You are on page 1of 5

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 54, NO.

4, AUGUST 2005

1607

Optical Tomography Imaging Based on Higher


Order Born Approximation of Diffuse Photon
Density Waves
Edgar Scherleitner and Bernhard G. Zagar

AbstractIn this paper, we introduce a tomographical system


suitable for imaging absorbers in strongly scattering media using a
single light sourcedetector pair. Their reflective alignment would
make it possible to implement both source and detector into a compact scan-head. Suitable probes would be soft tissue, some strongly
scattering biological probes, or translucent technical materials, for
example. An intensity-modulation technique is utilized to take advantage of the amplitude and delay alterations between the source
and the detector. The propagation of the photons is modeled by the
Helmholtz equation and then appropriate inverse calculations of
several measurements reveal the distribution of the absorption coefficient inside the investigated volume. Therefore a reconstruction
algorithm employing a Born approximation of higher order and
a Tikhonov-regularization scheme were used to solve the deduced
ill-posed and underdetermined system of equations. In this paper,
we present results of simulations as well as of measurements.
Index TermsBorn approximation, diffuse photon density wave,
optical tomography, tomography reconstruction.

I. INTRODUCTION

OMOGRAPHICAL imaging methods are very frequently


used for examinations of human tissue. Much research has
already been done to circumvent the negative effects of ionizing
radiation in the case of X-ray scans and the high expenses for
the hardware and operation of magnetic resonance imaging systems, respectively. One of the alternative methods performing
with visible light is called optical tomography, which can also
be adapted for some biological probes or translucent technical
materials.
Various biological tissues, translucent plastics, and turbid
fluids act strongly scattering but only weakly absorbing on visible light photons. For human tissue examinations, wavelengths
between 600 and 1000 nm are typically chosen [1], [2] to make
use of the optical window in this spectral range. The photons
injected by the source are repeatedly scattered before they
reach the detector; thus, their propagation is random and most
easily modeled by the diffusion approximation of the transport
equation (forward model) [3], [4]. To reconstruct the volume
distribution of the optical properties from several measurements at the boundary of the probe, the corresponding inverse
problem must be solved. The interesting optical parameters are
the absorption coefficient and the scattering coefficient that are

Manuscript received June 15, 2004; revised March 25, 2005.


The authors are with the Institute of Measurement Technology, Johannes Kepler University of Linz, A-4040 Linz, Austria.
Digital Object Identifier 10.1109/TIM.2005.851080

(a)

(b)

Fig. 1. Delineated propagation of intensity-modulated light in strongly


scattering media. (a) U and t are the amplitude and the delay difference,
respectively, for the subsequent inverse problem to be solved. (b) Investigated
semi-infinite volume with a DPDW schematically indicated in the sectional
drawing.

expressed by the complex valued index of refraction in every


discretized voxel of the investigated volume.
In general, there are three different essential methods named
after the form of the source signal: the constant wave method
(the source illuminates the probe with a continuous laser beam)
and the time- and frequency-domain methods. The time-domain
method applies a laser pulse which is typically broadened due
to the scattering behavior of the probe and possibly enclosed inhomogeneities. The frequency-domain method uses a harmonic
signal to modulate the source beam so that an amplitude ratio
and a time delay can be observed between the source and a
signal obtained at a different position of the probe.
The motivation of this research is to develop an inexpensive,
frequency-domain scanning system suitable for various biological probes as well as for some technical materials that fulfill
the optical requirements discussed in detail in Section II by
just using a single source and a single detector transversing
over the surface. The frequency-domain method is employed
because there is less hardware effort and faster calculations are
possible [5].
II. FUNCTIONALITY OF THE IMAGING SYSTEM
The emitted light of a laserdiode at 785 nm is intensity-modulated by a radio-frequency source at several hundred megahertz
and fiber-guided to the surface of the probe. In strongly scattering media, the propagation of such randomly walking photons can be macroscopically represented as an exponentially
damped spherical wave, a so-called diffuse photon density wave
(DPDW); see also Fig. 1. The wavelength of this DPDW is

