This research optimized parameters for breast cancer imaging with diffuse optical tomography using the NIRFAST software. The forward mesh parameters of node size, depth, and width were optimized to 0.75mm, 60mm, and 60mm, respectively, for a source-detector separation of 15mm. A regularization parameter of λ=1 provided the best reconstruction quality. Having more source-detector pairs improved absorption coefficient reconstruction accuracy. Future work will address frequency-domain measurements and crosstalk between absorption and scattering coefficients.

1. This research was funded by a generous gift from the Xerox Corporation.
Optimization of NIRFAST Parameters for Breast Cancer Imaging with Diffuse Optical Tomography
Abstract Optimization Process
Results
Conclusion & Future Research
Acknowledgements
Sophie (Xueqi) Zhang, Songfeng Han, Jingxuan Ren, Gabriel Ramirez, Ashley Proctor, Regine Choe – Department of Biomedical Engineering, University of Rochester, Rochester, NY
References
Introduction & Motivation
Forward Mesh Parameter Optimization
(a) (b)
Reconstruction Regularization Parameter (λ)
(a) 8 sources and 8 detectors with 15mm as the shortest source-detector separation;
(b) Moving probe 1 by Δx=Δy=2.5mm to get more coverage: obtained 52 sources and 39
detectors in the end.
λ""value" Max"
mua/"
mm,1"
Relative"
%"error"
Max"
mus’/"
mm,1"
Relative"
%"error"
Mu"
effective"
Relative"
%"error"
Expected" 0.04$ /$ 1.0$ /$ 0.35$ /$
0.01" 0.020$ 50%$ 2.2$ 115%$ 0.36$ 4%$
1" 0.014$ 65%$ 1.4$ 40%$ 0.25$ 27%$
5" 0.014$ 65%$ 1.4$ 40%$ 0.23$ 34%$
50" 0.012$ 71%$ 1.2$ 20%$ 0.21$ 41%$
$
!!!!!!value! Max!
mua/!
mm+1!
Relative!
%!error!
Max!
mus’/!
mm+1!
Relative!
%!error!
Mu!
effective!
Relative!
%!error!
Expected! 0.04$ /$ 1.0$ /$ 0.35$ /$
0.01! 0.031$ 23%$ 1$ 0$ 0.30$ 12%$
1! 0.043$ 7.5%$ 1$ 0$ 0.36$ 2.9%$
5! 0.044$ 10%$ 1$ 0$ 0.37$ 5.7%$
50! 0.045$ 13%$ 1$ 0$ 0.37$ 5.7%$
$
Example of reconstructed µa Images (CW: Continuous Wave; 0MHz)
Expected
λ=0.01, 1 ,5 and 50 were tested.
(a) (b)
[2] Durduran, T., R. Choe, W. B. Baker, and A. G. Yodh. "Diffuse Optics for Tissue Monitoring and Tomography." Rep. Prog. Phys. Reports on Progress
in Physics 73.7 (2010): 076701. Published 2 June 2010.
[1] Dehghani, Hamid, Matthew E. Eames, Phaneendra K. Yalavarthy, Scott C. Davis, Subhadra Srinivasan, Colin M. Carpenter, Brian W. Pogue, and
Keith D. Paulsen. "Near Infrared Optical Tomography Using NIRFAST: Algorithm for Numerical Model and Image Reconstruction." Communications in
Numerical Methods in Engineering. 25(6), 711-732 (2009)
The author gratefully thanks and acknowledges all the helpful discussions support and interactions with Songfeng Han, Jingxuan Ren, Gabriel
Ramirez, Ashley Proctor, Nathaniel Barber and Dr. Choe in Choe Lab at University of Rochester.
§ Optimized forward mesh parameters for source detector separation of 15mm: node size = 0.75mm,
depth= 60mm, width= 60mm.
§ Regularization parameter λ affects reconstruction image quality.
§ Having more source and detector pairs improves µa reconstruction.
