Histogram-based Res-UNet model for optical sectioning HiLo endo-microscopy

Area of Science:

• Biomedical Imaging
• Optical Microscopy
• Deep Learning

Background:

• Optical sectioning endo-microscopy is vital for deep brain imaging but faces limitations.
• Conventional methods are slow and require costly light sources.
• HiLo imaging offers faster acquisition and improved image quality.

Purpose of the Study:

• To introduce a novel histogram matching-based Res-UNet model for optical sectioning HiLo endo-microscopy.
• To improve image reconstruction quality and efficiency.
• To evaluate the model's performance against conventional methods.

Main Methods:

• Development of a histogram matching-based Res-UNet model.
• Application of the model to optical sectioning HiLo endo-microscopy.
• Comparative analysis with conventional ResNet models using SSIM and PSNR metrics.

Main Results:

• Substantial improvements in image reconstruction quality were achieved.
• Enhanced Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) metrics.
• PSNR > 30 dB and SSIM > 0.8 indicate image quality comparable to the HiLo system.

Conclusions:

• The histogram matching-based Res-UNet model offers high-quality, real-time reconstruction for optical sectioning.
• The method significantly outperforms conventional ResNet models.
• Future work will focus on extending the application to in-vivo imaging.