Professional Documents
Culture Documents
Biomedical Optics Laboratory, Department of Physics and Astronomy, Clemson University, Clemson, SC 29634
hjiang@clemson.edu
1. R. L. Barbour, H. L. Graber, Y. Wang, J. H. Chang, and R. Aronson, "A perturbation approach for optical
diffusion tomography using continuous-wave and time-resolved data," in Medical Optical Tomography, G.
Miller ed., SPIE Institute for Advanced Optical Technologies (SPIE Optical Engineering Press Vol. IS11,
1993), 87-120.
2. S. R. Arridge and J. C. Hebden, "Optical imaging in medicine: II. Modeling and reconstruction," Phys. Med.
Biol. 42, 841-853 (1997).
3. M. A. O'Leary, D. A. Boas, B. Chance, and A. G. Yodh, "Experimental images of heterogeneous turbid
media by frequency-domain diffusion-photon tomography," Opt. Lett. 20, 426-428 (1995).
4. X. D. Li, T Durduran, A. G. Yodh, B. Chance, and D. N. Pattanayak, "Diffraction Tomography for
biomedical imaging with diffuse-photon density waves," Opt. Lett. 22, 573-575 (1997).
5. C. L. Matson, "A diffraction tomographic model of the forward problem using diffuse photon density
waves," Optics Express 1, 6-12 (1997).
6. S. A. Walker, S. Fantini, and E. Gratton, "Image reconstruction by backprojection from frequency domain
optical measurements in highly scattering media," Appl. Opt. 36, 170-179 (1997).
7. S. B. Colak, D. G. Papaioannou, G. W. 't Hooft, M. B. van der Mark, H. Schomberg, J. C. J. Paasschens, J. B.
M. Melissen, and N. A. A. J. van Asten, "Tomographic image reconstruction from optical projections in light-
diffusing media," Appl. Opt. 36, 180-213 (1997).
8. K. D. Paulsen, H. Jiang, “Spatially-varying optical property reconstruction using a finite element diffusion
equation approximation,” Med. Phys. 22, 691-702 (1995).
9. H. Jiang, K. D. Paulsen and U. L. Osterberg, B. W. Pogue and M. S. Patterson, “Optical image
reconstruction using frequency-domain data: simulations and experiments," J. Opt. Soc. Am. A 13, 253-266
(1996).
10. K. D. Paulsen and H. Jiang, "Enhanced frequency-domain optical image reconstruction in tissues through total
variation minimization," Appl. Opt. 35, 3447-3458 (1996).
11. H. Jiang, K. D. Paulsen, U. L. Osterberg and M. S. Patterson, "Frequency-domain near-infrared photo
diffusion imaging: initial evaluation in multi-target tissue-like phantoms," Med. Phys. 25, 183-193 (1998).
12. D. T. Delpy, M. Cope, “Quantification in tissue near-infrared spectroscopy,” Phil. Trans. R. Soc. Lond. B 352,
649-659 (1997).
13. A. H. Hielscher, R. E. Alcouffe, and R. L. Barbour, "Comparison of finite-difference transport and
diffusion calculations for photon migration in homogeneous and heterogeneous tissues," Phys. Med. Biol. 43,
1285-1302 (1998).
14. M. Firbank, S. Arridge, M. Schweiger and D. Delpy, “An investigation of light transport through scattering
bodies with non- scattering regions,” Phys. Med. and Biol. 41 767-783 (1996).
1. Introduction
Current optical image reconstruction algorithms [1-11] are almost all based on the first-order
diffusion equation, which is only applicable in cases where the scattering is assumed to
dominate the absorption and the optode spacing is much larger than the inverse of the
reduced scattering coefficient. While these conditions are satisfied for almost all tissues in
the near-infrared region, there are concerns in the use of the first-order diffusion equation in
optical image reconstruction. The major concern is encountered in imaging multi-layered
brain tissue where a clear, non-scattering layer of cerebrospinal fluid (CSF) lies between the
inner skull table and the brain surface. The first-order diffusion equation fails to describe
light propagation in this region [12-14], which means that alternative approaches for light
modeling must be used. Another concern involved in optical image reconstruction using the
first-order diffusion equation is highly absorbing regions such as hematoma.
