Hyperspectral LSFM with DMD-only shaping and neural network reconstruction

Area of Science:

• Biomedical imaging
• Microscopy
• Optical engineering

Background:

• Conventional light sheet fluorescence microscopy (LSFM) faces challenges in quantitative imaging due to autofluorescence and spectrally overlapping fluorophores.
• Limitations in spectral resolution and signal loss hinder accurate fluorophore quantification in complex biological samples.

Purpose of the Study:

• To introduce a novel hyperspectral LSFM (HLSFM) approach, DNR-HLSFM, for robust fluorophore quantification.
• To overcome the limitations of conventional LSFM in handling autofluorescence and spectral overlap.
• To enable versatile, high-resolution, quantitative hyperspectral imaging in biomedical research.

Main Methods:

• Developed DNR-HLSFM, combining DMD-only light shaping and neural network reconstruction (DNR).
• Integrated structured illumination with a physical acquisition model, a denoising neural network, and non-negative unmixing.
• Eliminated the need for optical filters, enhancing spectral resolution and minimizing signal loss.

Main Results:

• Successfully demonstrated DNR-HLSFM for 3D imaging of zebrafish embryos.
• Achieved effective removal of autofluorescence from biological samples.
• Enabled precise separation of two spectrally overlapping red fluorophores, validating quantitative accuracy.

Conclusions:

• DNR-HLSFM significantly expands the capabilities of LSFM for quantitative hyperspectral imaging.
• This computational technique offers a versatile solution for complex biological imaging challenges.
• Publicly available data and code promote reproducibility and further advancements in the field.