Zero-shot learning for denoising and super-resolution in multifocal structured illumination microscopy

Compared to conventional widefield-illumination microscopy, multifocal structured illumination microscopy (MSIM) achieves a twofold enhancement in spatial resolution while exhibiting greater imaging penetration depth and superior optical sectioning capabilities. These technical advantages position MSIM as a prominent solution for three-dimensional (3D) super-resolution (SR) imaging of thick biological specimens. However, the intrinsically low signal-to-noise ratio (SNR) of raw MSIM data introduces reconstruction artifacts during SR processing. Such artifacts not only degrade image resolution and fidelity but also constrain imaging speed, thereby limiting the applications of MSIM in live-cell and tissue SR imaging. To address these challenges, this study proposes a novel approach termed ZSDN-MSIM: Zero-shot Deconvolution Networks (ZSDN) are implemented to perform denoising and deconvolution preprocessing on raw MSIM data, coupled with pixel reassignment post-processing for MSIM SR reconstruction. This methodology significantly enhances MSIM imaging quality under low-illumination conditions, demonstrating substantial value for advancing low-photodamage in vivo imaging applications.