels (where ) and this absorption is ignored outside of these vessels.
The linear dimensions of the vessels are much smaller than the linear dimensions of the body as a whole. Typically the distance between two neighboring large micro-vessels (of about 15 m in the diameter) is around 180 m in primate cerebral cortex . This distance is also used in 3D simulation of tumor growth and angiogenesis . Assume at the first approximation that the tissue is a nearly periodic structure. Moreover, consider the two-dimensional idealization of this periodic structure that is, the periodic set of the parallel strait narrow identic vessels separated by the homogeneous tissue. Let L be the macroscopic characteristic size and let be the distance between two neighboring vessels (strips). It means that the parameter stands here for the ratio of the distance between two neighboring vessels and the characteristic macroscopic size . It is assumed throughout that this ratio is a small parameter. Indeed, if the characteristic macroscopic size (L) is equal to 10 mm (it is a typical value of the diffuse optical tomography ) and the distance between the vessels is equal to 0.18 mm, then . The second parameter of the model is the ratio of the thickness of the vessels and the distance between neighboring vessels denoted . This as well is supposed to be a small parameter. Thus, in the discussed above structure we have . In the spectral window 500 nm–700 nm, the oxyhaemoglobin extinction coefficient shows a wide dynamic from 275 (mole/L) cm at 690 nm (the lowest) and 55 500 (mole/L) cm at 575 nm . In order to quantify the maximum variation we compare arterial blood (90 percent saturated) with pure water at 690 nm and 575 nm then the ratio of the absorption coefficient is respectively 300 and 400000. At 700 nm the absorption of tissue (without blood) is about ten times less than that of the pure water . In the visible window the maximum of this ratio can be more than 10000. That is why in the idealized model we consider the case when the absorption coefficient is equal to zero out of vessels. As we have mentioned above the light absorption process is described here by the Helmholtz equation (1), where is just non-dimensionalized absorption coefficient, equal to zero out of vessels and equal to the great (dimensionless) parameter within the vessels. Let us remind that the physical sense of this great parameter is the ratio of the absorption and diffusion effects, more exactly, the ratio of the absorption coefficient and the diffusion coefficient multiplied (i.e. the ratio is multiplied) by the square of the characteristic macroscopic size . Let the haemoglobin concentration be 150 g/liter. Then in order to convert the molar extinction coefficient to the absorption coefficient (in mm), one has to multiply it by 0.00054. At the wavelength 575 nm for the scattering coefficient close to 2.5 mm, is about 23000. Then and In the case of penetrating vessels with the average diameter close to 65 m, at the same wavelength, . Then the case when is greater than one can be found in a spectral window below 600 nm and for vessels of large diameter. In neurophotonics, the highly vascularized pie-matter correspond to the maximal values of . The highly vascularized tumors could be also a special case when is very high.
Models of this “composite medium” are widely considered in biophysics. Let us give a short review of these results. The partial differential equation (PDE) diffusion with absorption is one of the most important in life science, and in particular, in tissue optics . The researchers are interested in two types of results:
- when the optical properties are unknown, and they want to find them using observed measurements, it is the harder well-known inverse problem.
- when the optical properties are known, then they can use many strategies to calculate various quantities of interest (reflectance, transmission, fluence). This is often referred to as the forward problem.
But in all these approaches, the community of research in biophotonics uses the volumetric averaging of the absorption coefficient. This problem is crucial in the domain of in vivo diffuse optical tomography , , and is always present in in vivo optical neuromethods .
Our goal in the present paper is to show that the classical homogenization (averaging) approach to the solution of equation (1) leading to the approximation(2)where is often approximated to the volumic mean value of has some limitations. Indeed, we will show that it is right for some combination of magnitudes of parameters : it can be proved that but it is inapplicable for some other combinations. In this paper we do not construct the expansions for all possible combinations; we only prove the classical homogenization result in the caseand we show that the homogeneous model (2) is inapplicable in the case
It means that the “diluting” of vessels in multiscale modeling of light absorption of tissues may be applied only in justified cases, in particular, in case A. In fact, the situations when the coefficients of equations depend on two or more small parameters and it leads to a fail of classical homogenization are not too surprising now: first results of this type could be found in –. For some cases the non-classical asymptotic expansions were constructed. However, to our knowledge, such examples have not been constructed for rapidly oscillating large potentials with narrow support. The present paper tries to contribute in this important case of biophysical applications.