0018-9456/$20.00 2005 IEEE

1608

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 54, NO. 4, AUGUST 2005

induced by the modulation frequency and the optical parameters like the absorption coefficient and the scattering coefficient.
At each different position, the detector fiber picks up the local
signal and guides it to a high-speed avalanche photodiode; for
more detail, see [6]. For each sourcedetector position, the amplitude ratios and phase differences of these received DPDWs
with respect to the source contain information of possible inhomogeneities in the underlying region. Out of multiple measurements of this kind for various sourcedetector configurations,
a set of equations can be derived to reconstruct slice images of
the investigated volume to determine the position of inhomogeneities. For simplification, our approach is to only use one
source and one detector at a fixed distance, so that they could
easily be integrated in a scan-head. In this paper, we show in
the first instance results obtained by probes with spherical, embedded absorbers.
For the derivation of that set of equations, the propagation of
the DPDW is modeled with the diffusion equation [4] and simplified by certain assumptions, so that inverse calculations can
reveal variations of the absorption inside the observed volume.
The introduced system is compatible with probes that are homogeneously and strongly scattering but only weakly absorbing
the applied light. Absorbing inhomogeneities hidden up to more
than 2 cm (depending on their size) deep under the surface can
be localized. The lack of depth resolution could be improved
to a certain degree by a significant increase of the number of
measurements that would multiply the total effort, which is not
shown in this paper.
III. MATHEMATICAL THEORY
The optical properties of media that are translucent at the generated wavelength of the utilized light source are defined by
and the scattering coefficient ,
the absorption coefficient
respectively. A more illustrative quantity than , however, is
the reduced scattering coefficient , which is reduced to the
isotropically scattering part of , i.e.,
(1)
where
is the expected value of the cosine of scattering
angle .
In this section, we show a mathematical strategy to localize
absorbing spheres in otherwise homogeneously absorbing and
scattering media [1], [4].
Since the propagation of light in sufficiently strong scattering
and only weakly absorbing media can be represented by a diffusion process and the used light is intensity-modulated, the
Helmholtz diffusion equation in the frequency domain is employed [7], [8], [2], which is

with the electromagnetic propagation velocity of light in the


[cm s ], i.e., the speed
medium given by
of light in vacuum divided by the refractive index of the medium
at the chosen wavelength.
is the source term. The
, and the
wavelength of the DPDW is given by
real part of equals

(4)
where
is the diffusion coefficient (
) and for
must be small. For average values of soft
validity of (2)
human tissue, i.e.,
cm ,
cm and
at a modulation frequency of
MHz,
cm.
the wavelength of the DPDW becomes
To solve (2) for deviations from the homogeneous absorption
coefficient, we define
(5)
is the homogeneous part and
is the spatially
where
. According to (3), the
varying, additive component of
square of the wavenumber thus can be written as
(6)
Also, the photon density
is represented as a combination
of a homogeneous part
and a heterogeneous, scattered
, which originates at an existing inhomogeneity
part
at that particexcited by the incident photon density
ular location. One common way to make the problem solvable
is to employ the first-order Born approximation
(7)
at position caused by the
That is, the photon density
source at location can be approximated by the sum over a homogeneous part and a heterogeneous part which is supposed not
to strike any inhomogeneity again before reaching the detector
[1], [9]. Thus it follows from (2) together with (7) when the parameter is omitted for simplification
(8)
Because we only want to detect absorption parameter variations,
and
,
the homogeneous equation, considering
is subtracted from (8) and we are left with
(9)
The first-order Born approximation implies that
(9) becomes

so

(2)
(10)
is the photon density at the currently observed position
in the medium,
is the angular modulation frequency,
is the square of the wavenumber of the DPDW that is short
for

We simply assume that the medium is infinite, instead of semiis a solution of (10)
infinite; then the Green function
[10], i.e.,

(3)

(11)

SCHERLEITNER AND ZAGAR: OPTICAL TOMOGRAPHY IMAGING

1609

where
is the distance of a point from the detector position . In the reconstruction algorithm, however, the medium
was considered semi-infinite by introducing image sources. For
details and the influence of boundaries, see [1] and [8].
There is also an exponential approach, the so-called Rytov
approximation [1], [4], where the results would show a slightly
different sensitivity to absorbing inhomogeneities, but this is not
treated in this paper.
Equation (10) can be solved using the Green function of (11)
to arrive at the first-order approximated, continuous solution for
the heterogeneous part of the photon density
(12)
The investigated volume is mathematically discretized into
voxels so that (12) can be written as a system of linear equations where every measurement gives a row

..
.

