See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/338329918

Validation and Comparison of Monte Carlo and Finite Element Method in

Forward Modeling for Near Infrared Optical Tomography

Chapter in Advances in Experimental Medicine and Biology · January 2020

DOI: 10.1007/978-3-030-34461-0_39

CITATIONS READS

0 371

8 authors, including:

Jingjing Jiang Wuwei Ren

ETH Zurich ShanghaiTech University
43 PUBLICATIONS 147 CITATIONS 44 PUBLICATIONS 295 CITATIONS

SEE PROFILE SEE PROFILE

Alexander Kalyanov Scott Lindner

University of Zurich École Polytechnique Fédérale de Lausanne
70 PUBLICATIONS 203 CITATIONS 49 PUBLICATIONS 611 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Wuwei Ren on 09 May 2020.

The user has requested enhancement of the downloaded file.

Validation and Comparison of Monte Carlo and
Finite Element Method in Forward Modeling
for Near Infrared Optical Tomography

Jingjing Jianga,*, Wuwei Rena,b,* , Helene Islera, Alexander Kalyanova, Scott

Lindnera, Di Costanzo Mata Aldoa, Markus Rudinb, Martin Wolfa

a Biomedical Optics Research Laboratory (BORL), Department of Neonatology,

University Hospital Zurich (USZ), 8091 Zurich, Switzerland
b Institute of Biomedical Engineering, ETH Zurich, 8093 Zurich, Switzerland
*Shared first authorship

Abstract Near infrared optical tomography (NIROT) is a non-invasive imaging

technique to provide physiological information e.g. the oxygenation of tissue. For
image reconstruction in clinical and preclinical scenarios, models to accurately
describe light propagation are needed. This work aims to assess the accuracy and
efficiency of different models, which paves the way for an optimal design of model-
based image reconstruction algorithms in NIROT for realistic tissue geometries and
heterogeneities. Two popular simulators were evaluated: the Monte Carlo (MC)
method based MCX and the finite element method (FEM) based Toast++. We
compared simulated results with experimental data measured on a homogeneous
silicone phantom with well-calibrated parameters. The laser light was focused on
the center of the phantom surface and images were captured by a CCD camera in
both reflection and transmission modes. For transmittance measurements, the two
models showed good agreement. Both achieve a cosine similarity of ~99%. In
contrast, for reflectance measurements, FEM results deviated more from the
measured values than MC, yielding similarity values of 86% and 94%, respectively.
This study recommends the use of MC for NIROT in reflection mode and both MC
and FEM yield excellent results for transmission mode.

1. Introduction

To image e.g. the oxygenation of tissue by high resolution near infrared optical
tomography (NIROT), an accurate model for describing light propagation in tissue
is critical irrespective of the actual algorithm used for the image reconstruction. The
gold standard model for light propagation in tissue is the radiative transfer equation
(RTE) [1]. However, the RTE is impractical because it cannot simply be solved
analytically. The Monte Carlo (MC) method provides an accurate solution to the
RTE by simulating the path of photons through the tissue. But MC is
2

computationally expensive and hence time consuming. By using parallel computing

techniques, the state-of-the-art MC-based toolkit, MCX, has improved this situation
considerably [2]. Therefore, image reconstruction based on MC has become
practical, in particular, when combined with high speed methods such as scaling
methods, where the MC simulation is based on known reduced scattering
coefficients (𝜇𝑠′ ), and the simulation results are scaled by absorption coefficients
(𝜇𝑎 ) [3]. However, when the 𝜇𝑠′ values are also unknowns, the current hardware
limits the use of MC particularly for a large number (e.g. >1000) of source-detector
pairs and for time-domain data. An alternative method is to approximate the RTE
by the diffusion theory. Simple analytical solutions exist for homogeneous infinite
and semi-infinite geometries. For more complex geometries the diffusion equation
can be solved numerically e.g. by finite element methods (FEM) [4]. The objective
of this study was to model light propagation comparing the two currently popular
simulators, a MC based model MCX [2] and a FEM based model Toast++ [4]. For
model validation experiments have been carried out using silicone phantoms: the
results of forward simulations by the two simulators were compared with the
measured intensity distribution for both reflection and transmission mode.

