Professional Documents
Culture Documents
2003 Hybrid Monte Carlo-Diffusion Method For Light
2003 Hybrid Monte Carlo-Diffusion Method For Light
The heterogeneity of the tissues in a head, especially the low-scattering cerebrospinal fluid 共CSF兲
layer surrounding the brain has previously been shown to strongly affect light propagation in the
brain. The radiosity-diffusion method, in which the light propagation in the CSF layer is assumed
to obey the radiosity theory, has been employed to predict the light propagation in head models.
Although the CSF layer is assumed to be a nonscattering region in the radiosity-diffusion method, fine
arachnoid trabeculae cause faint scattering in the CSF layer in real heads. A novel approach, the
hybrid Monte Carlo– diffusion method, is proposed to calculate the head models, including the low-
scattering region in which the light propagation does not obey neither the diffusion approximation nor
the radiosity theory. The light propagation in the high-scattering region is calculated by means of
the diffusion approximation solved by the finite-element method and that in the low-scattering region
is predicted by the Monte Carlo method. The intensity and mean time of flight of the detected light
for the head model with a low-scattering CSF layer calculated by the hybrid method agreed well with
those by the Monte Carlo method, whereas the results calculated by means of the diffusion approx-
imation included considerable error caused by the effect of the CSF layer. In the hybrid method, the
time-consuming Monte Carlo calculation is employed only for the thin CSF layer, and hence, the
computation time of the hybrid method is dramatically shorter than that of the Monte Carlo method.
© 2003 Optical Society of America
OCIS codes: 170.0170, 170.3660, 170.3890.
兰
and c is the speed of light in the CSF layer. Sum-
⌫ 2共q兲 ⫽ ⌫ 1共p兲 F共p, q兲dA 1, (4) ming all the contributions from the inner surface of
As-in the superficial layer, As-in, yields the mean time of
具t 13共q兲典 ⫽ 再兺
p
关具t 12共p兲典 ⫹ t pq共q兲兴⌫ pq共q兲 冎 冒兺
p
⌫ pq共q兲 (6) distribution of the incoming flux from the CSF. This
problem was previously investigated by Firbank et
where ⌫pq共q兲 is the flux from p to q. The mean time al.9 Figure 4 shows the angular distribution of the
of flight, 具t14共s兲典, of the flux, ⌫4共s兲, from the brain light exiting from the superficial layer predicted by
surface was obtained from this mean time distribu- the MC method. The refractive-index mismatch on
tion and the mean time of flight, 具tqs共s兲典, of light that the superficial–CSF-layer boundaries was ignored
traveled in the brain layer from q to s calculated with and a hundred million photons were injected to ob-
the FEM: tain the result. The distribution of the surface ra-
再兺 冎 冒兺
diance is proportional to the cosine of the exiting
zenithal angle for the exiting light that is isotropi-
具t 14共s兲典 ⫽ 关具t13共q兲典⫹具tqs共s兲典兴⌫qs共s兲 ⌫ qs共s兲, (7) cally distributed, whereas the distribution of the ra-
q q
diance caused by the photon density inside a
where ⌫qs共s兲 is the flux at s from q. The mean time scattering body can be considered proportional to
of flight of flux, ⌫5共r兲, can be calculated with Eq. 共6兲, cos 共1 ⫹ cos 兾0.71兲. The angular distribution of
and the mean time of flight in the superficial layer the transmitted light shown in Fig. 4共a兲 indicates
was calculated with the FEM. These processes were that we should consider the 共cos 兾0.71兲 term for the
iterated until changes in the mean time of flight of first calculation of the light propagation in the CSF
the detected light become negligible. layer.9 The angular distribution of the light re-
Since the light propagating in the CSF layer is flected from the high-scattering medium is almost
hardly scattered, the angular dependence of the light proportional to cos as shown in Fig. 4共b兲. Conse-
radiated to the CSF layer significantly affects the quently, the term of 共cos 兾0.71兲 for the light source
distribution in the MC method was only considered in sults suggest that the diffusion approximation is not
the first iteration of the hybrid calculation. valid in the low-scattering CSF layer of the head
model and that it is not appropriate to employ the
3. Results and Discussion FEM to calculate the light propagation in the head
model with the CSF.
