DEEP-squared: deep learning powered De-scattering with Excitation Patterning.

Area of Science:

• Biomedical Optics
• Microscopy
• Deep Learning

Background:

• Nonlinear optical microscopy, particularly point scanning multiphoton microscopy, faces throughput limitations for in vivo deep tissue imaging.
• Existing widefield imaging modalities are faster but typically limited to optically cleared or thin specimens.
• Previous widefield methods like DEEP (De-scattering with Excitation Patterning) encoded spatial information but required hundreds of patterned excitations for de-scattering at depth.

Purpose of the Study:

• To introduce DEEP², a deep learning-based model designed to accelerate de-scattering in deep tissue imaging.
• To significantly improve the throughput of widefield nonlinear optical microscopy.
• To enable faster in vivo imaging of biological structures at greater depths.

Main Methods:

• Development of DEEP², a deep learning model utilizing patterned multiphoton excitation.
• Training and application of the model to de-scatter images using significantly fewer patterned excitations compared to previous methods.
• Validation through numerical simulations and experimental imaging studies, including in vivo mouse models.

Main Results:

• DEEP² successfully de-scatters images using only tens of patterned excitations, a reduction from hundreds required previously.
• Achieved an improvement in throughput by nearly an order of magnitude compared to the original DEEP method.
• Demonstrated effective in vivo cortical vasculature imaging up to 4 scattering lengths deep in live mice.

Conclusions:

• DEEP² represents a substantial advancement in deep tissue imaging, overcoming throughput limitations of traditional methods.
• The deep learning approach enables efficient de-scattering, making widefield nonlinear optical microscopy more practical for in vivo applications.
• This technology holds promise for accelerated and deeper visualization of biological processes in living organisms.