2. Reconstruction algorithm
In this paper we are interested only in optical image reconstruction in the steady-state,
continuous-wave domain. The algorithm in the frequency- or time-domain can be developed
in a similar manner. The third-order diffusion equations derived from the Boltzmann photon
transport equation can be stated as[16-18]:
25 60 60
−∇ ⋅ D(r)∇Φ (1) (r) + ∇⋅ D(r)∇Φ ( 2) (r) − 5µ′t (r)Φ( 2) (r) − ∇⋅ D(r)∇1Φ ( 3) (r) − ∇ ⋅D(r)∇ 2 Φ (4) (r) = 0 (2)
7 7 7
10 90
∇ ⋅ D(r)∇1Φ (1) (r) − ∇ ⋅ D(r)∇ 1Φ (2 ) (r) + ∇ ⋅ D(r)∇Φ ( 3) (r) − 10 µt′ (r)Φ ( 3) (r) = 0 (3)
7 7
1 5 45
∇ ⋅ D(r)∇ 2 Φ (1) (r) − ∇ ⋅ D(r)∇2 Φ ( 2) (r) + ∇ ⋅ D (r)∇ Φ (4) (r) − 5µt′ (r)Φ ( 4) (r) = 0 (4)
2 7 7
We must choose appropriate boundary conditions (BCs) in order to solve Eqs. (1)-
(4). In this study we apply Type III BCs to the first component, Φ(1) , i.e.,
−D ˆn ⋅ ∇Φ(1) = αΦ (1) , where nˆ is the unit normal vector for the boundary surface and α is a
coefficient that is related to the internal reflection at the boundary, and employ Type I BCs
(2) ( 3) ( 4)
to the remaining components, i.e., Φ = Φ = Φ = 0 .
Making use of finite element discretizations, the discretized forms of Eqs. (1)-(4)
can be written as
N P Q P P P
∑ Φ (1)
j − ∑ D p φ p ∇φ j ⋅ ∇φ i − ∑ µ q φ q φ j φi + Φ(2)
j ∑ D pφ p∇φ j ⋅ ∇φ i (3)
− 6Φ j ∑ D pφ p ∇1 φ j ⋅ ∇φi − 6Φ j
(4)
∑ D p φ p ∇ 2φ j ⋅ ∇φ i
j =1 p= 1 q =1 p =1 p =1 p= 1
(1) (2) (3) (4)
= − Sφ i − ∫ Dˆn ⋅ ∇Φ φi ds + ∫ Dˆn ⋅ ∇Φ φ ids − 6∫ Dˆn ⋅ ∇1Φ φ ids − 6∫ Dˆn ⋅ ∇2Φ φ ids (5)
N P
25 (2) P
5 P 60 P
60 (4) P
∑ Φ (1)j ∑ Dp φp ∇φ j ⋅ ∇φ i − Φj ∑ D pφ p∇φ j ⋅ ∇φ i + ∑ D −1
p φ q φ j φi + Φ(j3) ∑ D p φ p ∇1φ j ⋅ ∇φ i − Φj ∑ Dp φ p ∇ 2 φ j ⋅∇φ i
j =1 p =1 7 p= 1 3 p= 1 7 p =1 7 p =1
N P
10 P
90 P
10 P
∑ Φ (1)
j − ∑ D p φ p∇1 φ j ⋅∇φ i + Φj
(2)
∑ D p φ p∇ 1φ j ⋅ ∇φ i −
(3)
Φj ∑ D p φ p ∇φ j ⋅∇φ i + ∑ D −p 1 φ pφ j φ i
j =1
p =1 7 p =1 7 p =1 3 p =1
N 1 P
5 (2) P
45 (4) P
5 P
∑ Φ(1)j − ∑ D pφ p ∇ 2 φ j ⋅ ∇φ i + Φj ∑ D pφ p ∇ 2 φ j ⋅ ∇φ i − Φj ∑ D p φ p∇φ j ⋅ ∇φi + ∑ D −1p φ p φ j φ i
j =1
2 p=1 7 p=1 7 p=1 3 p=1
1 5
D ˆn ⋅ ∇ 2 Φ(1) φ ids + ∫ Dnˆ ⋅ ∇ 2 Φ(2) φ ids −
45
Dnˆ ⋅ ∇Φ (4) φi ds
(8)
2∫ 7 ∫
=−
7
where indicates integration over the problem domain and Φ (1)-(4), D, and µa have been
expanded as the sum of coefficients multiplied by a set of locally spatially-varying
Lagrangian basis functions φ φ p , and φ q . ∫ expresses integration over the boundary
surface. N is the node number of a finite element mesh. The expansions used to represent D
and µa are P and Q terms long where P ≠ Q ≠ N in general; however, in the study reported
here P=Q=N.
