Neighbor2Mean: a self-supervised denoising method via local spatial redundancy for live-cell fluorescence microscopy

Area of Science:

• * Live-cell fluorescence microscopy
• * Deep learning for image analysis
• * Biological imaging techniques

Background:

• * Live-cell microscopy requires low light to prevent phototoxicity, leading to low signal-to-noise ratio (SNR) images.
• * Supervised deep learning needs high-SNR ground truth data, which is challenging for dynamic live-cell experiments.
• * Existing self-supervised methods have limitations in dynamic samples or introduce bias, hindering fine structure recovery.

Purpose of the Study:

• * To develop a self-supervised denoising method for live-cell fluorescence microscopy.
• * To improve image quality and SNR without relying on paired ground truth data.
• * To overcome limitations of existing temporal and spatial self-supervised denoising techniques.

Main Methods:

• * Introduction of Neighbor2Mean (N2M), a self-supervised denoising algorithm.
• * N2M exploits local spatial redundancy in single noisy images.
• * Utilizes local mean aggregation to construct low-variance pseudo-targets for training.

Main Results:

• * N2M achieves significant denoising performance in live-cell fluorescence microscopy.
• * Demonstrated a 5.59 dB PSNR gain compared to state-of-the-art self-supervised methods.
• * Achieved a 0.16 SSIM improvement over the SN2N baseline, indicating enhanced structural fidelity.

Conclusions:

• * N2M offers an effective self-supervised approach for live-cell image denoising.
• * The method successfully enhances image quality by leveraging local spatial information.
• * N2M shows potential for improving visualization of fine subcellular structures in low-light microscopy.