Noise reduction in ultra-low light digital holographic microscopy using neural networks

Abstract

Live cell imaging is challenging because the difficult balance of maintaining both cell viability and high signal to noise ratio throughout the entire imaging duration. Label free quantitative light microscopy techniques are powerful tools to image the volumetric activities in living cellular and sub-cellular biological systems, however there are minimal ways to identify phototoxicity. In this paper, we investigate the use of neural network to restore quantitative digital hologram micrographs at ultra-low light levels down to 0.06 𝑚𝑊/𝑐𝑚2 which approximately two orders of magnitude lower than sunlight. By developing an adaptive image restoration method specifically tailored for digital holograms, we demonstrated the 2x improvement in SSIM over existing denoising methods. This demonstration could open up new avenues for high resolution holographic microscopy using deep ultraviolet coherent sources and achieve high-resolution imaging with ultralow light illumination Show more