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

Area of Science:

• Biomedical Imaging
• Microscopy Technology
• Optical Physics

Background:

• Multifocal structured illumination microscopy (MSIM) provides enhanced resolution and depth for 3D super-resolution imaging of biological specimens.
• Low signal-to-noise ratio (SNR) in raw MSIM data leads to reconstruction artifacts, limiting imaging speed and fidelity.
• Current limitations hinder MSIM's application in live-cell and tissue super-resolution imaging.

Purpose of the Study:

• To develop a novel preprocessing method for MSIM data to overcome SNR limitations.
• To enhance the quality and speed of MSIM super-resolution reconstruction.
• To enable low-photodamage in vivo imaging applications using MSIM.

Main Methods:

• Implementation of Zero-shot Deconvolution Networks (ZSDN) for denoising and deconvolution of raw MSIM data.
• Integration of ZSDN preprocessing with pixel reassignment for MSIM super-resolution reconstruction (ZSDN-MSIM).
• Evaluation of the proposed method under low-illumination conditions.

Main Results:

• The ZSDN-MSIM approach significantly improves the imaging quality of MSIM data.
• Artifacts introduced by low SNR are substantially reduced.
• Enhanced image fidelity and resolution are achieved, particularly under low-illumination settings.

Conclusions:

• ZSDN-MSIM effectively addresses the SNR challenges in MSIM imaging.
• The methodology facilitates high-quality, low-photodamage super-resolution imaging.
• This advancement holds significant potential for in vivo live-cell and tissue imaging applications.