Accelerated Richardson-Lucy deconvolution by unmatched projection pairs for 3D fluorescence microscopy

Area of Science:

• Microscopy and Imaging Science
• Computational Imaging
• Biophysics

Background:

• The Richardson-Lucy deconvolution (RLD) algorithm enhances fluorescence microscopy image sharpness.
• High computational cost of RLD limits its use with large 3D datasets and real-time visualization.

Purpose of the Study:

• To introduce an accelerated RLD (WB-ARL) method for efficient 3D fluorescence microscopy.
• To improve scalability and reduce computational demands for high-throughput imaging.

Main Methods:

• Developed WB-ARL using a Wiener-Butterworth unmatched backprojector.
• Flattened spectral product and suppressed high-frequency noise.
• Implemented CUDA acceleration for further speed enhancement.

Main Results:

• WB-ARL required over 10-fold fewer iterations than conventional RLD with high-fidelity reconstruction.
• CUDA acceleration increased speed 40-fold, with WB-ARL showing a 400x iterative advantage.
• Demonstrated robustness to noise and optical aberrations.

Conclusions:

• WB-ARL enables high-resolution, high-fidelity 3D imaging with reduced computational cost.
• Offers a scalable solution for high-throughput fluorescence microscopy.
• Facilitates real-time volumetric visualization in microscopy.