2. Methods

2.1 Silicone Phantom experiments

For the experiment, we fabricated a homogeneous slab silicone phantom with a

dimension of 60 × 30 × 15 mm3. The 𝜇𝑎 of the phantom was 0.007 𝑚𝑚−1 and 𝜇𝑠′
was 0.87 𝑚𝑚−1 according to published procedures [5]. The schematic of the
experimental setup is illustrated in Fig. 1. Light was emitted from a solid-state laser
(B&W Tek, Newark, USA) operated at 670 nm and collimated before entering an
imaging chamber. The beam was then directed to the sample surface by a scan head
(ScanLab, Puchheim, Germany), which contained two galvanometric driven
mirrors. By deflecting the beam via mirrors, reflection and transmission mode
operation can be implemented by illuminating the top or bottom surface,
respectively. After interaction with tissue, backscattered light was captured with a
16-bit CCD camera (ANDOR Corporation, Northern Ireland) equipped with a
macro lens (Nikkor, Tokyo, Japan).

2.2 Simulation with MCXLAB

We modeled the above measurement with MCXLAB, a MATLAB version of MCX

 3

with all core codes in C and CUDA C. We employed a discretized volume of

15 × 30 × 60 voxels with voxel size of 1 𝑚𝑚 3 for simulating the silicone phantom.
The same optical properties were assigned to each voxel. Detectors of 2x2 mm2 size
covered the whole area of one side of the volume. A light source with Gaussian
profile was placed on the center of the opposite side in transmission mode and on
the same plane in reflection mode. To account for the continuous wave nature of
the measurement, we utilized 5 × 107 photons to achieve smooth results. Traveling
through the volume, the photons experienced multiple scattering and absorption
events. The computation of their propagation was parallelized on a Geforce Titan
xp GPU (Nvidia, USA) yielding an acceleration by a factor >1000 as compared to
CPU based computation [2].
Once a photon escaped the boundary defined by detectors, the track was saved
and tagged with the corresponding detector ID. The path length of the track s was
used for calculating survival probability P of the photon given by 𝑃 = 𝑒 −𝜇𝑎 𝑠 . After
obtaining P for each detected photon, the energy for each detector can be calculated
by accumulating P for photons detected by the same detector. This requires a loop
over the list of photon tracks, which was facilitated by a MEX compilation of C
code.

2.3 Simulation with Toast++

Tissue is highly scattering at the wavelength utilized in NIROT and, therefore, light
propagation can be approximated as a diffusion process. FEM numerically solves
diffusion equations for complex boundaries. We implemented FEM with Toast++
in MATLAB on a computer with an Intel Core i7@ 2.60 GHz and 8.00 GB RAM.
First, we generated tetrahedral meshes for the slab phantom with an open-source
meshing tool, iso2mesh [6]. To study the influence of the meshing configuration in
FEM, we created a coarse mesh containing 7793 nodes and a refined mesh
consisting of 107293 nodes. The illumination source was set to the center of the top
and the bottom of the mesh for reflection and transmission mode, respectively.
30 × 15 virtual detector pixels were set to cover the top surface of the object, given
that each pixel has a dimension of 2 × 2 mm2, and 𝜇𝑠′ and 𝜇𝑎 were assigned to the
mesh. With all the setup information and presumed optical properties, a system
matrix was generated. System matrix describes the linear relationship between the
illumination source and light intensity within the simulated object [4].