A. Validity of Diffusion Equation for Low-Scattering
Medium B. Transmittance and Reflectance Ratios
Figures 5 and 6 shows the intensity and mean time of In the hybrid MC– diffusion method, the incoming flux
flight of the detected light for a homogeneous model from the CSF layer such as ⌫2共q兲, ⌫3共r兲 and ⌫5共r兲 can be
calculated by the FEM and MC method as a function obtained from the transmittance and reflectance ra-
of source– detector spacing, respectively. The trans- tios. We do not have to recalculate the light propa-
port scattering coefficient of the homogeneous model gation in the CSF by the MC method in each iterative
was 0.1 and 1.0 mm⫺1, and the absorption coefficient process once the transmittance and reflectance ratios
of both models was 0.002 mm⫺1. Since the diffusion are obtained. The transmittance and reflectance ra-
equation is not valid in a low-scattering medium, the tios as a function of the detected position are shown in
intensity and mean time of flight for the low- Fig. 7. The angular distribution of the point source on
scattering model calculated with the FEM are signif- the surface of the CSF layer is assumed to be propor-
icantly underestimated. The results of the FEM for tional to cos . These ratios can be used to calculate
the model with a scattering coefficient is 1.0 mm⫺1 the incoming flux from the CSF layer except for the
agreed well with that of the MC method. These re- first calculation in the iterative processes. In case of
the nonscattering CSF layer, the transmittance ratio is ing is shown in Fig. 8. The results calculated by the
only shown because no reflection can occur in the non- hybrid MC– diffusion method and FEM are compared
scattering CSF layer. The total intensity of the trans- with those by the MC method to discuss the validity
mittance decreases with the increasing scattering of the methods. Figure 8共a兲 shows the detected in-
coefficient of the CSF layer, whereas the total intensity tensity for the model with the nonscattering CSF
of the reflectance increases with the increasing scat- layer. The result of the hybrid radiosity-diffusion
tering coefficient. The distribution of both the trans- method is also shown in Fig. 8共a兲 since that method
mittance and the reflectance of light propagated in the can be applied only to the model with the nonscatter-
CSF layer caused by a point source on the boundary ing CSF. The intensity calculated by the hybrid ra-
converges with the increasing scattering coefficient of diosity and hybrid MC methods and FEM agree with
the CSF layer. Since the geometry of the CSF layer in that by the MC method up to a source– detector spac-
the head model was a simple slab, the transmittance
ing of 15 mm. Once the source– detector spacing
and reflectance ratios at any position were the same.
exceeds 15 mm, the result of the FEM does not match
The MC method calculation for 50 million incident
photons to obtain the transmittance and reflectance the MC method results, whereas the results by the
ratios for each model took approximately 10 min in hybrid methods are in good agreement with that by
computation time on a dual 450 MHz Pentium III PC. the MC method. Since the diffusion equation is not
valid in the nonscattering region, the influence of the
C. Validity of Hybrid Monte Carlo–Diffusion Method CSF layer should be considered for calculation of the
The intensity of detected light on the surface of the propagation of light detected at more than 15 mm
head model as a function of the source– detector spac- from the source. The detected intensity for the
model with the low-scattering CSF layer calculated mean time of flight for the model with the nonscat-
with the FEM is slightly overestimated compared tering CSF layer predicted by the MC method and
with the MC method results beyond the source– the hybrid methods increases slowly for source–
detector spacing of 20 mm, whereas the results from detector spacing greater than 30 mm and levels off,
the hybrid MC method agree well with the MC whereas that by the FEM remains approximately
method results, as shown in Figs. 8共b兲 and 8共c兲. The constant for the source– detector spacing greater
error in the FEM results for the head model decreases than 20 mm as shown in Fig. 9共a兲. The FEM un-
with the increasing scattering coefficient of the CSF derestimates the mean time of flight for the light
layer. The diffusion equation holds in the scattering detected at more than 20 mm from the source be-
medium with a transport scattering coefficient of 1.0 cause of the collapse of the diffusion approximation
mm⫺1 as shown in Fig. 5共b兲. The detected intensity in the CSF layer. The slope of the mean time of
calculated with the FEM agrees with the MC method flight for the head models with the low-scattering
results as well as the hybrid MC method in the case CSF layer with 0.1 and 0.2 mm⫺1 transport scat-
in which the transport scattering coefficient of the tering coefficients predicted by the MC method
CSF layer is 1.0 mm⫺1. slightly decreases beyond the source– detector spac-
The mean time of flight for the head models as a ing of 30 mm. This tendency can be observed in
function of the source– detector spacing calculated the experimentally measured mean time of flight
by the MC method, the FEM, and the hybrid MC for adult foreheads.11 The mean time of flight cal-
method is shown in Fig. 9. The result of the hybrid culated by the hybrid MC method agrees well with
radiosity method is also shown in Fig. 9共a兲. The the MC method results, whereas that by the FEM