..
.

..

..
.

..
.

(13)
which must be solved for the deviations of the absorption coeffor all voxels. is the number of independent
ficient
measurements at the various sourcedetector combinations. The
matrix consists of weight functions for each voxel and each
sourcedetector combination . These weight functions are
(14)
is the Green function for the distance between
where
is the incident
the voxel and the detector position and
photon density at due to the source at . The voxels are cubic
with sides of length .
result from the measureThe complex values of
ments of amplitude and delay differences. To obtain
in (13), one must still subtract the homogeneous part
from the measurements. This homogeneous part is in general
not exactly known and only varies when the sourcedetector dismust be estimated properly and
tance is changed. So
subtracted unless the sourcedetector distance stays constant, as
is the case in our system. Furthermore, their absolute values are
not required to only obtain contrast images.
To improve the reconstruction by Born approximations of
, the approximation of first order in (12)
higher order
again. The solution
is substituted by
is
is then obtained by iteration. For the initial value
is calculated to be inserted
supposed to be zero. Then
in the next step

(15)
After solving (13) for the
vector,
lated for every voxel to be inserted into (15).

can be calcu-

Fig. 2. Schematic of the volume for simulation with two embedded, absorbing
spheres ( = 0:3 cm) utilized for the simulations. Grayscale slices show the
phase of the DPDW at the current sourcedetector position.

Considering the system (13), it becomes clear that it is large,


ill posed, and underdetermined and can only be solved by appropriate methods [10][12] like singular value decomposition,
regularization procedures, or iterative algorithms, e.g., conjugate gradient implementations. We have chosen the Tikhonovregularization method since it performs better with the iterative
algorithm introduced above.
Because of the ill-posedness of the system, small interferences can result in severe artifacts in the reconstructed images,
so that the tradeoff between the fidelity of the reconstruction and
the restriction by the regularization must be adjusted carefully.

IV. SIMULATION RESULTS


In order to be able to later calibrate the real measurements,
in a first step, a set of simulated data (amplitude and delay differences) for different geometric configurations was generated
using the finite-elements software FEMLAB. Effort was put into
creating data that comply as precisely as possible with the subsequent described measurements (Section V). The discretization,
however, was left to FEMLABs automatic mesh generator .
The goal of our simulations is to find the most efficient geometries of the scan head and to predict the detectability dependent on the size and depth of the absorbers and other parameters
of the system like the modulation frequency and the wavelength
of the utilized light, for example.
In Fig. 2, a schematic of the volume data with two embedded,
spherical absorbers is depicted. The coordinates ( , , ) of
the spheres are (1, 0.5, 0.8) and ( 1, 1, 1.2) and their diameter is 0.3 cm. The optical parameters were set to those of
soft human tissue (see Table I). The simulations calculate the
wavelength of the DPDW (
MHz) between the source

1610

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 54, NO. 4, AUGUST 2005

TABLE I
SET

OF PARAMETERS USED IN THE SIMULATIONS AS WELL AS THE


MEASUREMENTS. THESE VALUES CORRESPOND APPROXIMATELY TO THOSE OF
SOFT HUMAN TISSUE

Fig. 4. Specimen with an acrylic grid of holes to fix the source and detector
fibers at predetermined locations.

