September 1, 2009 / Vol. 34, No.

17 / OPTICS LETTERS 2679

Accuracy of the Born approximation in calculating

the scattering coefficient of biological
continuous random media
İlker R. Çapoğlu,1,* Jeremy D. Rogers,1 Allen Taflove,2 and Vadim Backman1
1
Biomedical Engineering Department, Northwestern University, 2145 Sheridan Road, Evanston, Illinois 60208, USA
2
Electrical Engineering and Computer Science Department, Northwestern University, 2145 Sheridan Road,
Evanston, Illinois 60208, USA
*Corresponding author: capoglu@ieee.org
Received June 9, 2009; revised July 14, 2009; accepted July 28, 2009;
posted August 7, 2009 (Doc. ID 112526); published August 31, 2009
A rigorous error analysis is presented for the scattering coefficient of biological random continuous media in
the Born (or single-scattering) approximation. The analysis is done in two dimensions (2-D) for simplicity of
numerical computation. Scattering coefficients of various tissue-like random media are numerically calcu-
lated via statistical finite-difference-time-domain analysis. The results are then checked against analytical
formulas for the scattering coefficient in the Born approximation. The validity ranges for the correlation
length and the refractive index fluctuation strength of the medium are clearly identified. These 2-D results
show promise for future 3-D investigations. © 2009 Optical Society of America
OCIS codes: 170.3660, 050.1755.

The scattering coefficient 共␮s兲 of a random medium is sponse in a single simulation. The tremendous effi-
defined as the total scattered power per unit volume ciency offered by FDTD in simulating very complex
of the medium under unit incident parallel-beam media with arbitrary refractive index distributions is
light intensity. It is a commonly used theoretical and an attractive feature for studying biological tissues.
experimental parameter in many scattering prob- Recently, it has become possible to obtain more de-
lems in physics and engineering. It is well known tailed and accurate data on the RI distributions of
that ␮s plays an important role in the multiple- biological tissue [8–10]. The information thus gath-
scattering theory for discrete random media [1,2]. ered can be used to construct a random model for the
For continuous random media, the theoretical tools biological medium and fed as an input to such elec-
are relatively less developed. Nevertheless, for tromagnetic modeling tools as FDTD.
weakly scattering media such as biological tissue, a In the following, we first derive analytical equa-
perturbation (or Born) expansion of the scattered tions for the scattering coefficient ␮s in the Born ap-
wave proves to be valuable. Retaining only the first proximation. We then numerically calculate the scat-
term in this expansion is usually termed the Born (or tering coefficient ␮s of various 2-D continuous
single-scattering) approximation [3]. In this approxi- random media using statistical FDTD analysis and
mation, the scattering coefficient ␮s as defined above compare the results with theoretical values. We de-
becomes an essential element in the multiple- fine a meaningful measure of the “error” in the scat-
scattering theory of a continuous random medium tering coefficient and plot this error for a range of
[4,5]. It is important to keep in mind, however, that random medium parameters.
this approximation is valid only for weak refractive- The normalized RI fluctuation of a medium is de-
index fluctuations. For stronger fluctuations, the fined by ⌬n共¯␳兲 = 共n共¯␳兲 − n0兲 / n0, where n0 is the average
very definition of ␮s (scattered power per unit vol- RI of the medium. In this Letter, we consider a spe-
ume) gradually loses its meaning, as other wave ef- cific random (or stochastic) model for the normalized
fects such as localization take effect [4]. In this Let- RI fluctuation and discuss the implications of this
ter, we investigate the accuracy of the Born model in detail. In this model, the normalized RI fluc-
approximation in calculating ␮s in continuous ran- tuation is represented by a 2-D statistically homoge-
dom media with refractive index (RI) properties re- neous Gaussian random field with correlation
sembling biological tissue. Although we consider 2-D
random media for ease of numerical computation, our
results should be valuable for inferring approximate
Bn共⌬␳兲 = ␴n2
⌬␳
lc
K1 冉 冊
⌬␳
lc
, 共1兲
criteria for the full 3-D case, for which a theoretical
model was presented in [6]. in which ⌬␳ is the distance between two points, ␴n is
For the assessment of theoretical approximations, the fluctuation strength, lc is the correlation length,
one needs a rigorous numerical solution to the light- and K1共 · 兲 is the modified Bessel function of second
scattering problem. One of these numerical tech- kind and order 1. The specific choice for the correla-
niques is the finite-difference time-domain (FDTD) tion function is of comparatively little concern, be-
method [7]. This method relies on the numerical so- cause the accuracy of the Born approximation is more
lution of full-vector Maxwell’s equations directly in influenced by lc and ␴n than by the exact shape of the
the time domain, thereby providing a broadband re- correlation function.
0146-9592/09/172679-3/$15.00 © 2009 Optical Society of America
2680 OPTICS LETTERS / Vol. 34, No. 17 / September 1, 2009

