Beruflich Dokumente
Kultur Dokumente
Huabei Jiang
We present a finite-element-based algorithm for reconstruction of fluorescence lifetime and yield in turbid
media, using frequency-domain data. The algorithm is based on a set of coupled diffusion equations that
describe the propagation of both excitation and fluorescent emission light in multiply scattering media.
Centered on Newton’s iterative method, we implemented our algorithm by using a synthesized scheme
of Marquardt and Tikhonov regularizations. A low-pass spatial filter is also incorporated into the
algorithm for enhancing image reconstruction. Simulation studies using both noise-free and noisy data
have been performed with the nonzero photon density boundary conditions. Our results suggest that
quantitative images can be produced in terms of fluorescent lifetime and yield values and location, size,
and shape of heterogeneities within a circular background region. © 1998 Optical Society of America
OCIS codes: 170.0170, 170.3010, 170.6960, 170.3650, 170.3660, 170.3880.
F G
However, the correlation between the diseased tis-
sues and measured optical signals has been weak- iv
¹ z @Dx~r!¹Fx~r, v!# 2 max~r! 2 Fx~r, v!
ened by background light absorption and scattering, c
which are nonspecific to disease. Hence the speci-
ficity and the sensitivity for detection might be im- 5 2S~r, v!, (1)
F G
proved if these background signal contributions could
iv
be eliminated from the fluorescent, endoscopic mea- ¹ z @Dm~r!¹Fm~r, v!# 2 mam~r! 2 Fm~r, v!
surement. Toward this end, fluorescence lifetime c
and yield measurements in tissues demonstrate flu- 1 1 ivt~r!
orescence lifetime imagesyspectra that are indepen- 5 2h~r!max3mFx~r, v! , (2)
dent of tissue absorption and scattering in vivo. 1 1 v2t~r!2
Although a number of groups have begun to explore where Fx,m is the photon density for excitation ~sub-
the methods needed to extract fluorophore lifetimes script x! or fluorescent light ~subscript m!, Dx,m is the
from deep within tissue,23–29 the development of such diffusion coefficient, max,m is the absorption coefficient
techniques has been hampered since absorption and due to contributions from both nonfluorescing chro-
scattering distort both the intensity and the lifetime mophores and fluorescent dye, max3m is the absorption
kinetics of reemitted fluorescent light. This distor- coefficient for the excitation light due to the contri-
tion or convolution is complex in nature because the bution from fluorescent dye, v is the modulation fre-
optical property distribution can be quite nonuniform quency, c is the velocity of light in the medium, and h
and in general is not known a priori. This complex- and t are the fluorescent quantum yield and lifetime.
ity necessitates the use of detailed analysis of the S~r, v! is the excitation source term in Eq. ~1!, which
multiply scattered excitation and fluorescent light to for a point source can be written as S 5 S0d~r 2 r0!,
reconstruct the lifetime images in tissues. where S0 is the source strength and d~r 2 r0! is the
Studies in fluorescent reconstructions in multiply Dirac-delta function for a source at r0. Note that a
scattering media or tissues have just begun. Several single-exponential fluorescence decay has been as-
groups have recently reported their simulated results sumed in the source term for fluorescent light @right-
in this regard, using a simple integral method,30 hand side of Eq. ~2!#; multiexponential time decay can
perturbation methods,31,32 and a finite difference al- be handled by a simple extension. The diffusion co-
gorithm.33 In this paper we report a finite-element- efficient can be written as
based reconstruction algorithm for imaging both
1
fluorescence lifetime and yield in thick tissues, us- Dx,m~r! 5 , (3)
ing frequency-domain data. The algorithm is con- 3@max,m~r! 1 msx,m9~r!#
structed under the same framework of our algorithms
for optical imaging in which excitation data are where msx,m9~r! is the reduced scattering coefficient.
For known optical properties and fluorescent life-
used.34,35 The finite-element method has proved to
time and yield, Eqs. ~1! and ~2! become standard
be powerful in studies of both forward and inverse
boundary value problems for the spatially varying
photon migration problems by our own research and photon densities of excitation and emission light sub-
others.10,34 – 42 Based on a set of coupled diffusion ject to appropriate boundary conditions ~BC’s!. In
equations that describe the propagation of both exci- this paper we use the nonzero photon density BC’s
tation and fluorescent emission light in multiply scat- or type III BC’s, which have proved to be most accu-
tering media, the algorithm is centered on Newton’s rate:35,43 2Dx,m¹Fx,m z n̂ 5 aFx,m, where n̂ is the unit
iterative method where a synthesized scheme of Mar- normal vector for the boundary surface and a is a
quardt and Tikhonov regularizations is used. A low- coefficient that is related to the internal reflection at
pass spatial filter is also incorporated into the the boundary.
algorithm to enhance image reconstruction. Simu- In general, the purpose of fluorescent diffusion to-
lation studies in which both noise-free and noisy ~to mography is to recover all distributions including
as much as 5% noise! data are used have been per- Dx,m, max,m, t, and h. We focus on the reconstruction
formed with the nonzero photon density boundary of t and h distributions in this study. Reconstruc-
conditions. Images reconstructed are quantitative tions of other parameters can be easily included in
in terms of fluorescent lifetime and yield values and the algorithm. Thus, making use of finite-element
location, size, and shape of heterogeneities within a discretizations of Eqs. ~1! and ~2!, we can obtain two
circular background region. matrix equations in terms of a discrete set of spatially
distributed fluorescent lifetime and yield and photon
density values:
2. Reconstruction Algorithm @Ax#$Fx% 5 $bx%, (4)
In the frequency domain it is known that propagation
of both excitation and fluorescent emission light in @Am#$Fm% 5 $bm%. (5)
K S D L
here; i.e., we Taylor expand both the real and the
iv imaginary parts of Fm about an assumed ~t, h! dis-
~ax!ij 5 2Dx¹cj z ¹ci 2 max 2 cj ci , (6)
c tribution that is a perturbation away from some other
distribution ~t̃, h̃! so that a discrete set of photon
K
~am!ij 5 2Dm¹cj z ¹ci 2 mam 2 S iv
c
D L
cj ci , (7)
density values can be expressed as
]Fm~R! ]Fm~R!
