UCLA

The microscopy imaging technique has been employed as the gold-standard method for diagnosing numerous diseases for hundreds of years. However, the dependence on high-end optical components of traditional optical microscopes may often limit their usage in many applications. Recent developments in deep learning-enabled computational imaging techniques have revolutionized the field achieving both faster speed and higher image quality while maintaining the simplicity of the optical system.In the first part of this dissertation, a set of novel deep learning-enabled microscopy imaging techniques is introduced to perform super-resolution, color holography, and quantitative polarization imaging, which aims for improving the performance of the existing optical system. Firstly, deep learning was adopted to enhance the resolution of both pixel size-limited and diffraction-limited coherent imaging systems, providing a rapid, non-iterative method to improve the space-bandwidth product of coherent imaging systems. Then, a high-fidelity color image reconstruction method using a single hologram is presented, where deep learning was used to simultaneously eliminate the missing-phase-related artifacts and correct the color distortion. In addition, another deep learning-enabled holographic polarization microscope is demonstrated, which can obtain quantitative birefringence retardance and orientation information of the specimen from a phase-recovered hologram from one polarizer/analyzer pair. In the second part of this dissertation, deep learning is further applied to various biological imaging or sensing applications enabling these systems to perform virtual histology staining, cell classification, and pathogen detection. A digital staining technique is first demonstrated to transform the quantitative phase images (QPI) of label-free tissue sections into images equivalent to the brightfield microscopy images of the same tissue sections that are histologically stained. Next, using time-lapse lensless speckle imaging and a deep learning classifier, a computational cytometer is shown to rapidly detect magnetic bead-conjugated rare cells of interest in three dimensions (3D). Lastly, two deep learning-based pathogen detection frameworks are presented. A bacterial colony-forming-unit (CFU) detection system exploiting a thin-film-transistor (TFT)-based image sensor array is firstly shown which can save ~12 hours compared to the Environmental Protection Agency (EPA)-approved methods. Then, a stain-free quantitative viral plaque assay framework is presented which could automatically detect the first cell lysing events due to the viral replication as early as 5 hours after the incubation and achieved a >90% detection rate for the plaque-forming units (PFUs) with 100% specificity in <20 hours, providing major time savings compared to the traditional plaque assays that take ≥48 hours.