HD2Net: A Deep Learning Framework for Simultaneous Denoising and Deaberration in Fluorescence Microscopy

Abstract

Fluorescence microscopy is essential for biological research, offering high-contrast imaging of microscopic structures. However, the quality of these images is often compromised by optical aberrations and noise, particularly in low signal-to-noise ratio (SNR) conditions. While adaptive optics (AO) can correct aberrations, it requires costly hardware and slows down imaging; whereas current denoising approaches boost the SNR but leave out the aberration compensation. To address these limitations, we introduce HD2Net, a deep learning framework that enhances image quality by simultaneously denoising and suppressing the effect of aberrations without the need for additional hardware. Building on our previous work, HD2Net incorporates noise estimation and aberration removal modules, effectively restoring images degraded by noise and aberrations. Through comprehensive evaluation of synthetic phantoms and biological data, we demonstrate that HD2Net outperforms existing methods, significantly improving image resolution and contrast. This framework offers a promising solution for enhancing biological imaging, particularly in challenging aberrating and low-light conditions.