Lens-Free Computational Microscopy for Disease Diagnosis
- Author(s): Zhang, Yibo
- Advisor(s): Ozcan, Aydogan
- et al.
The optical microscope has served as a workhorse for the medical community for hundreds of years and has been one of the gold-standards for the diagnosis of numerous diseases. However, the high cost and bulkiness of lens-based microscopes largely restrict them to advanced imaging labs in centralized clinics; whereas the limited throughput increases the time and labor for the diagnostician to reach a diagnosis, which in turn increases the cost of diagnosis.
In this dissertation, a novel imaging modality, namely lens-free computational microscopy, and the techniques and methods that enable its applications to medical diagnostic tasks will be introduced. With no lenses in the optical setup, the lens-free microscope directly captures defocused holographic patterns of the sample using an optoelectronic image sensor, and computationally recovers the sample’s image with image reconstruction algorithms. This unique design makes the lens-free microscope extremely cost-effective, wide-field, robust and compact. Centered around lens-free microscopy, three application domains are covered in this dissertation: bright-field pathology slide imaging for cancer diagnosis (pathology), polarized microscopy for gout/pseudogout diagnosis (rheumatology), and rapid screening of parasite infection in bodily fluids (microbiology). First, a multi-height-based phase retrieval method was developed for lens-free microscopy, which enabled imaging of pathology slides of tissue sections and cell smears for diagnosis of various diseases such as cancers. In addition, an absorbance spectrum estimation-based colorization method was developed to provide accurate colorization of lens-free microscopy images of stained tissues and cells, and an edge sparsity-based autofocusing algorithm improved the automation and robustness of the image processing pipeline. Furthermore, using chemical tissue clearing, i.e., the simplified CLARITY method, wide-field three-dimensional imaging of thick tissues was demonstrated, which could potentially be used for cell phenotyping and clinical diagnosis. Next, on the basis of some of these techniques, a polarization-sensitive lens-free microscopy platform was created, which could image birefringent crystals in the synovial fluid for high-throughput diagnosis of gout and pseudogout. Lastly, a novel optical technique for sensitive and high-throughput detection of parasitic infections is developed, named lensless time-resolved holographic speckle imaging, which analyses the motility patterns of parasites in the bloodstream or other bodily fluids using deep learning.