Fig. 3. Reconstruction of simulated absorption data obtained by the geometry


shown in Fig. 2. Slice images of the absorption in depths from Z = 0 to 1:6 cm
(bright areas have little absorption). The position of the two spheres is indicated
by an x. The upper left image is forced to be white because no results can be
obtained directly at the boundary.

and the detector slightly incorrect, because the semi-infinity of


the volume was ignored. Instead of the expected value of 28.5
cm directly calculated as described in Section III, the simulation yields 24.3 cm. The wavelength was calculated from the
phase difference between the source and the detector with no
spheres inside the volume; the phase values with embedded absorbers are indicated in the two grayscale slices in Fig. 2. The
mismatch in the wavelengths has no effect on the reconstructed
images since the sourcedetector separation is always held constant at 1.5 cm and thus gives merely a nonvarying deviation
component.
Fig. 3 depicts a fifth-order reconstruction of the simulated
scenario shown in Fig. 2. There the same algorithm as for the
measured data was employed, which discretizes the volume into
voxels of 0.2 cm sides. The two strong absorbers are both detected since the cubic dark areas mean a higher absorption. The
cm is always forced to white since no results
image at
can be obtained directly at the boundary. The sphere in a depth
of 1.2 cm in the upper right edge of each image is much brighter
and hardly visible, but one has to remember that its radius is only
0.3 cm. A problem is the unsatisfactory depth resolution, for excm where none
ample, the spheres are already seen at
of them should be present yet. Due to this reflective alignment
it is, of course, also hardly possible to obtain information on the

volume behind a strong absorber, which leads to defective images in deeper regions. An administrable approach to improve
the depth resolution would probably be to increase the number
of measurements and also to vary the separation between the
source and the detector [10] to obtain different sensitivities regarding the depth. This would, however, again complicate the
scan-head. Moreover, it must be considered that larger distances
between the source and the detector can heavily decrease the
signal-to-noise ratio, which would nullify the advantage of a
better depth resolution immediately.
V. MEASUREMENT RESULTS
For the measurements, a cube formed resin phantom was
molded with sides of 9 cm with an acrylic-glass grid of holes on
the top for fixing the source and detector fibers at predetermined
well-defined locations; see Fig. 4. The optical properties are the
same as summarized in Table I chosen for the simulations. A
perfectly absorbing sphere of 1 cm in diameter has been invisibly hidden inside the cube during production. Its center is at 1.2
cm depth.
Then the grid was fixed somewhere on the surface above the
area of the sphere and a series of measurements was taken,
where the source and the detector fibers were positioned on the
grid at 28 different hole pairs, so that the distance between the
source and the detector was constant at 1.5 cm. The result of a
third-order reconstruction is seen in Fig. 5. Higher order reconstructions would too strongly amplify errors occurring by the
current measurement setup. The coordinates of the sphere relative to the center of the grid resulted in ( 1, 0.5, 1.2), which
were reconstructed fairly well except for the depth or -coordinate, of course, as discussed before.
The injected light was produced by a laserdiode at
nm and a power of 8 mW. For the diameter of
the used fiber of 1 mm, this results in a power density of about
1 W cm , which is even slightly under the allowable limit of

SCHERLEITNER AND ZAGAR: OPTICAL TOMOGRAPHY IMAGING

1611

REFERENCES

Fig. 5. Reconstruction of measurements on a specimen with an embedded


absorbing sphere ( = 1 cm) at the coordinates ( 1, 0.5, 1.2). Slice images
of the absorption in depths from Z = 0 to 1:6 cm with the real position of
the sphere indicated by an x. The first image is forced to be white because no
results can be obtained directly at the boundary.

1.1 W cm for skin at a wavelength of 632.8 nm (norm DIN


EN 608254), which gives a sufficient safety margin.
The signal was picked up by an avalanche photodiode, and
amplitude and delay differences between the source and the detector signal were obtained by heterodyne downconversion, subsequent filtering and further processing in a PC using the software package MATLAB [6].

VI. CONCLUSION AND FUTURE WORK


In this paper, we have presented a system with an optical
scan-head capable of locating absorbers in turbid media like biological tissue as well as some translucent technical materials.
The forward model was evolved by the diffusion approximation.
The solution of the corresponding inverse problem was obtained
by higher order Born approximations. Therefore, we have developed a regularized reconstruction algorithm that worked well
on both simulated data and real measurements. In order to find
a truncation criterion of the iterative inverse calculations, we
are currently working on a finite difference method for the forward problem [13] to be executed after each iteration step, thus
making it possible to reach some desired accuracy of the distribution of both absorption and scattering coefficients.
It is planed to build a prototype of the scan-head described
above moved by a two-dimensional servo motor system to enable an exact and fast positioning of the fiber pair. This will be
tested on a new phantom that also contains weak absorbers and
scatterers, both of specific shapes.