§ Future research: frequency-domain measurements for simultaneous reconstructions of µa and µs
’.
However, the crosstalk issue between of µa and µs
’ needs to be solved first.
Near Infrared Fluorescence and Spectral Tomography (NIRFAST) is a
software package that models photon propagation in turbid medium and
reconstructs 3D images of optical coefficients based on data[1]. The ultimate
goal of this research is to design an imaging probe for breast cancer. Before
designing the probe, we first optimized various parameters in NIRFAST, such
as the node size, depth and width of finite element mesh in forward model,
regularization parameter λ in inverse model and source-detector pattern to
obtain adequate reconstructed images within minimum computational time.
This work demonstrates the method of comparing different probe geometries
using NIRFAST.
Diffuse Correlation Spectroscopy (DCS) and
Diffuse Optical Spectroscopy (DOS) are diffuse
optical instruments that implement diffusion
model of light transport, which separates
absorption (µa) and scattering coefficients (µs
’)
of tissues, thus providing information on blood
flow, oxygenated and deoxygenated
hemoglobin concentrations. The µa value of
tumor is higher than that of normal tissues due
to angiogenesis[2]. Under a clinical setting, an
optical probe containing source and detector
fibers that are connected to DCS/DOS is
placed on patients’ breasts to measure tissue
hemodynamics non-invasively. The optimal
source-detector pattern can be designed to
cover a range of the area of interest and tumor
depth. Thus, an adequate probe design is
essential for clinical use of diffuse optical
instruments.
Figure 1.(a) DCS is being used to
take a measurement on a patient. (b)
An optical probe is placed on a
breast.
Node size
Forward mesh depth (z-axis)
Forward mesh width (x-axis)
Forward & Inverse Problem
Solving an Inverse Problem
Input Output
Light intensity &
phase at each
detector from
diffuse optical
instrument or
simulation.
Inverse
Problem
Solver
A 3-D
distribution of
reconstructed
optical
coefficients.
Generate Simulated Data with NIRFAST
Input
Forward
Model
Output
Intensity Map Phase Map
Intensity Phase
Extract at detector
Probe Example
Expected 3D µa Distribution
• µa
tumor=0.04mm-1
• µa
breast=0.01mm-1
• Radius of the tumor=5mm
• Depth of the tumor=7.5mm
(a) Coarse mesh (b) Fine mesh.
(a) Depth=20mm (b) Depth=100mm
(a) Width=120mm (b) Width=40mm
Probe Configuration
(a) (b)
Node size
Relative Intensity Error Relative Phase Angle Error
Adequate node size = 0.75mm for SD=15mm
Depth
Relative Intensity Error Relative Phase Angle Error
Adequate depth = 60mm for SD=15mm
Width
Relative Phase Angle ErrorRelative Intensity Error
Adequate width = 60mm for SD=15mm
Probe Comparison (CW)
Expected
Probe 1 Probe 2
Parameter Crosstalk Issue
µa
µs
'
Probe 1 Probe 2
Depth=0mm Depth=3.5mm Depth=5.5mm Depth=7.5mm Depth=10.5mm Depth=12.5mm
Reconstructed
(λ=1,CW)
Variation of Regularization Parameter λ (CW)
λ=1λ=0.01 λ=5
Expected
λ=1 at depth = 7.5mm
λ=1 at depth
= 7.5mm
Expected
Reconstructing µa Only
(CW: 0MHz)
Reconstructing µa & µs
’
(140MHz)
CW Results
Probe 1
Probe 1
Goal: To find adequate parameter values with
minimum computational time.
NIRFAST utilizes finite element
method (FEM) to simulate light
propagat ion in biological
tissues[1], In this case, a breast
with a tumor is simulated. Here
we used forward model inside
NIRFAST to generate data as
input for inverse problem solver.
λ=50
140MHz Results
Compared with the solution to photon diffusion equation for a homogeneous semi-infinite medium.
(µa=0.01mm-1, µs
’=1.0mm-1)