Dumas, John Paul. Compressive endoscopy - a computational imaging approach for fiber-bundle-based imaging systems. Retrieved from https://doi.org/doi:10.7282/t3-ndf9-0215
DescriptionCompressed sensing (CS) is a signal processing technique that provides a theoretical framework for accurately reconstructing discrete signals from fewer samples than traditionally dictated by the Shannon-Nyquist theorem. In the context of imaging, CS enables the recovery of images with more resolved points than pixels in the physical sensor. This capability is appealing for minimally invasive biomedical imaging applications that suffer from poor image quality due to inherent constraints on the size and type of hardware that can be used. The goal of this dissertation is to adapt the CS framework for use in endoscopy platforms, providing a path toward higher resolution minimally invasive imaging.
Endoscopes commonly use coherent fiber optic bundles to facilitate in vivo imaging. The image quality with these fiber-bundle-based endoscopes is limited because of manufacturing challenges that restrict achievable fiber density and spacing. Chapter 1 reviews endoscopy technologies and current fiber-bundle- based imaging techniques. The general field of computational imaging is then discussed, including a specific focus on CS-based and spectral imaging approaches that may overcome limitations in fiber bundle imaging.
Chapter 2 identifies and addresses some practical challenges that are not anticipated by CS theory or simulations. Computational imaging based on the CS framework, or compressive imaging (CI), was evaluated with a test platform that introduced intensity modulation at a conjugate image plane. It was demonstrated that a CS model accounting for system-specific practical limitations, like optical aberration, is an efficient way to implement highly parallel CI.
An imaging architecture with intensity modulation at a conjugate image plane is one approach for CI, but the development of different CS mathematical models has given rise to various different CI architectures. Chapter 3 provides a comparison of different architectures with the application of endoscopy in mind. An experimental comparison of two candidate architectures was performed, and it was determined that an architecture with coded masks at a conjugate image plane is a good option for translation to endoscopy.
Chapters 2 and 3 developed imaging methods for CI in test platforms where image quality was limited by the number of pixels in a low-resolution sensor. Chapter 4 translates these methods for fiber-bundle-based imaging where image quality is limited by the number of fibers in an imaging bundle. The fiber bundle was considered as a low-resolution sensor array where the number of resolved points in an image is limited by the number of fibers in the bundle. CI was evaluated in a fiber-bundle-based imaging platform for compressive endoscopy, which was demonstrated for fluorescence imaging and resolved 17 points for each fiber in the bundle.
While CI resolves multiple pixels within the diameter of each fiber in thebundle, inter-fiber cladding that binds fibers together blocks information that is not recovered by traditional CI methods. Chapter 4 presents a solution to this missing information problem using spectral coding with a distal prism to ensure information from all sample positions is transmitted through the fibers and captured by proximal color camera. Integrating spectral coding into the compressive endoscopy platform allowed for image reconstruction that filled in 80% of the missing information. Additionally, spectral coding implemented as a standalone technology without CI generated images with three times more resolved points than images from traditional fiber-bundle-based imaging methods. This snapshot spectral coding approach improves image quality in fiber-bundle-based imaging without requiring distal electromechanical components. The computational imaging methods developed here serve as an initial step toward analyzing tissue morphology using CS principles and spectral coding for in vivo optical biopsy.