3. Results

We generated transmittance and reflectance data with MCXLAB and

FEM/TOAST++. It took 3.27 s on average to generate the forward results with
MCXLAB. With Toast++, the procedure of calculating the system matrix took
4

1.07 s for the coarse mesh and 17.69 s for the refined mesh. The detected images
were obtained by dividing the illumination profile with the system matrix. For the
coarse and refined meshes, this last step took less than 0.5 s. FEM based Toast++
has the advantage of avoiding the time-consuming step of recalculating the system
matrix for different source patterns, whereas MC has to be run for each new source
pattern. MCXLAB was able to compete with Toast++ with respect to
computational speed. The reason is the acceleration of the MC by the advanced
GPUs. This result is true when combining a small number of sources with a large
number of detectors, which is the case for many NIROT instruments based on
continuous wave CCD cameras or time-resolved SPAD cameras.
To validate the models, we compared simulation and experimental data for both
2D images (Fig. 2) and two directional 1D plots (Fig. 3). The overall distributions
appear similar in transmission mode whereas, in reflection mode, the values from
FEM deviated more from the measured data, especially in the central area.
We calculated the cosine similarity between simulated 𝐼𝑠 and measured results
𝐼m ,

𝐼 ∙𝐼
cosine similarity = ∥𝐼 m∥ ∥𝐼s (1)
m s∥

to quantify the accuracy of the simulation results (Table 1).

Table 1. Cosine similarity between simulated (MC, FEM) and experimental forward results

Transmission mode Reflection mode

MC 0.988 0.942
FEM with fine mesh 0.987 0.868
FEM with coarse mesh 0.983 0.865

For transmission mode, the results from MC and FEM agree well with the
experiments. However, in reflection mode, MC outperformed FEM with 7.4%
higher accuracy. In this study, FEM results obtained with the coarse mesh were not
inferior to those obtained with the refined mesh in transmission and reflection mode.

4. Discussion, conclusion and outlook

We compared the MC-based MCX and the FEM based Toast++ for modeling light
propagation in tissue with experimental results for the continuous wave mode.
Based on these results, MCX is recommended for simulating the reflectance
measurement. In most clinical applications, reflectance will be the relevant mode.
FEM is not as accurate because, in typical NIROT setups, sources and detectors are
placed too close together to apply FEM [7]. FEM relies on the diffusion
 5

approximation, which is not correct for short source-detector distances. When

working in transmission mode, e.g. for breast cancer diagnosis, both MCX and
Toast++ will yield appropriate modeling results.
In a next step, we will compare the two simulation approaches for time resolved
measurements and compare the results with data acquired with our newly developed
in-house TR NIROT system.

Acknowledgments This work was supported by the Clinical Research Priority Programs (CRPP)
Tumor Oxygenation and Molecular Imaging Network Zürich (MINZ) of the University of Zurich,
the Swiss Cancer Research grant KFS-3732-08-2015 and the Swiss National Science Foundation
project 159490.

References

1. Welch AJ and Gemert MJC (2010), Optical-Thermal Response of Laser- Irradiated Tissue.
second ed. Springer 2 edition
2. Fang Q, Boas DA (2009) Monte Carlo Simulation of Photon Migration in 3D Turbid Media
Accelerated by Graphics Processing Units. Opt. Express, 17:20178–20190
3. Zhu C. Liu Q (2013) Review of Monte Carlo Modeling of Light Transport in Tissues. J.
Biomed. Opt. 18:050902.
4. Schweiger Mand, Arridge SR (2014) The Toast++ software suite for forward and inverse
modeling in optical tomography, J Biomed Opt. 19(4): 040801
5. Ren W. (2018). STIFT: a modular software platform for simulation, optimization and
reconstruction in fluorescence molecular tomography. (PhD thesis), ETH Zurich
6. Fang Q, Boas DA (2009). Tetrahedral Mesh Generation from Volumetric Binary and Gray-
Scale Images. IEEE International Symposium on Biomedical Imaging: From Nano to Macro,
Vols 1 and 2, 1142-1145.
7. Yoo M, Liu F and Alfano RR (1990). When does the diffusion- approximation fail to describe
photon transport in random-media. Phys. Rev. Lett. 64, 2647–2650.
 6

Fig. 1. Schematic of the experimental setup.

Fig. 2. Forward results of MC simulated (a, b), FEM simulated (c, d) and phantom experiments
(e, f) in transmission mode (a, c, e) and reflection mode (b, d, f)

Fig. 3. Comparison between measured, MC and FEM simulated data along central lines.

View publication stats