In order to simplify technical details of the analysis we consider the boundary value problem for the Helmholtz equation set in a layer with the Dirichlet conditions on the boundary of the layer; we assume that the right-hand sides of the equation and of the boundary condition do not depend on . In this case we may seek a solution independent of as well. The Helmholtz equation takes the following form:(3)where stands for , is a small positive parameter, such that, is an integer.
We consider a boundary condition corresponding to a constant solution in the case of the constant absorption coefficient and of a constant right-hand side . Then . So, these boundary conditions are:(4)
Similar problems were addressed by many authors –; however the potentials considered in these works are typically or . In this work, there are three parameters and we study all relevant asymptotic regimes. We follow the ideas and methods of –, ,  in the sense that we construct asymptotic expansions for and analyze all the important asymptotic regimes of the parameters .
Mention that the Helmholtz equation (1) (or (3)) is not a perfect model for the light absorption process: it is not more than an approximation of a more adequate model based on the radiation transfer equation . Indeed, one can introduce the average (over all directions ) for the diffused radiation intensity
In the case of the described above plane geometry the diffusion approximation of the radiation transfer has a form where and are the coefficients corresponding to the light absorption and dispersion respectively, is the blood heat radiation intensity and is the external radiation intensity, is the thickness of the irradiated skin strip, is the dimension factor. Introducing the new variable corresponding to the relative optical depth of the radiation processwe get for (here ):where and functions depend on , solution of equation
The choice of the boundary conditions (4) is not too important for our analysis: the asymptotic approach developed below can be applied in the case of other conditions, for instance,corresponding to a constant solution for equation (3) with replaced by its volumic mean oror the periodicity conditions (mention that the diffusion approximation of the radiation transfer may loose its precision near the boundary).
Mathematical statement of the problem
Let be smooth enough (this assertion will be formulated more precisely later). We are interested in studying the asymptotic behaviour as , , and of solving the following boundary value problem:(6)(7)Note that we have the following a priori estimate for solution of equation (6) with boundary condition (7) replaced by the homogeneous oneProposition 0.1. There exists a constant independent of small parameters such thatHere may be taken equal to . A well-known imbedding theorem gives:
Returning now to equation (6) with non-homogeneous boundary condition (7) we can present the solution as a sum of the solution of equation (6) with homogeneous boundary condition and the solution of the homogeneous equation (6) with non-homogeneous boundary condition (7). Applying the maximum principle argument, we get:
Proposition 0.2. There exists a constant independent of small parameters such that solution of problem (6)-(7) satisfies:Remark 0.1. If boundary condition (7) is replaced by another one:with some numbers and then the estimate of Proposition 0.2 takes form:
If , then the estimate is proved. If , then the sign of is constant on the whole interval ; consider the case Then is increasing and and so,and we get (8). In the same way we consider the opposite case when
Asymptotic expansion in the case
Consider the -th level approximation of given by an ansatz from :(9)where are 1-periodic functions, , and is a smooth function which will be sought in a form(10)Substituting the expression for from (9) in (6) we obtain(11)where(12)and is given by(13)In writing expressions (11), (12) we followed the convention that if at least one of the two indices or , is negative.
We choose being a constant such that the equation (12) has a solution, i.e.(14)where Also, periodic solutions of equation(15)are chosen to be of average zero for all and such that, ; is chosen equal to . We get: , , for all , is a 1-periodic solution of equation
We solve the family of equations (15) by determining for each fixed the solutions for all . Due to the form of the equation (15) the behavior of and will be similar.
With this background, we now state the principal result concerning the 1-periodic functions .
Lemma 0.1. Let . For each (of the form or ) and ,
- Functions are piecewise polynomials on with respect to the partition .
- There exists a constant independent of the small parameters such that, for , , , and .
Proof. The proof is by induction on .
Step (i). Let . That is, . We get: , , satisfy equations (15) with the last two terms equal to zero; so , for all , for all and the assertion of lemma for is evident.
