Shunan Wu
BME PhD Defense Presentation

Date: 2025-03-24
Time: 03/24 8:00 PM-9:00 PM (EST, UTC-4) / 03/25 8:00 AM-9:00 AM (Beijing Time, UTC+8)
Location / Meeting Link: https://gatech.zoom.us/j/3058457734?omn=91442671485

Committee Members:
Xunbin Wei, PhD (Advisor); Shu Jia, PhD (Co-Advisor); Peng Xi, PhD; Changhui Li, PhD; Jiajia Luo, PhD


Title: Large-Field High-Temporal-Resolution Microscopic Imaging: Deconvolution and Dynamic Range Improvement

Abstract:
The discovery of cells has revolutionized life sciences by revealing the fundamental building blocks of organisms and uncovering the intricate mechanisms that govern biological processes. Optical microscopy, particularly fluorescence microscopy, has become essential for visualizing live-cell dynamics with high spatial and temporal resolution. For high spatial resolution or 3D fluorescence microscopy, enormous data volumes and intensive computations make reconstruction several orders of magnitude slower than acquisition, which is unacceptable for real-time monitoring and high-throughput processing. Besides, high-temporal-resolution fluorescence microscopy faces significant challenges: high-speed acquisition necessitates short exposure times and reduced bit depths, which lead to photon starvation, motion blur, and a compressed dynamic range. These limitations can obscure critical details and compromise quantitative analyses, posing a formidable barrier to capturing rapid, volumetric biological events with the desired fidelity. In this study, we present two complementary computational methodologies designed to overcome these obstacles in high-temporal-resolution microscopy. The first approach, the PEARL deconvolution method, introduces a transformative acceleration framework for three-dimensional image reconstruction. By employing third-order gradient extrapolation for local gradient approximations, PEARL dramatically reduces the number of iterations required for deconvolution. Our evaluations on both simulated and experimental datasets—spanning subcellular imaging of organelles to larger-scale imaging of organoids—demonstrate that PEARL can significantly cut reconstruction processing times. This substantial acceleration enables high-throughput, long-term, and fast volumetric monitoring, thereby facilitating the study of dynamic biological phenomena without compromising spatial resolution. Complementing this, we have developed the Self-FM-HDR reconstruction method to address the limited dynamic range inherent in high-temporal-resolution imaging systems. High-speed cameras, such as sCMOS and SPAD detectors, often operate at reduced bit depths to achieve kilohertz frame rates, resulting in images that suffer from under-saturation and increased noise. Self-FM-HDR overcomes these drawbacks by fusing multiple low-bit-depth exposures through advanced noise suppression and artifact mitigation strategies, thereby reconstructing high dynamic range images. This method effectively preserves fine structural details and restores an intensity range comparable to traditional 16-bit imaging, ensuring that both dim and bright regions are accurately captured even under photon-limited conditions. Together, these methodologies address the dual imperatives of speed and image quality in high-temporal-resolution fluorescence microscopy. PEARL deconvolution accelerates volumetric reconstruction, making fast, high-resolution imaging feasible, while Self-FM-HDR ensures that the full richness of the sample’s fluorescence is retained. The integration of these techniques lays a robust foundation for next-generation high-temporal-resolution imaging systems, capable of capturing live biological processes with unprecedented clarity and temporal precision.