Deep Learning-Assisted Multiphoton Microscopy to Reduce Light Exposure and Expedite Imaging in Tissues With High and...

Area of Science:

• Biomedical Imaging
• Optical Microscopy
• Deep Learning

Background:

• Two-photon excitation fluorescence (2PEF) imaging provides insights into tissue function.
• Phototoxicity necessitates reduced light exposure during 2PEF imaging.
• Lowering excitation light compromises image quality by reducing fluorescence emission.

Purpose of the Study:

• To apply deep learning (DL) super-resolution techniques to 2PEF images acquired with low light exposure.
• To generate high-resolution images of retinal and skin tissues from low-light 2PEF data.

Main Methods:

• Analyzed U-Net and patch-based regression DL methods.
• Used paired low- and high-resolution skin (550 images) and retina (1200 images) datasets.
• Evaluated DL performance using Mean Squared Error (MSE) and Structural Similarity Index Measure (SSIM).

Main Results:

• For skin, the patch-based method yielded lower MSE (3.768) and higher SSIM (0.824) than U-Net.
• For retina, the patch-based method achieved lower MSE (27,611) and higher SSIM (0.636) than U-Net.
• The patch-based method was computationally slower (303 seconds) than U-Net (<1 second).

Conclusions:

• Deep learning effectively reduces excitation light exposure in 2PEF imaging.
• DL preserves critical image quality metrics despite reduced light levels.
• DL methods facilitate the translation of 2PEF imaging for in vivo applications in light-sensitive tissues.