Dana H. Brooks

Data Specific Spatially Varying Regularization for Multimodal Fluorescence Molecular Tomography

Fluorescence molecular tomography (FMT) allows in vivo localization and quantification of fluorescence biodistributions in whole animals. The ill-posed nature of the tomographic reconstruction problem, however, limits the attainable resolution. Improvements in resolution and overall imaging performance can be achieved by forming image priors from geometric information obtained by a secondary anatomical or functional high-resolution imaging modality such as X-ray computed tomography or magnetic resonance imaging. A particular challenge in using image priors is to avoid the use of assumptions that may bias the solution and reduced the accuracy of the inverse problem. This is particularly relevant in FMT inversions where there is not an evident link between secondary geometric information and the underlying fluorescence biodistribution. We present here a new, two step approach to incorporating structural priors into the FMT inverse problem. By using the anatomic information to define a low dimensional inverse problem, we obtain a solution which we then use to determine the parameters defining a spatially varying regularization matrix for the full resolution problem. The regularization term is thus customized for each data set and is guided by the data rather than depending only on user defined a priori assumptions. Results are presented for both simulated and experimental data sets, and show significant improvements in image quality as compared to traditional regularization techniques.

A Statistical Approach to Inverting the Born Ratio

We examine the problem of fluorescence molecular tomography using the normalized Born approximation, termed herein the Born ratio, from a statistical perspective. Experimentally verified noise models for received signals at the excitation and emission wavelengths are combined to generate a stochastic model for the Born ratio. This model is then utilized within a maximum likelihood framework to obtain an inverse solution based on a fixed point iteration. Results are presented for three experimental scenarios: phantom data with a homogeneous background, phantoms implanted within a small animal, and in vivo data using an exogenous probe.

New techniques for data fusion in multimodal FMT-CT imaging

We examine approaches to the incorporation of anatomic structural information into the inverse problem of fluorescence molecular tomography (FMT). Using an appropriate relationship between anatomic and reconstruction image resolution, we build an inverse problem parameterized along the anatomical segmentation. These values serve as the basis for two new regularization techniques. The first regularizes individual voxels in proportion to the importance of the underlying segments in reducing the residual error. The second is based on a well known statistical interpretation of Tikhonov-type regularization in which the statistical prior is defined implicitly as the solution to a PDE whose structure is based on the anatomical segmentation. Results are shown using both techniques for a simulated experiment within the chest cavity of a mouse.

Detection of the dermis/epidermis boundary in reflectance confocal images using multi-scale classifier with adaptive texture features

Reflectance confocal microscopy is an emerging modality for dermatology applications, especially in-situ and bedside detection of skin cancers. Work to date has concentrated on hardware development and validation by clinicians in comparison with standard histological staining. As this technology gains acceptance, the development of automated processing methods becomes more important. We concentrate here on detection of the dominant internal feature of the skin, the epidermis/dermis boundary, a complex corrugated 3-dimensional layer marked by optically subtle changes and features. We adopt a machine learning approach to this segmentation problem, using a hierarchical multi-scale classifier with sophisticated on-line feature selection, to minimize the required expert labeling and maximize the range of potential features in the face of high inter- and intra-subject variability and low optical contrast. Initial results indicate the ability of our approach to recover the complex 3-D boundary surface.

Localizing the dermis/epidermis boundary in reflectance confocal microscopy images with a hybrid classification algorithm

Confocal reflectance microscopy is an emerging modality, for dermatology applications, especially for in-situ and bedside detection of skin cancers. As this technology gains acceptance, automated processing methods become increasingly important to develop. Since the dominant internal feature of the skin is the epidermis/dermis boundary, it has been chosen as the initial target for this development. This boundary is a complex corrugated 3D layer marked by optically subtle changes and features. Indeed, even trained clinicians in an attempt at validation of our early work, were unable to precisely and reliably locate the boundary within optical resolution. Thus here we propose to detect two boundaries, a lower boundary for the epidermis and an upper boundary for the dermis thus trapping the epidermis/dermis boundary. We use a novel combined segmentation/classification approach applied to z-sequences of tiles in the 3D stack. The approach employs a sequential classification on texture features, selected via on-line feature selection, to minimize the labeling required and to cope with high inter- and intra-subject variability and low optical contrast. Initial results indicate the ability of our approach to find these two boundaries successfully for most of the z-sequences from the stack.

Differential equation-driven regularization for joint FMT-CT imaging

A primary motivation for multi-modal imaging is to improve reconstructions for low resolution functional modalities using high resolution structural information. Most such approaches assume that the anatomic and functional images share a common physical structure. For fluorescence molecular tomography (FMT), however, this may be only approximately valid. We thus present and analyze a regularization scheme that allows more flexible use of anatomic images. Using parallels between regularization and statistical modeling, we develop a stochastic PDE that shares information across structural boundaries. Simulations indicate that our approach is capable of obtaining more accurate reconstructions than methods treating each tissue independently.

Modeling DIC Microscope Images of Thick Objects Using a Product-of-Convolutions Approach

A three-dimensional forward model which attempts to capture phase delays has been developed to simulate microscope images from thick heterogeneous transparent objects. Intensity images were calculated successfully predicting the appearance of Difference Interference Contrast images.

A statistical method for inverting the Born ratio

We examine a statistical solution to the problem of fluorescence molecular tomography (FMT) using the Born ratio. Experimentally verified noise models for the fluorescence and excitation signals are combined to generate a stochastic model for the Born ratio. This model is then utilized within a maximum likelihood (ML) framework to obtain a fixed point iteration about a linear least squares problem. Results for experimental data are shown, including phantoms implanted within a small animal, and in-vivo data using an exogenous probe

A Product-of-Convolutions Model for Three-Dimensional Microscopy, Comparison to Born and Rytov Models

Three-dimensional imaging by a microscope is important in the study of three-dimensional structures such as embryos. In this work we present a three-dimensional forward model and a comparison to Born and Rytov models is presented.