where
∂Φ(1)
1 ∂Φ 1(1) ∂Φ1(1) ∂Φ1(1) ∂Φ1(1) ∂Φ1(1)
∂D L L
∂D2 ∂D N ∂µ a,1 ∂µa ,2 ∂µa ,N
1
∂Φ(1)
2 ∂Φ (1)
2 L
∂Φ (1)
2 ∂Φ(1)
2 ∂Φ (1)
2 L
∂Φ(1)
2
ℑ = ∂D1 ∂D2 ∂D N ∂µ a,1 ∂µa ,2 ∂µa ,N (10)
M(1) M O M M M O M
∂Φ M ∂Φ (1)
M ∂Φ (1)
M ∂Φ(1)
M ∂Φ (1)
M
(1)
∂ΦM
L L
∂D1 ∂D2 ∂D N ∂µ a,1 ∂µa ,2 ∂µa ,N
and ∆χ = (∆D1 , ∆D2 ,...∆D N ,∆µ a,1, ∆µa ,2 ,...∆µa, N )T is the update vector for the optical
property profiles. Φ o = (Φ(1),o
1 ,Φ(21),o ,...Φ (1),o
M )
T
and Φ c = (Φ1(1), c , Φ(1),
2
c
,...Φ (1),c T
M ) , where
Φ(1),o
i and Φ (1),c
i are observed and calculated average diffused photon density for i=1, 2, ...
M boundary locations. Note that only the first component or the average diffused photon
density, Φ(1) , is used in Eq. (9) since it is the dominant component [16], and other
components, Φ (2)−( 4) , are set to zeros at the boundary. In Eq. (9), the decomposition of the
ill-conditioned matrix ℑT ℑ is stabilized by a synthesized Marquardt and Tikhonov
regularization scheme [8,9].
C
R1: 21.5 mm
G
R2: 12.5 mm
A E F B
Fig. 1. (a) Geometry of the test case under study; (b) reconstructed D image for the first test case;
(c) reconstructed µ a image for the first test case.
In this section we use numerical examples to test the reconstruction algorithm described in
Section 2. The test geometry, shown in Fig. 1 (a), consists of a circular background region
(radius=21.5 mm) with an embedded circular target (radius=6.25 mm) offsetting 5 mm. The
examples include two test cases with different optical properties assigned in the embedded
target and background regions. For the first case, the optical properties for the target are
µ ′s =2.0 mm-1, µ a=0.012 mm-1; the optical properties for the background are µ ′s =1.0 mm-1,
µ a=0.006 mm-1. For the second case, the optical properties for the target are µ s′ =0.01 mm-1,
µ a=0.005 mm-1; the optical properties for the background are µ s′ =1.0 mm-1, µ a=0.01 mm-1.
The first case is used to just demonstrate the implementation of our third-order
In the examples, the "measured" data were generated using a forward higher-order
diffusion model with the exact D and µ a in place. Fig. 1 (b, c) shows the D and µ a images for
the first case reconstructed under conditions of no noise. As can be seen, the images are
clearly recovered.
(a) (b)
(c) (d)
Fig. 2. (a) Recovered D image for the second case using the third-order codes; (b) recovered µ a
image for the second case using the third-order codes; (c) recovered D image for the second case
using the first-order codes; (b) recovered µ a image for the second case using the firs-order codes;
For the second case, 2% noise has been added to the “measured” data. Fig. 2 (a, b)
presents the successfully reconstructed D and µ a images for the second case. In order to
provide a comparison, D and µ a images reconstructed using our first-order codes are
displayed in Fig. 2 (c, d). A number of observations can be made from Fig. 2. The almost
non-scattering, void-like target can be qualitatively recovered for both D and µ a images using
the third-order codes [Fig. 2 (a, b)], whereas it cannot be correctly recovered using the first-
order codes [Fig. 2 (c, d)]. Interestingly, the target location for both D and µ a images
4. Conclusions
Acknowledgments
This work was supported in part by the National Institutes of Health (R55 CA78334) and the
Greenville Hospital System/Clemson University Biomedical Cooperative.