Deep Learning-Based Image Restoration and Super-Resolution for Fluorescence Microscopy: Overview and Resources.

Area of Science:

• Biophysics
• Cell Biology
• Computational Imaging

Background:

• Fluorescence microscopy is vital for live-cell dynamics but suffers from signal degradation due to phototoxicity, bleaching, and noise.
• These limitations reduce signal-to-noise ratio and image resolution, hindering detailed cellular and molecular analysis.

Purpose of the Study:

• To review classical and deep learning (DL) methods for improving fluorescence microscopy image quality.
• To provide researchers with accessible tools and resources for applying DL in live-cell imaging.

Main Methods:

• Overview of traditional image processing techniques for denoising and deconvolution.
• Exploration of deep learning (DL) algorithms applied to fluorescence microscopy data.
• Compilation of open-source databases and code repositories for DL in microscopy.

Main Results:

• DL approaches effectively mitigate noise and enhance resolution in fluorescence microscopy.
• Identified and summarized key DL methods for denoising, deconvolution, and super-resolution.
• Provided a practical DL-based image denoising project for easy implementation.

Conclusions:

• Deep learning offers powerful solutions to overcome limitations in fluorescence microscopy.
• The review equips researchers with resources to implement and develop DL applications for advanced live-cell imaging.
• Facilitates broader researcher engagement in DL for biological imaging research.