A convex 3D deconvolution algorithm for low photon count fluorescence imaging.

Area of Science:

• Microscopy
• Image Processing
• Biophysics

Background:

• Deconvolution enhances 3D fluorescence microscopy images but can create artifacts under low light conditions.
• Existing methods often fail when noise models are violated, particularly with low photon counts.
• Biological specimens often exhibit complex structures that challenge standard deconvolution techniques.

Purpose of the Study:

• To develop a novel deconvolution method tailored for 3D fluorescence imaging in the low-light regime.
• To address reconstruction artifacts caused by noise models being violated in low photon count scenarios.
• To provide a robust deconvolution tool for biological research involving low-light imaging.

Main Methods:

• Utilized a mixed Poisson-Gaussian noise model accounting for photon shot and camera read noise.
• Formulated a convex loss function solved via the alternating direction method of multipliers (ADMM).
• Employed a Hessian-based regularizer optimized for biological specimen features and incorporated on-the-fly noise parameter estimation.

Main Results:

• The proposed method significantly reduces artifacts in simulated and experimental low-light 3D fluorescence images.
• Demonstrated effective deconvolution for images with peak intensities as low as tens of photoelectrons per voxel.
• Validated performance in live cell imaging, showcasing practical applicability.

Conclusions:

• The developed deconvolution algorithm offers superior performance for low-light 3D fluorescence microscopy of biological samples.
• On-the-fly noise parameter estimation simplifies the workflow by removing the need for manual calibration.
• This method serves as a valuable tool for advancing biological research requiring high-fidelity imaging under low photon conditions.