Estimating Signal Detectability in a Model Diffuse Optical
Imaging System
Stefano Young, Matthew A. Kupinski, Abhinav K. Jha
University of Arizona College of Optical Sciences, 1630 E University Blvd Tucson AZ 85721
Author e-mail address: syoung@optics.arizona.edu
Abstract: Diffuse optical imaging (DOI) researchers need metrics for quantifying signal
detectability to assess different hardware configurations. Using Monte Carlo and statistical model
observers, we estimated DOI signal detectability to compare source, signal, and detector
parameters.
© 2010 Optical Society of America
OCIS codes: (110.3000) Image quality assessment; (170.3660) Light propagation in tissues
1. Introduction
Diffuse optical imaging (DOI) remains limited in its clinical scope because the spatial resolution is low and the
imaging problem is ill-posed. Researchers have suggested that it may not be necessary to solve the ill-posed and
underdetermined imaging problem if one can address a specific clinical task [1]. By optimizing the DOI system
parameters for a specific clinical task, the acquired image data carries the maximum possible task-specific
information. Researchers need methods for quantifying the performance of DOI systems for a particular clinical
task and optimizing the system hardware accordingly. Our group developed a model framework that employs
Monte Carlo and statistical model observers to compute task-based figures of merit for DOI systems. This summary
demonstrates our task-based approach and shows quantitative comparisons for different source, signal, and detection
parameters. For a SKE/BKE signal detection task [2], we quantified our model system’s performance and its
dependence on source diameter (for a uniform beam), number of sources (for individual acquisitions), number of
input source angles, signal location within the tissue, detector size, and detector resolution. The trends for all system
configurations matched with our physical intuitions. These findings suggest that we can quantify signal detectability
in a model DOI system using a task-based approach with mathematical observers. To extend our model approach
for use in optimizing clinical DOI systems, our approach should incorporate faster and more accurate models for the
forward problem and noise sources. To generate more realistic image data in less time, we are developing a non-
uniform RTE solver which may be substituted for Monte Carlo in our model framework.
2. Methods: Forward problem and figure of merit
An imaging system can be described by the following general equation:
(1)
where is the data, is a nonlinear operator, is the object, and is the system noise (not necessarily additive)
[1]. We modeled a 3D slab phantom ( ) with centered and non-centered embedded signals at different depths (Fig.
1) and generated image data ( ) using the open-source tMCimg monte carlo modeling software ( ) [3]. The
source(s) occupied the -plane at the top surface of the slab phantom. We computed figures of merit from the raw
projection data; we did not perform image reconstruction because reconstruction adds unwanted artifacts when our
main purpose is system optimization.
OSA / BIOMED/DH 2010
BSuD26.pdf