Abstract
Fluorescence microscopy reveals dynamic biological processes but remains limited by photobleaching and phototoxicity. Illumination generates contrast yet simultaneously damages fluorophores and cells, restricting imaging duration and interpretability. Lower-ing excitation intensity reduces harm but yields sparse, noisy photon data that challeng-es high-fidelity reconstruction. This dissertation develops an ultralow-light imaging framework that enables non-invasive, long-term fluorescence microscopy. It introduces near-zero-photon bioimaging, which redesigns the pipeline from photon detection to computational reconstruction. By combining photon-efficient acquisition and AI models optimized for sparse statistics, the system achieves irradiance over 10,000× lower than conventional microscopy while sus-taining kilohertz frame rates. These advances eliminate photobleaching and open imag-ing regimes inaccessible under standard illumination.
Operating in this regime also enables new resolution capabilities, termed photon super-localization, a method that uses the multipixel signatures of single-photon events to as-sign them with subpixel accuracy on a high-density virtual grid. This overcomes the usu-al field-of-view versus resolution trade-off of moderate-NA, long-working-distance objec-tives, enabling high-resolution imaging across large fields without hardware changes or additional photons.
Finally, this dissertation establishes the first biologically grounded metric for phototoxici-ty. An integrated microscopy-transcriptomics assay detects light-induced cellular stress before morphological damage or cell-growth-related effects appear, providing a rigorous framework for evaluating illumination conditions.
Together, these contributions extend the physical limits of fluorescence microscopy, es-tablishing a unified platform for high-resolution, long-duration, non-invasive live-cell im-aging under ultralow-light conditions.