Doctoral Research at USC

Molecular imaging is a promising field that allows noninvasive in vivo visualization of specific molecular targets and pathways in living organisms. In recent years its sphere of influence has extended beyond early diagnosis of diseases as it has emerged as a valuable tool for drug discovery and development. Fluorescence molecular tomography (FMT) is a molecular imaging modality that exploits the specificity of fluorescent biomarkers, which absorb and emit light with a downward shift in the frequency. It generates volumetric functional images using near‐infrared light, which is safe and non‐ionizing, at a substantially lower cost than competing modalities like positron emission tomography (PET) and Single photon emission computed tomography (SPECT). While these very attractive features make this functional imaging modality well suited for preclinical research, the FMT problem is confounded by the high degrees of absorption and scattering of photons propagating through tissue, which make the inverse problem ill‐posed. The focus of my PhD research under the mentorship of Dr. Richard Leahy at the University of Southern California was to develop computational methods that enable us to reconstruct high resolution images of molecular targets in vivo in small animals and to develop a novel FMT system for small animal imaging.

---

Fast Forward Models

To accurately and efficiently predict the photon fluence on the animal surface, we have developed a fast perturbative approach based on the Born approximation to compute the system forward model with reasonable accuracy. Our method works for complex animal geometries and takes into consideration the heterogeneity of tissue optical properties. This work was accepted for oral presentation at the SPIE Medical Imaging Conference at San Diego in February, 2008.

Links: SPIE 2008 Paper

---

Optimal Illumination Patterns

We have also developed a systematic approach for designing efficient illumination patterns for FMT. Typically, multiple sets of data are acquired by illuminating the animal with different spatial patterns of near‐infrared light. However the choice of these patterns in most experimental setups is ad hoc and suboptimal. The key idea of our approach is to parameterize the illumination patterns and to formulate a constrained optimization problem that seeks to improve the conditioning of the Fisher information matrix for the FMT forward model. This work was selected for oral presentation at the IEEE Symposium on Biomedical Imaging at Boston in July, 2009, and was published in the journal Physics in Medicine and Biology.

Links: ISBI 2009 Paper, PMB 2010 Paper

Imaging System Development

In collaboration with Prof. Simon Cherry's lab at UC Davis, we have built a prototype imaging system that uses a CCD camera for collecting multispectral fluorescence data and a compact conical mirror to provide a view of the entire surface of a mouse sitting on a transparent stage. We have demonstrated that the conical mirror geometry offers higher light collection efficiency and have developed a calibrated mapping model to extract surface fluorescence data from the mirror image. We have successfully used our system to conduct in vivo experiments. We published these results in the journal Optics Express in April, 2009. Our paper was selected to appear in the May issue of the Virtual Journal of Biomedical Optics.

Link: Optics Express Paper

We have developed a second FMT imaging system in a three-way collaboration with Prof. Cherry's Lab and Cambridge Research and Instrumentation (CRi) (now a subsidiary of PerkinElmer). This system uses a pyramidal mirror setup for multi-view data acquisition and generates computer-controlled spatial patterns of illumination using a Texas Instruments DLP® projector. A set of preliminary phantom studies performed using this was presented at the OSA BIOMED Meeting at Miami in April, 2010.

Link: BIOMED 2010 Paper

---

DigiWarp

We developed a three-step procedure named DigiWarp to warp the Digimouse atlas (a realistic mouse atlas based on CT and cryosection data of a male nude mouse) to fit the surface profile data. This method preserves organ information in the mouse tessellation thus allowing us to account for the optical inhomogeneity of animal tissue in our forward model. This work was also selected for oral presentation at the IEEE Symposium on Biomedical Imaging at Boston in July, 2009, and subsequently published in Physics in Medicine and Biology.

Links: ISBI 2009 Paper, PMB 2010 Paper

Sparsifying and Smoothing Priors

Owing to excessive absorption and scattering of light through tissue, fluorescence tomographic image reconstruction is inherently ill-conditioned. A suitable regularizing penalty function that restricts the search space to a specific solution set is essential to facilitate reconstruction. Typically, fluorescent probes are locally concentrated within specific areas of interest (e.g., inside tumors). The commonly used L2 norm penalty generates the minimum energy solution, which tends to be spread out in space. Instead, we have developed an approach involving a combination of the L1 and total variation (TV) norm penalties, the former to suppress spurious background signals and enforce sparsity and the latter to preserve local smoothness and piecewise constancy in the reconstructed images. Our article describing this work was published as a Featured Article in Physics in Medicine and Biology.

Link: PMB 2012 Paper

In Vivo Experiments

We performed several preliminary in vivo experiments using the conical mirror setup and several of the computational tools described above. Shown below is the reconstruction result from an FMT study performed on a nude mouse with a xenograft melanoma (DX-6) dorsally located on the right flank. The reconstructed image shows fluorescence from the tumor bulge, some fluorescence from dye accumulating in the bladder, and some skin autofluorescence.