The random sample with correlation (1) is illumi- ues ranging between 1.35 and 1.38 are encountered
nated by a plane wave from direction k̂i, and the ra- in the literature for the average RI (or the cytoplasm
RI), n0, of biological cells [8,10]. Fluctuation
diated scattered field is observed at direction k̂o. If strengths 共␴n兲 ranging from 0.007 to 0.022 are also
the electric-field polarization vector êi of the plane reported in the mentioned studies. In the following
wave lies along the axis of invariance, the excitation examples, we use the common value of n0 = 1.38 for
is TM, or scalar. Otherwise, the excitation is TE, or
the average RI. We consider different values of ␴n
vector. In 3-D space, scattering parameters such as
and lc, and calculate the error in the scattering coef-
the total or differential scattering cross section are
ficient predicted by Eqs. (4) and (5).
defined per unit volume [1]. Since we are concerned
The FDTD model applied in this Letter uses the
with a 2-D geometry, the same parameters are de-
same techniques as described in [7]. This yields a
fined per unit area. The differential scattering cross
computed differential scattering cross section for ca-
section per unit area ␴共k̂o , k̂i兲 is defined as [1] nonical scatterers within ±1 dB of the exact solution
for all scattering angles over a dynamic range exceed-
␴共k̂o,k̂i兲 = 具兩f共k̂o,k̂i兲兩2典/S, 共2兲
ing 50 dB. Specifically, the FDTD simulations are
carried out as follows. A differential area of dimen-
in which f共k̂o , k̂i兲 is the scattered wave amplitude at
sions L ⫻ L is placed in the center of a 2-D FDTD grid
direction k̂o, S is the differential surface area, and with grid spacing ⌬ = 13.3 nm and time step ⌬t
the mean 具·典 is taken over the ensemble of differential = 0.98共⌬ / c兲 / 冑2. The grid is terminated by a convolu-
areas with correlation (1). Using methods similar to tion perfectly matched layer [7] of thickness 266 nm.
those in [1], a closed-form expression for ␴共k̂o , k̂i兲 can A plane wave with TE polarization is sourced into the
be obtained in the Born approximation. For TM (sca- FDTD grid using the total-field/scattered-field ap-
lar) excitation, proach [7]. The results for TM incidence are almost
identical and are omitted for brevity. The electric
2␴n2 k3l2c
␴TM共k̂o,k̂i兲 = , 共3兲 field of the plane wave is a Gaussian-modulated sinu-
共1 + 4k2l2c sin2共␪/2兲兲2 soidal pulse in time, the spectral amplitude of which
is above −20 dB of its maximum between 400 nm and
in which k = 共␻ / c兲n0 is the wavenumber correspond- 700 nm in vacuum, which corresponds to the visible
ing to the average RI of the medium and ␪ is the range of the electromagnetic spectrum. The radiated
angle between k̂i and k̂o. The corresponding result for far field obtained via the near-field-to-far-field trans-
TE excitation is simply Eq. (3) multiplied by a dipole former [7] is normalized by the spectrum of the plane
factor cos2 ␪. Finally, the scattering coefficient ␮s is wave, numerically integrated over all angles, divided
found by integrating the differential scattering cross by the differential area, and finally averaged over
200 realizations of the random medium to obtain the
section per unit area ␴共k̂o , k̂i兲 over 0 ⬍ ␪ ⬍ 2␲, scattering coefficient ␮sTE.
4␴n2 k3l2c 共1 + 2k2l2c 兲␲ In Fig. 1, the normalized scattering coefficient ␮s / k
␮sTM = , 共4兲 is given for two different klc values and a range of
共1 + 4k2l2c 兲3/2 normalized differential-area dimensions, kL. In Figs.
1(a) and 1(b), the correlation lengths are lc = 0.3␭, and
␮sTE = ␴n2 关− 1 − 4k4l4c + 8k6l6c + 冑1 + 4k2l2c + 2k2l2c 共− 3 lc = 3.45␭, respectively, which correspond to cellular-
scale fluctuations in the visible-light spectrum. The
+ 2冑1 + 4k2l2c 兲兴␲/kl2c 共1 + 4k2l2c 兲3/2 . 共5兲 value of ␴n is 0.02 for both figures. The solid horizon-
tal line denotes the ␮s value predicted by the Born
In the following, we test the accuracy of Eqs. (4) and approximation [Eqs. (4) and (5)], and the dashed line
(5) in tissuelike 2-D media by comparison with rigor- denotes the Rayleigh limit [12], where the scattering
ous numerical results obtained using statistical coefficient is proportional to the sample area: ␮s
FDTD analysis.
= ␴n2 k3L2 / 2. It is seen that the Born approximation is
For the statistical FDTD analysis of scattering
from a random medium, averaging is needed over inaccurate at the two extremes of kL: For small kL,
many samples with the correlation specified in Eq. there is not enough spatial averaging for the correla-
(1). The random samples are generated using an tion (1) to take effect. For large kL, multiple scatter-
inverse-Fourier-transform approach, in which the ing effects become dominant. This latter case is sig-
2-D spatial Fourier transform of the random medium nified by decreased mean-free path ls, which is
is generated first and inverse transformed to spatial customarily defined in transport theory as ls = 1 / ␮s.
domain for the final result. The method is based on a When ls becomes comparable with the sample dimen-
straightforward generalization of the principle that sion L, the Born approximation becomes less accu-
the Fourier transform of a stationary random process rate, as seen in the large-kL regions of Figs. 1(a) and
is nonstationary white noise with variance equal to 1(b). In summary, the Born approximation is valid for
the power-spectral density of the random process. sample sizes that are larger than lc (for adequate
Grayscale images of some random samples generated spatial averaging) and smaller than ls (for single
using this method can be found in [11]. scattering.) If no such range for L exists [as in Fig.
The RI distribution of biological media has been 1(b)], the Born approximation ceases to be valid. The
the focus of many experimental investigations. Val- sample size can thus be eliminated from the validity
 September 1, 2009 / Vol. 34, No. 17 / OPTICS LETTERS 2681