Step (ii). Let and let all , and with be bounded by , where stands for the integer part of , and is a constant independent of the small parameters. Consider equation (15) for . For it takes formwith the piecewise polynomial right hand side; moreover, the primitive (with vanishing mean value) of this right hand side is also bounded by the same order. That is why, integrating this equation twice and keeping every time the vanishing mean for the primitive, we prove the assertion of lemma for . In the same way we prove it for .
Now we apply again (15), express from this equation and prove by induction on that all , , the right hand sides and so, finally, and are bounded by . The lemma is proved.
Define now functions from boundary value problems:(24)and(25)(26)Here is supposed to be greater than . These equations have a formwhere are some real numbers and is a smooth function. In the same way as in the first section we prove thatand so,where the constants and are independent of small parameters.
Remark 0.2. Functions depend on small parameters but their asymptotic expansion can be constructed by classical boundary layer technique (see ). For instance, if , then has a form:
In the same way we construct by induction the expansions of functions , . Assume that , . Consider the right hand side having the form of a linear combination with some bounded coefficients of functions having a form:where are times differentiable functions independent of small parameters and , such that, there exist constants independent of small parameters satisfying for any real inequalities
The right hand sides of the boundary conditions as well have a similar form: they are some linear combinations with bounded coefficients of constants and having a form:where are independent of small parameters.
The expansion of has a similar form of a linear combination with bounded coefficients of functionswhere satisfy equations (functions with the negative indices are equal to zero)and exponentially decaying at infinity functions satisfy equationsand boundary conditionsexponentially decaying at infinity functions satisfy equationsand boundary conditionsand of the exponents
Applying the induction on , the explicit expressions for the right hand sides of the problems (24),(25),(26) for and the estimates of Lemma 0.1 we prove that there exist constants independent of small parameters, such that,(27)
So, from the triangle inequality we getTheorem 0.1. Assume that there exists a positive real such that, is bounded by a constant independent of small parameters. Let belong to . Then there exists a constant independent of small parameters such that
Consider now the very first term of the expansion and we get:
Mention that the error of order is much smaller than that is of order , and so the relative error of the approximation of the exact solution by is small.
Remark 0.3. Here we have considered the case when stands for a large parameter tending to the infinity. The case when is a positive finite constant may be considered by a classical technique  and the estimate of Corollary 0.3 becomes
Asymptotic analysis: case
Here we will construct an example where the behavior of the solution is completely different from the behavior described in the previous section. In particular, the solution of the problem is different from the behavior of the solution of the homogenized equation. For the sake of simplicity consider the right hand side . Consider an auxiliary function defined on the interval as a function from independent of small parameters such that it equals to zero on the interval and it equals to one on . Let us keep the same notation for the 2-periodic extension of this function on . Consider an approximation for having a form:(29)where and extend it periodically to . By a simple calculation we find that satisfies the boundary conditions exactly and the equation with a discrepancy of order , i.e. . Assume that there exists a positive real such that, . Then for any positive , .
Mention that for the right hand side of this bound is much smaller than the values of approximation , and so the relative error of the approximation of the exact solution by is small.
Results and Discussion
The main observations on the asymptotic analysis
We record some observations in the following remark.
In the case (A), the leading term is equal to the solution of the homogenized equation (2), that is the constant plus two exponents rapidly decaying from the boundary: their contribution can be neglected at the distance of order of . In the case (B) the approximate solution is completely different: it is periodic piecewise quadratic function of order except for the small intervals where the potential is large. That is why the homogeneous model (2) is inapplicable in this case.
Thus, the homogenized model (2) may be applied only in the case , when it is justified theoretically. The homogenized model may be inapplicable in the case if is not small.
On the effective absorption coefficient, the volumetric mean and the total absorption coefficient
Biological tissues are highly heterogeneous media at the microscopic scale. For years, the patterns of the mammalian cortex micro-angioarchitecture have attracted the interest of many research groups , –. At different scales the blood vessels have very different physical characteristics. Moreover the absorption coefficient is great in the vessels and small out of the vessels in UV and visible parts of the spectrum. At the end they induce multiscale complex photon propagation and absorption. On the other hand the experimental data are obtained mainly at the macroscopic scale, and so one of the main questions is how to make a passage from the micro-scale to the macro-scale. The main mathematical tool of such passage from one scale to another (the up-scaling) is the homogenization theory, see , , ,  and the references there. Normally, the microscopic description of the heterogeneous medium can be replaced by an equation with constant coefficients. This equation is called the homogenized equation and the constant coefficients are effective coefficients. This homogeneous approximation for the heterogeneous medium is justified if the solutions of both models are close in some norms. In some cases a heterogeneous medium cannot be approximated by a homogeneous one , , , . Then the notion of an effective coefficient cannot be introduced in the above sense.
