The fiber-optic imaging system enables imaging deeply into hollow tissue tracts or organs of biological objects in a minimally invasive way, which are inaccessible to conventional microscopy. It is the key technology to visualize biological objects in biomedical research and clinical applications. The fiber-optic imaging system should be able to deliver a high-quality image to resolve the details of cell morphology in vivo and in real time with a miniaturized imaging unit. It also has to be insensitive to environmental perturbations, such as mechanical bending or temperature variations. Besides, both coherent and incoherent light sources should be compatible with the imaging system. It is extremely challenging for current technologies to address all these issues simultaneously. The limitation mainly lies in the deficient stability and imaging capability of fiber-optic devices and the limited image reconstruction capability of algorithms. To address these limitations, we first develop the randomly disordered glass-air optical fiber featuring a high air-filling fraction (~28.5 %) and low loss (~1 dB per meter) at visible wavelengths. Due to the transverse Anderson localization effect, the randomly disordered structure can support thousands of modes, most of which demonstrate single-mode properties. By making use of these modes, the randomly disordered optical fiber provides a robust and low-loss imaging system which can transport images with higher quality than the best commercially available imaging fiber. We further demonstrate that deep-learning algorithm can be applied to the randomly disordered optical fiber to overcome the physical limitation of the fiber itself. At the initial stage, a laser-illuminated system is built by integrating a deep convolutional neural network with the randomly disordered optical fiber. Binary sparse objects, such as handwritten numbers and English letters, are collected, transported and reconstructed using this system. It is proved that this first deep-learning-based fiber imaging system can perform artifact-free, lensless and bending-independent imaging at variable working distances. In real-world applications, the gray-scale biological subjects have much more complicated features. To image biological tissues, we re-design the architecture of the deep convolutional neural network and apply it to a newly designed system using incoherent illumination. The improved fiber imaging system has much higher resolution and faster reconstruction speed. We show that this new system can perform video-rate, artifact-free, lensless cell imaging. The cell imaging process is also remarkably robust with regard to mechanical bending and temperature variations. In addition, this system demonstrates stronger transfer-learning capability than existed deep-learning-based fiber imaging system.
| Identifer | oai:union.ndltd.org:ucf.edu/oai:stars.library.ucf.edu:etd-7598 |
| Date | 01 January 2019 |
| Creators | Zhao, Jian |
| Publisher | STARS |
| Source Sets | University of Central Florida |
| Language | English |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Electronic Theses and Dissertations |
Page generated in 0.4425 seconds