Deep learning-enabled filter-free fluorescence microscope

Area of Science:

• Biomedical Imaging
• Optical Microscopy
• Computational Biology

Background:

• Optical filters are crucial in fluorescence microscopy for signal isolation and noise reduction.
• Current optical filtering methods increase system complexity, size, and cost, hindering high-speed and multifluorescence imaging.
• There is a need for advanced techniques to simplify fluorescence microscopy while maintaining imaging quality.

Purpose of the Study:

• To develop a filter-free fluorescence microscopic imaging technique using deep learning.
• To enable automatic fluorescence channel selection and accurate fluorescence prediction post-acquisition.
• To demonstrate the potential of deep learning-based spectral filtering in biomedical applications.

Main Methods:

• Implemented a deep learning model for digital spectral filtering in fluorescence microscopy.
• Developed algorithms to compute color changes from spectral shifts and excitation scattering for fluorescence prediction.
• Validated the technique on cells and tissues labeled with various fluorophores across different magnification powers.

Main Results:

• Achieved filter-free fluorescence microscopic imaging with automatic channel selection.
• Demonstrated accurate fluorescence prediction with robust sensitivity and specificity.
• Obtained consistent results compared to the reference standard, validating the technique's reliability.

Conclusions:

• Deep learning-based digital spectral filtering offers a viable alternative to traditional optical filters in fluorescence microscopy.
• The developed technique simplifies microscopic systems, making them suitable for high-speed and multifluorescence imaging.
• This approach has broad potential for advancing other biomedical fields like cytometry and endoscopy.