In particular, in the present paper we prove that if product is small then the homogenized model (2) is justified and the effective absorption coefficient is equal to the volumetric mean value of function . The smallness of the value of this product means that the relative error of the approximation of the heterogeneous medium by the homogeneous one is of order of this product. The effective absorption coefficient is an important quantity because the detailed knowledge of the tissue macroscopic optical properties is essential for an optimization of optical methods i.e. for modeling the color of skin, of port-wine stains and for tumor detection; it helps to adapt an appropriate photodynamic therapy, in particular, some laser treatment.
On the other hand the present paper shows the limitations of the homogenized models for the absorption problems. Indeed, if product is not small then the homogenized model (2) as well as other homogenized models (with other possible constant values of the effective absorption coefficient) do not approximate the initial microscale model, because the solution of problem (2) does not oscillate for any choice of the effective absorption coefficient, while the solution of equation (6) with boundary condition (7) rapidly oscillates (see (29)).
This theoretical argument is confirmed by some physical reasons and by experimental observations. Many authors , , ,  show that the assumption of a homogeneous distribution of blood in the tissue may strongly overestimate the total blood absorption when absorption is high and/or the vessels have sufficient diameter. For large vessels less of light reaches the center of the vessel, and the absorbers in the center of a vessel contribute less and less to the total attenuation of the light. However, the total blood absorption is an important qualitative characteristic of the absorption process and so it should be calculated with a great precision. Let us apply the above asymptotic analysis provided for equation (6) with boundary condition (7) and calculate the total blood absorption in cases A and B. Let us define the total absorption coefficient as a ratio of the integrals
For the particular case of a constant coefficient we get evidently . In the case A substituting the leading term of an asymptotic expansion we get that the asymptotic behavior of is given by an approximate formula , i.e. it is close to the volumetric mean of that is, . In the case B the value of this total absorption coefficient is much less than the volumetric mean (that was observed in the discussed above experiments): the direct computations for the approximation (29) show that with
The authors of papers , , ,  characterize the fall of the total absorption coefficient by the correction factor and discuss the situations when this factor is different from 1. In our asymptotic analysis we see that in the case A but in the case B .
- 1. Risser L, Plouraboue F, Cloetens P, Fonta C (2009) A 3D-investigation shows that angiogenesis in primate cerebral cortex mainly occurs at capillary level. International Journal of Developmental Neuroscience 27: 185–196.
- 2. Shirinifard A, Gens JS, Zaitlen BL, Poplawski NJ, Swat M, et al. (2009) 3D multi-cell simulation of tumor growth and angiogenesis. PLoS ONE 4(10): e7190.
- 3. Vignal C, Boumans T, Montcel B, Ramstein S, Verhoye M, et al. (2008) Measuring brain hemodynamic changes in a songbird: responses to hypercapnia measured with functional MRI and near-infrared spectroscopy. Phys Med Biol 53: 2457–2470.
- 4. http://omlcogiedu/spectra/hemoglobin/.
- 5. Ramstein S, Vignal C, Mathevon N, Mottin S (2005) In-vivo and non-invasive measurement of songbird head's optical properties. Applied Optics 4: 6197–6204.
- 6. Arridge SR, Cope M, Delpy DT (1992) The theoretical basis for the determination of optical pathlengths in tissue - temporal and frequency-analysis. Physics in Medicine and Biology 37: 1531–1560.
- 7. Gibson AP, Austin T, Everdell NL, Schweiger M, Arridge SR, et al. (2006) Three-dimensional whole-head optical passive motor evoked responses in the tomography of neonate. Neuroimage 30: 521–528.
- 8. Mottin S, Laporte P, Cespuglio R (2003) The inhibition of brain NADH oxidation by chloramphenicol in the freely moving rat measured by picosecond time-resolved emission spectroscopy. J Neurochem 84: 633–642.