[1] D. Boas, Diffuse photon probes of structural and dynamical properties


of turbid media: Theory and biomedical applications, Ph.D. dissertation, Dept. Biochemistry and Biophysics, Univ. Pennsylvania, Philadelphia, PA, 1996.
[2] R. Gaudette, Constrained reconstruction techniques for diffuse optical
tomography, Ph.D. dissertation, Athinoula A. Martinos Center for
Biomedical Imaging, Charlestown, MA, 2000.
[3] M. S. Patterson, B. Chance, and B. C. Wilson, Time resolved reflectance and transmittance for the noninvasive measurement of tissue
optical properties, Appl. Opt., vol. 28, no. 12, pp. 23312336, 1989.
[4] M. A. OLeary, Imaging with diffuse photon density waves, Ph.D.
dissertation., Dept. Biochemistry and Biophysics, Univ. Pennsylvania,
Philadelphia, PA, 1996.
[5] S. Willmann, Streulichtspektroskopie mit Hilfe von Photonendichtewellen, Ph.D. dissertation, Heinrich-Heine-Universitt Dsseldorf, Dsseldorf, Germany, 1999.
[6] E. Scherleitner and B. G. Zagar, Diffraction tomography adapted to
technical applications, in Proc. XVII IMEKO World Congr., Dubrovnic,
Croatia, 2003, pp. 207212.
[7] V. Ntziachristos, A. H. Hielscher, A. G. Yodh, and B. Chance, Diffuse
optical tomography of highly heterogeneous media, IEEE Trans. Med.
Imag., vol. 20, no. 6, pp. 470478, Jun. 2001.
[8] J. C. J. Paasschens and G. W. t Hooft, Influence of boundaries on the
imaging of objects in turbid media, Opt. Soc. Amer., vol. 15, no. 7, pp.
17971812, 1998.
[9] A. Ishimaru, Wave Propagation and Scattering in Random
Media. New York: IEEE Press, 1997.
[10] R. Gaudette et al., A comparison study of linear reconstruction techniques for diffuse optical tomographic imaging of absorption coefficient, Phys. Med. Biol., vol. 45, pp. 10511070, 2000.
[11] B. W. Pogue, M. S. Patterson, and T. J. Farrell, Forward and inverse calculations for 3-D frequency-domain optical tomography, Proc. SPIE,
vol. 2389, pp. 328339, 1995.
[12] W. H. Press, S. A. Teulkolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in C++. Cambridge, U.K.: Cambridge Univ. Press,
2002.
[13] Y. Yao and Y. Wang, Frequency-domain optical imaging of absorption
and scattering distributions by a Born iterative method, Opt. Soc. Amer.,
vol. 14, pp. 325342, 1997.

Edgar Scherleitner was born in Kirchdorf, Austria,


in 1973. He received the diploma degree in mechatronics from Johannes Kepler University, Linz, Austria, in 1999, where he is currently pursuing the Ph.D.
degree.
In 2001 he joined the Department for Measurement Technology, Johannes Kepler University, as a
Research Assistant. His current research interests are
tomographical imaging methods, signal acquisition,
and data processing.

Bernhard G. Zagar was born in Klagenfurt, Austria,


in 1957. He received the diploma and Ph.D. degrees
in electrical engineering from the Technical University of Graz, Austria, in 1983 and 1988, respectively,
and the M.S. degree in computer science from University of California, Davis, in 1988.
In 1983, he joined the Department of Electrical
Engineering, Technical University of Graz, where
he was an Associate Professor in electrical measurement and instrumentation until 2001. In July
2001, he became a full Professor for measurement
technology at the Johannes Kepler University, Linz, Austria. From 1986 to
1987, he was a Research Assistant and during 1994 a Research Associate at the
University of California, Davis, in the Department of Electrical Engineering
and Computer Science. His research interests include digital signal and image
processing and applications of lasers for noncontacting measurements.

You might also like