Deep learning-enhanced fluorescence microscopy via degeneration decoupling.

Area of Science:

• Microscopy
• Image Reconstruction
• Computational Imaging

Background:

• Deep learning enhances fluorescence microscopy but often ignores physical imaging laws.
• End-to-end deep learning methods require challenging, specific training data.
• Existing methods struggle with low photon counts and artifacts.

Purpose of the Study:

• To develop a novel deconvolution algorithm for fluorescence microscopy.
• To improve resolution and generalization by incorporating physical imaging models and deep learning priors.
• To address challenges associated with low photon budgets and image artifacts.

Main Methods:

• Proposed a deconvolution algorithm combining an imaging model, deep learning priors, and the alternating direction method of multipliers.
• Decoupled reconstruction into deblurring and noise/artifact suppression sub-problems.
• Utilized deep learning priors learned from general datasets and a variance stabilizing transform.

Main Results:

• The novel algorithm demonstrated superior performance in resolution enhancement compared to state-of-the-art methods.
• Achieved better generalization across various measurement data types.
• Effectively handled image degeneration caused by low photon budgets.

Conclusions:

• The proposed algorithm offers a robust and effective approach for fluorescence microscopy image reconstruction.
• Incorporating physical models and generalizable deep learning priors enhances fidelity and performance.
• This method overcomes key limitations of prior deep learning reconstruction techniques.