Distortion Correction and Denoising of Light Sheet Fluorescence Images

Area of Science:

• Neuroscience
• Microscopy
• Image Processing

Background:

• Light Sheet Fluorescence Microscopy (LSFM) is crucial for volumetric brain imaging in neurobiology.
• LSFM images often contain artifacts and distortions that hinder accurate 3D reconstruction.
• Image enhancement is essential for optimizing LSFM data quality.

Purpose of the Study:

• To develop and validate a comprehensive image processing pipeline for LSFM data.
• To correct slice-by-slice artifacts and distortions before 3D volume reconstruction.
• To improve the quality and interpretability of LSFM images for neurobiological research.

Main Methods:

• A three-step enhancement process: deblurring, automatic contrast enhancement, and convolutional denoising auto-encoder.
• A novel auto-encoder approach for correcting axial distortion using bead calibration images.
• Implementation of skip connections in the auto-encoder for efficient noise reduction.

Main Results:

• The proposed pipeline effectively reduces noise, particularly mixed Poisson-Gaussian noise.
• Axial distortion correction was achieved using a dedicated auto-encoder.
• The method demonstrated superior denoising performance compared to existing techniques.

Conclusions:

• The developed pipeline offers a complete solution for LSFM image enhancement.
• Results show significant improvements in image quality and denoising capabilities.
• This advancement has the potential to enhance the interpretation of complex biological data from LSFM.