Fig. 2. Percentage error caused by the Born approxima-

tion in calculating the TE scattering coefficient ␮sTE.

In this Letter, we have presented a rigorous assess-

ment of the Born approximation in calculating the
scattering coefficient of biological random media us-
ing the FDTD method, which is a full-vector electro-
magnetic simulation tool widely used in the scatter-
ing analysis of complex media. A validity condition
␴n2 共klc兲2 Ⰶ 1 was derived and numerically justified for
the specific correlation function (1). Although the re-
sults are for 2-D random media, they provide valu-
able information regarding the range of validity of
the Born approximation in a general 3-D setting,
which remains to be investigated in a future study.

This work was supported by the National Insti-

tutes of Health (NIH) grants R01EB003682,
Fig. 1. Normalized TE scattering coefficient ␮sTE / k as a R01CA128641, and R01CA112315.
function of the normalized sample dimension kL: (a) small-
error case, (b) large-error case.
References
criterion by writing lc Ⰶ ls = 1 / ␮s. Using Eqs. (4) and 1. A. Ishimaru, Wave Propagation and Scattering in
(5), this can be rewritten as Random Media (Wiley-IEEE, 1999).
2. M. Mishchenko, L. Travis, and A. Lacis, Multiple
␴n2 共klc兲3 Ⰶ 1 for klc Ⰶ 1 Scattering of Light by Particles (Cambridge U. Press,
. 共6兲 2006).
␴n2 共klc兲2 Ⰶ 1 for klc Ⰷ 1
3. M. Born and E. Wolf, Principles of Optics, 7th ed.
As a heuristic measure of the error in the Born ap- (Cambridge U. Press, 1999).
proximation, we consider the minimum percentage 4. L. Ryzhik, G. Papanicolaou, and J. Keller, Wave
difference between the FDTD result and the Born ap- Motion 24, 327 (1996).
proximation in Fig. 1. A third-order polynomial is fit- 5. B. van Tiggelen and H. Stark, Rev. Mod. Phys. 72, 1017
(2000).
ted to the FDTD plot to reduce the error caused by
6. J. D. Rogers, I. R. Capoglu, and V. Backman, Opt. Lett.
spurious minima. In reference to expression (6), and 34, 1891 (2009).
considering the fact that klc ⬎ 1 is frequently satisfied 7. A. Taflove and S. C. Hagness, Computational
in biological tissue, the error is plotted with respect Electrodynamics, 3rd ed. (Artech House, 2005).
to ␴nklc in Fig. 2. The curves are seen to coincide for 8. J. Beuthan, O. Minet, J. Helfmann, M. Herrig, and G.
different values of ␴n and klc, which lends more sup- Muller, Phys. Med. Biol. 41, 369 (1996).
port to the criterion 9. F. Charriere, A. Marian, F. Montfort, J. Kuehn, T.
Colomb, E. Cuche, P. Marquet, and C. Depeursinge,
␴n2 共klc兲2 Ⰶ 1. 共7兲 Opt. Lett. 31, 178 (2006).
10. N. Lue, G. Popescu, T. Ikeda, R. R. Dasari, K.
An immediate consequence of Fig. 2 is that a maxi- Badizadegan, and M. S. Feld, Opt. Lett. 31, 2759
mum 20% error can be attained for the scattering co- (2006).
efficient in the Born approximation if ␴nklc ⬍ 0.1. 11. I. R. Capoglu and V. Backman, presented at the IEEE
Extrapolating the above method to 3-D, it follows AP-S International Symposium, Charleston, S.C., June
from [6] that the corresponding criterion for 3-D is 1–5, 2009.
␴n2 共klc兲2 Ⰶ 1, which is the same as expression (7) for 12. H. C. Van De Hulst, Light Scattering by Small
2-D. Particles (Dover, 1981).