Computational microscopy methods to image at high resolution over large areas

Abstract

Modern optical imaging faces a fundamental trade-off between achieving high resolution and maintaining a large field of view—a limitation that spans various modalities, from fluorescence and X-ray imaging to brightfield microscopy. However, the emergence of powerful computational resources such as GPUs, alongside new classes of image sensors like SPAD arrays and multi-camera systems, has catalyzed rapid advancement in computational imaging. This paradigm integrates algorithmic reconstruction and optical design, enabling breakthroughs beyond traditional resolution limits. In this thesis, we aim to address the high-resolution, large-field-of-view imaging challenge by combining innovations in all three core elements of the optical system—illumination, optics, and detection—with modern computational methods.We develop and apply a range of techniques rooted in phase retrieval, Fourier optics, first Born approximation, optical aberration theory, and geometric optics to enhance both the hardware and software aspects of microscopy systems. Our work begins with improving Fourier ptychography using SPAD arrays for faster acquisition and extends to vectorial phase retrieval for polarization-sensitive imaging of anisotropic samples. We also introduce system-level innovations by designing new lenses and incorporating fiber bundle arrays to overcome vignetting effects in wide-area imaging. Collectively, this thesis demonstrates a versatile and scalable framework for high-throughput, high-resolution optical imaging, offering practical solutions to long-standing trade-offs in microscopy. Our contributions expand the capabilities of computational optics and set a foundation for future applications in digital pathology, neuroscience, and live-organism imaging.