DOI: 10.1002/lpor.71702 ISSN: 1863-8880

Learning to Self‐Interfere: Zero‐Shot Deconvolution Enables Replica‐Free, Low‐Coherence Common‐Path Quantitative Phase Microscopy

Zhuoshi Li, Zezhong Zhang, Yao Fan, Runnan Zhang, Yefeng Shu, Jiasong Sun, Lei Zhang, Ran Ye, Piotr Zdańkowski, Maciej Trusiak, Malgorzata Kujawinska, Qian Chen, Chao Zuo

ABSTRACT

Digital holographic microscopy (DHM) enables full‐field, label‐free, and high‐precision quantitative phase imaging (QPI), yet its performance is often limited by speckle noise, parasitic artifacts, and vibration sensitivity resulting from the highly coherent illumination and separate reference beam path. While reducing illumination coherence facilitates suppressing speckle and parasitic fringes, it typically requires a common‐path self‐interference configuration, leading to replica images that hinder quantitative analysis of dense specimens. Here, we introduce replica‐free total‐shear three‐beam common‐path digital holographic microscopy (RF‐CPDHM), which resolves this long‐standing dilemma by coupling physical modeling with a deconvolution‐prior zero‐shot learning strategy. By formulating replica removal as an ill‐posed inverse problem and solving it in an unsupervised manner, RF‐CPDHM achieves high‐resolution, high‐SNR, and full‐field QPI without any ground truth or training data. The polarization‐grating‐based three‐beam interference configuration provides intrinsic achromaticity, enabling low‐coherence illumination to suppress speckle and parasitic fringes while maintaining high phase sensitivity. Experiments on a USAF resolution target demonstrate a half‐pitch resolution of 308 with a , 0.8 NA objective lens. Moreover, RF‐CPDHM enables quantitative visualization of organelle dynamics, including lipid droplets and vesicles, in PLC cells under oleic acid treatment, demonstrating its broad potential for label‐free quantitative studies in cell biology, pathology, and pharmacology.

More from our Archive