You are here

ECE PhD Defense: Vivian Pera

13
Nov

206 Lake Hall

November 13, 2015 9:00 am to 11:00 am
November 13, 2015 9:00 am to 11:00 am
Title: "Designing Novel Biophotonic Imaging Concepts: Signal Processing Algorithms and Hardware Implementation"

 

Abstract: 

In this work, we consider the design of novel biophotonic imaging concepts from both a signal processing and hardware perspective. With respect to the former, we employ classic and recent signal processing approaches to (1) explore the optimization of system design for a novel time-domain hyperspectral tomographic imager; and (2) develop novel, intrinsically-regularized algorithms for the processing of fluorescence molecular tomography (FMT) data from two instrument prototypes. With respect to the latter, we consider the construction of a custom microscope to be used for label-free enumeration and characterization of circulating cells in vivo.

FMT is an optical technique that uses near-infrared light to perform quantitative, three-dimensional imaging of fluorophores in whole animals noninvasively. It is a variant of diffuse optical tomography (DOT) and is becoming an important tool in preclinical imaging of small animals. In the first project, we evaluated the potential of the Cramér-Rao lower bound (CRLB) to serve as a design metric for diffuse optical imaging systems. The CRLB defines the best theoretical precision of any estimator for a given data model; it is often used in the statistical signal processing community for feasibility studies and system design. Computing the CRLB requires inverting the Fisher information matrix (FIM). However, this matrix is usually ill-conditioned (and often underdetermined) in the case of ill-posed inverse problems like FMT or DOT. We regularized the FIM by assuming that the object to be imaged was a point target and assessed the ability of point-target CRLBs to predict system performance in silico. We found that these bounds are not good predictors of imaging performance across different system configurations, even in a relative sense, and that agreement between the trends predicted by the CRLBs and imaging performance obtained with reconstruction algorithms that rely on a different regularization approach cannot be assumed a priori.

In the second and third projects, we developed intrinsically-regularized data processing algorithms for two small-animal FMT instruments under construction in our lab.  By "intrinsically-regularized," we mean that regularization is accomplished only by physically relevant assumptions (e.g., optical intensity is nonnegative) without relying on user-specified, often empirically determined, parameters. The first instrument we call a "diffuse fluorescence flow cytometer"; it can detect and localize very rare fluorescently-labeled circulating cells in the limb of a mouse. We developed a maximum-likelihood algorithm for tomographic image reconstruction that significantly out-performed a standard image reconstruction method, particularly for deep-seated targets, and achieved close to 150 µm accuracy in a 3 mm diameter cross-sectional area with only 12 measurements. The algorithm is also suitable for real-time implementation.

For the second instrument, a time-domain hyperspectral tomographic fluorescence imager, we proposed an intrinsically-regularized demixing algorithm which incorporates ideas from sparse subspace clustering and compressed sensing. It uses a suitable "library" of fluorophore signatures to compute a nonnegative least-squares estimate of each fluorophore signal in the sample. The algorithm does not require a regularization parameter, even when the library is rank-deficient. In simulations, we simultaneously demixed four fluorophores with closely overlapping spectral and temporal profiles in a 25 mm diameter cross-sectional area with a root-mean-squared error of less than 3% per fluorophore.

Finally, in the fourth project we constructed an infinity-corrected brightfield microscope to be used, along with spontaneous Raman and absorption spectroscopy measurements, for label-free enumeration and characterization of circulating cells in mice in vivo. We showed that the resolution of our microscope was better than ~2 µm, and that we were able to image flowing microspheres in a microfluidic chip at linear speeds of up to 3.1 mm/s, which is consistent with the upper limit of speeds expected in blood vessels in a mouse ear.

 

Advisors: Professor Mark Niedre & Professor Dana Brooks

Committee Members:
Professor Mark Niedre
Professor Dana Brooks
Professor Charles DiMarzio