UCLA

Recent developments of deep learning-enabled image transformation and object detection in microscopic images has revolutionized traditional computational imaging techniques and outperformed many digital image processing algorithms in both speed and quality. This dissertation introduces a set of novel deep learning techniques for cross-modality image super-resolution, virtual histological staining, and early bacterial colony using time-lapsed coherent microscopic images. This dissertation first introduces a deep learning-based method to correct distortions introduced by mobile-phone-based microscopes is introduced, which facilitates the production of high-resolution, denoised and color-corrected images, matching the performance of benchtop microscopes with high-end objective lenses, also extending their limited depth-of-field. Inspired mobile-phone microscope to benchtop microscope image transformation, a deep learning-enabled super-resolution framework across different fluorescence microscopy modalities is also demonstrated. Using this framework, the resolution of wide-field images acquired with low-numerical-aperture (NA) objectives were improved to match the resolution that is acquired using high-NA objectives. The framework was further applied to cross-modality super-resolution transformation of confocal microscopy images to match the resolution acquired with a stimulated emission depletion (STED) microscope, and transformation of total internal reflection fluorescence (TIRF) microscopy images of subcellular structures within cells and tissues to match the results obtained with a TIRF-based structured illumination microscope. The similar cross-modality image transformation framework can also transform autofluorescence images of unlabeled tissue sections into the equivalence of the bright-field images captured with histologically stained versions of the same samples. A blind comparison, by board-certified pathologists, of this virtual staining method and standard histological staining using microscopic images of human tissue sections of the salivary gland, thyroid, kidney, liver, and lung, and involving different types of stain, showed no major discordances. Other than image transformation, a deep learning-based live bacteria detection system was also developed which periodically captures coherent microscopy images of bacterial growth inside a 60-mm-diameter agar plate and analyses these time-lapsed holograms for the rapid detection of bacterial growth and the classification of the corresponding species. This system shortens the detection time of Escherichia coli and total coliform bacteria in water samples by >12 h compared to the Environmental Protection Agency (EPA)-approved methods, achieved a limit of detection (LOD) of ~1 colony forming unit (CFU)/L in ≤9 h of total test time. This platform is highly suitable for integration with the existing methods currently used for bacteria detection on agar plates.