Biomedical Engineering Reference
In-Depth Information
where A, the scattering prefactor, depends on the number and size of the scatterers
and b, the scattering exponent, depends on the size of the scatterers. Incorporating
these models in the multispectral reconstructions allows to directly map functional
and structural parameters simultaneously. The method has been demonstrated to
improve robustness and accuracy in quantitative functional imaging of the breast
[ 21 , 67 ].
Similarly, the use of anatomical priors from auxiliary clinical imaging modality
can dramatically increase the spatial and quantitative accuracy of optical imaging.
The incorporation of structural priors in optical imaging can be traced back to
the work of Chang et al. [ 68 ]. Based on simulations, they demonstrated that by
using an anatomically accurate optical (AAO) model derived from a segmented
MRI data, optical reconstructions were able to retrieve the pathology accurately.
Since, the introduction of spatially localized anatomical areas in the optical image
formation flow has become a significant area of research for the optical tomography
community. The many different methods devised to include anatomical priors can
be clustered in two main categories, hard and soft priors.
Hard priors represent the most straightforward method in implementing the
anatomical priors. In these approaches, the volume to be imaged is segmented in
different structures based on the anatomical modality data. Hard prior methods
enforce the anatomical boundaries onto the optical imaging reconstructions. All
voxels belonging to the same spatial domain are then correlated in the optical
reconstructions, reducing dramatically the number of unknowns to estimate. The
implementation of hard priors varies from assuming that all segmented tissues have
homogenous optical properties [ 69 ], that some segmented tissues are homogenous,
and that compromised tissues are heterogeneous [ 70 ] or utilize homogenous
tissue-dependent optical properties obtained in a first step as initial guesses for
higher-resolution optical reconstructions in a second step [ 71 ]. All these methods
have been proven to be robust even in the case of noisy optical data [ 57 ].
However, hard priors imply that the anatomical structure of the probed tissue is
known and that there is a spatial correlation between the anatomical and optical
edges. In realistic scenarios, structural or physiological correlation between the
true optical solution and a prior image from a different medical imaging modality
cannot be guaranteed. This can be due to the different contrast mechanisms revealed
by each modality and/or geometrical deformation occurring during the different
examination protocols. Then, it is expected that enforcing incorrect priors can
lead to nontrustworthy optical reconstructions. To alleviate this issue, methods
incorporating soft priors have been proposed.
Conversely to hard priors, reconstruction techniques based on soft priors do not
force the anatomical boundaries to coincide with the optical domains. Thus, changes
of the optical parameters across boundaries are allowed, reducing the impact of
imperfect/incomplete anatomical information on the optical reconstructions. Guven
et al. [ 72 ] proposed a Bayesian formulation minimizing the incorrect bias from the
prior anatomical data but preserving the prior information leading toward the true
solution. Intes et al. [ 67 ] extended this method to functional imaging of the breast
Search WWH ::




Custom Search