Physics-guided self-supervised reconstruction for structured illumination microscopy

2026-05-07

PRE

Junli Wu

Qiurong Yan *

Junyuan Yin

Siying Huang

Haoran Zhang

Zhiqiang Wen

Xiaolong Luo

DOI：10.1016/j.optlaseng.2026.109850

Original

Original request for help

Save

Abstract

En 中文

• PG-SIM embeds a differentiable SIM forward model into a self-supervised reconstruction framework for robust structured-illumination microscopy. • A dual-domain objective combines Poisson-Gaussian measurement-domain data fidelity with frequency-domain consistency to suppress aliasing leakage near the OTF cut-off. • Dedicated PG-SIM-3/6/9 models are trained for protocol-specific 3-, 6-, and 9-frame acquisitions rather than subsampling from a common 9-frame sequence. • On BioSR simulations, PG-SIM-6 improves the FRC-derived resolution from 199.8 nm for wide-field imaging to 128.6 nm. • Under matched 3 × 3 synthetic inputs, PG-SIM-9 outperforms APC-SIM, SparseDeconv-SIM, and PRS-SIM, while hardware experiments show favorable FWHM/PVR behavior on fixed-cell and semiconductor-phantom data.

Keywords:

PG-SIM

structured illumination microscopy

self-supervised reconstruction

frequency-domain consistency

aliasing suppression

Physics-guided self-supervised reconstruction for structured illumination microscopy

Abstract

Journal

Organization

Cited Papers

Citing Papers