and the entries in column vectors $bx,m% and $Fx,m% are Fm~R!~t̃, h̃! 5 Fm~R!~t, h! 1 Dt 1 Dh 1 · · ·,
]t ]h
M (11)
~bx!i 5 2^Sci & 1 a ( ~F ! r c c ds,
x j j i (8)
]Fm~I! ]Fm~I!
Fm~I!~t̃, h̃! 5 Fm~I!~t, h! 1
j51
Dt 1 Dh 1 · · ·,
]t ]h
7 D8
L
K N 1 2 iv ( tc l l (12)
( (
S(
~bm!i 5 2 hkckmax3m ~Fx!j cj ci l51
L 2 where Dt 5 t̃ 2 t and Dh 5 h̃ 2 h. Fm~R! and Fm~I!
k51 j51
11v 2
tlcl are the real and the imaginary parts of Fm. If the
l51
assumed lifetime and yield distributions are close to
M
the true profiles, the left-hand side of Eqs. ~11! and
1a ( ~F
j51
! r cj cids,
m j (9)
~12! can be considered as true data ~either imposed or
observed! and the relationship truncated to produce
Fx,m 5 $~Fx,m!1, ~Fx,m!2, · · · ,~Fx,m!N%T, (10)
)Dx 5 Fmo 2 Fmc, (13)
where ^ & indicates integration over the problem do-
main and Fx,m, t, and h have been expanded as the where ) is the Jacobian matrix consisting of the de-
sum of coefficients multiplied by a set of locally spa- rivatives of Fm with respect to t or h at each boundary
tially varying Lagrangian basis functions cj , cl, and observation node34,35; Dx is the vector that expresses
ck. r expresses integration over the boundary sur- perturbations of t and h; and Fmo and Fmc are the
face where type III BC’s have been applied. ~Fx,m!i observed and the computed fluorescent photon den-
is the photon density at node i, N is the node number sity at the boundary. All these matrix and vectors
of a finite-element mesh, and M is the boundary node can be written as
3 4
]Fm,1~R! ]Fm,1~R! ]Fm,1~R! ]Fm,1~R! ]Fm,1~R! ]Fm,1~R!
··· ···
]t1 ]t2 ]tL ]h1 ]h2 ]hK
]Fm,1~I! ]Fm,1~I! ]Fm,1~I! ]Fm,1~I! ]Fm,1~I! ]Fm,1~I!
··· ···
]t1 ]t2 ]tL ]h1 ]h2 ]hK
· · ·· · · · ·· ·
)5 ·
·
·
· · ·
·
·
·
·
· · ·
· , (14)
]Fm,M~R! ]Fm,M~R! ]Fm,M~R! ]Fm,M~R! ]Fm,M~R! ]Fm,M~R!
··· ···
]t1 ]t2 ]tL ]h1 ]h2 ]hK
]Fm,M~I! ]Fm,M~I! ]Fm,M~I! ]Fm,M~I! ]Fm,M~I! ]Fm,M~I!
··· ···
]t1 ]t2 ]tL ]h1 ]h2 ]hK
Dx 5 ~Dt1, Dt2, · · · DtL, Dh1, Dh2, · · · DhK!T, (15)
~R! ~I! ~R! ~I! ~R! ~I!
Fm 5 $@Fm #1 , @Fm #1 , @Fm #2 , @Fm #2 · · · @Fm #M , @Fm #M % ,
o o o o o o o T
(16)
Fmc 5 $@Fm~R!#1c, @Fm~I!#1c, @Fm~R!#2c, @Fm~I!#2c · · · @Fm~R!#Mc, @Fm~I!#Mc%T, (17)
number. Note that the expansions used to repre- and @Fm~R,I!#i0 and @Fm~R,I!#ic are observed and cal-
sent the lifetime and yield profiles in Eq. ~9! are K and culated real ~or imaginary! parts of the fluorescent
L terms long where K Þ L Þ N in general; however, photon density @based on the estimated ~t, h! distri-
in the research reported here K 5 L 5 N. bution# for i 5 1, 2, . . . , M boundary locations. tl for
To form images from presumably uniform initial l 5 1, 2, . . . , L and hk for k 5 1, 2, . . . , K are the
estimates of the fluorescent lifetime and yield distri- reconstruction parameters for the lifetime and the
butions, we need a way of updating t and h from their yield profiles.
starting values. There are typically two methods Left multiplying Eq. ~13! by the transpose of ) and
Lifetime Yield
F (S DG
N 2 1y2
1 xexact 2 xreconstructed
,
N i51 xexact
Table 1. Geometric Information Derived from the Reconstructed Images in Different Noise Conditionsa
0% Noise 4.5 0.2 4.6 0.3 12.8 12.7 12.9 13.1 1.01 0.98
5% Noise 4.3 0.4 4.2 0.8 13.0 13.2 12.2 11.5 0.98 1.06
a
X and Y refer to the x and y coordinates ~in millimeters! of the target center, respectively. EF and GH are the transect length ~i.e.,
the recovered target diameter in millimeters! of the target region along the x and the y directions, respectively ~see Fig. 1!.