Intelligent Beam Optimization for Light-Sheet Fluorescence Microscopy through Deep Learning.

Area of Science:

• Optical microscopy
• Biomedical imaging
• Computational optics

Background:

• Light-sheet fluorescence microscopy (LSFM) offers high-resolution 3D imaging of cleared tissues.
• LSFM image quality depends critically on illumination beam characteristics.
• Optimizing illumination beams (e.g., Gaussian, Bessel, Airy) for LSFM is challenging due to differing objectives.

Purpose of the Study:

• To develop an illumination beam tailored to enhance deep learning model performance for LSFM.
• To integrate the LSFM illumination model with a variable phase mask into deep learning training.
• To improve cell detection accuracy in large-scale LSFM datasets.

Main Methods:

• A deep learning-based approach was used to jointly optimize the illumination beam and a phase mask.
• The physical LSFM illumination model was integrated into the training of a cell detection network.
• Simulations and experimental validations were performed to assess imaging quality and cell detection.

Main Results:

• Joint optimization of the phase mask and deep learning model continuously improved image quality.
• The developed method demonstrated substantial enhancements in imaging quality compared to traditional Gaussian light sheets.
• Improved image quality led to significantly better cell detection performance.

Conclusions:

• Designing microscopy illumination through a computational, deep learning-driven approach can optimize imaging performance.
• This method offers a novel strategy for advancing optical design in microscopy.
• The tailored illumination beam boosts the efficacy of deep learning models for analyzing large imaging datasets.