- 9. Liu H, Chance B, Hielscher AH, Jacques SL, Tittel FK (1995) Influence of blood-vessels on the measurement of hemoglobin oxygenation as determined by time-resolved reflectance spectroscopy. Medical Physics 22: 1209–1217.
- 10. Svaasand L, Fiskerstrand EJ, Kopstad G, Norvang LT, Svaasand EK, et al. (1995) Therapeutic response during pulsed laser treatment of port-wine stains: dependence on vessel diameter and depth in dermis. Lasers in Medical Science 10: 235–243.
- 11. Firbank M, Okada E, Delpy DT (1997) Investigation of the effect of discrete absorbers upon the measurement of blood volume with near-infrared spectroscopy. Physics in Medicine and Biology 42: 465–477.
- 12. Verkruysse W, Lucassen GW, de Boer JF, Smithies D, Nelson JS, et al. (1997) Modelling light distributions of homogeneous versus discrete absorbers in light irradiated turbid media. Physics in Medicine and Biology 42: 51–65.
- 13. Bradu A, Sablong R, Julien C, Tropres I, Payen JF, et al. (2001) In vivo absorption spectroscopy in brain using small optical fiber probes: effect of blood confinement. In: Papazoglou TG, Wagnieres GA, editors. pp. 85–90. 2001 Jun 19-21; Munich, Germany Spie-Int Soc Optical Engineering.
- 14. van Veen RLP, Verkruysse W, Sterenborg H (2002) Diffuse-reflectance spectroscopy from 500 to 1060 nm by correction for inhomogeneously distributed absorbers. Optics Letters 27: 246–248.
- 15. Bakhvalov N, Panasenko G (1989) Homogenisation: averaging processes in periodic media. Mathematical problems in the mechanics of composite materials. Mathematics and its Applications (Soviet Series): Kluwer, Dordrecht/London/Boston.
- 16. Panasenko G (1990) Multicomponent averaging of processes in strongly inhomogeneous structures. (in Russian) Mat Sb 181: 134–142; translation in Math. USSR-Sb. (1991) 69(1): 143–153.
- 17. Panasenko G (1995) Multicomponent homogenization of the vibration problem for incompressible media with heavy and rigid inclusions. C R Acad Sci Paris Série I Math 321(8): 1109–1114.
- 18. Allaire G, Capdeboscq Y, Piatnitski A, Siess V, Vanninathan M (2004) Homogenization of periodic systems with large potentials. Archive for Rational Mechanics and Analysis 174: 179–220.
- 19. Bensoussan A, Lions JL, Papanicolaou G (1978) Asymptotic analysis for periodic structures. Studies in Mathematics and its Applications: North Holland, Amsterdam.
- 20. Campillo F, Kleptsyna M, Piatnitski A (2001) Homogenization of random parabolic operator with large potential. Stochastic Processes and Their Applications 93: 57–85.
- 21. Fleming WH, Sheu SJ (1997) Asymptotics for the principal eigenvalue and eigenfunction of a nearly first-order operator with large potential. Annals of Probability 25: 1953–1994.
- 22. Weinstein MI, Keller JB (1985) Hills equation with a large potential. SIAM Journal on Applied Mathematics 45: 200–214.
- 23. Panasenko G (2005) Multiscale Modelling for Structures and Composites. Dordrecht: Springer.
- 24. Panasenko G (2008) Homogenization for periodic media: from microscale to macroscale. Physics of Atomic Nuclei 4: 681–694.
- 25. Case KM, Zweifel PF (1967) Linear transport theory. Reading, MA: Addison-Wesley.
- 26. Bar T (1980) The vascular system of the cerebral cortex. Adv Anat Embryol Cell Biol 59: 1–65.
- 27. Duvernoy HM, Delon S, Vannson JL (1981) Cortical blood-vessels of the human-brain. Brain Research Bulletin 7: 519–579.
- 28. Cassot F, Lauwers F, Fouard C, Prohaska S, Lauwers-Cances V (2006) A novel three-dimensional computer-assisted method for a quantitative study of microvascular networks of the human cerebral cortex. Microcirculation 13: 1–18.
- 29. Talsma A, Chance B, Graaff R (2001) Corrections for inhomogeneities in biological tissue caused by blood vessels. Journal of the Optical Society of America a-Optics Image Science and Vision 18: 932–939.