Image Reconstruction in Light-Sheet Microscopy: Spatially Varying Deconvolution and Mixed Noise

Area of Science:

• Microscopy
• Image Processing
• Computational Science

Background:

• Light-sheet microscopy data is often degraded by spatially varying blur.
• Microscopy images are affected by a combination of Poisson and Gaussian noise.
• The point spread function (PSF) in light-sheet microscopy varies due to excitation sheet and detection objective interactions.

Purpose of the Study:

• To develop an advanced deconvolution technique for light-sheet microscopy.
• To model the image formation process considering the interaction between excitation and detection.
• To create a robust variational model for handling mixed noise types.

Main Methods:

• A new image formation model is proposed, incorporating the excitation-detection interaction.
• A variational model is formulated using an infimal convolution of noise fidelities to manage Poisson and Gaussian noise.
• The primal-dual hybrid gradient (PDHG) algorithm is adapted for efficient inverse problem solving.

Main Results:

• Convergence rates and a discrepancy principle are established for the infimal convolution fidelity.
• The proposed method demonstrates superior image reconstruction compared to existing techniques.
• Successful validation on both simulated and real light-sheet microscopy data.

Conclusions:

• The developed deconvolution method effectively handles spatially varying blur and mixed noise in light-sheet microscopy.
• The novel variational approach and PDHG algorithm application offer significant improvements in image quality.
• This work advances the capabilities of quantitative imaging in light-sheet microscopy.