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

Optical sectioning endo-microscopy has become a crucial tool for deep brain imaging, but conventional methods face challenges such as time-consuming scanning processes and the need for expensive light sources. HiLo imaging addresses these issues by providing faster acquisition and high-quality images. In this study, we introduce a histogram matching-based Res-UNet model for optical sectioning HiLo endo-microscopy. By applying our model, we achieve substantial improvements in image reconstruction quality compared to the conventional ResNet model. Our evaluation demonstrates significant enhancements in the reconstructed images' structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) metrics. The PSNR values exceeding 30 dB and SSIM values surpassing 0.8 at various depths indicate that our method achieves image quality comparable to the HiLo system. Importantly, while our approach demonstrates high-quality, real-time reconstruction capabilities using ex-vivo samples, we are actively planning to extend our research to in-vivo imaging applications, which will further enhance the practical implications of our work.