Mu-net: Multi-scale U-net for two-photon microscopy image denoising and restoration

Area of Science:

• Neuroscience
• Biomedical Imaging
• Computer Science

Background:

• Two-photon microscopy (2PM) enables 3D neural imaging of deep cortical regions.
• 2PM images often suffer from noise (blur, white noise, photo bleaching) and limited processing effectiveness due to laser scanning artifacts.
• Existing denoising methods struggle with the unique noise characteristics of 2PM data.

Purpose of the Study:

• To develop an advanced algorithm for denoising three-dimensional (3D) images acquired via two-photon microscopy (2PM).
• To enhance the quality and clarity of neural imaging data from deep brain tissues.

Main Methods:

• A novel algorithm based on deep convolutional neural networks (CNNs) was proposed.
• The model utilizes multiple U-nets in a coarse-to-fine strategy for multi-scale noise removal.
• Fully 3D convolutional operations enable end-to-end learning without pre/post-processing.

Main Results:

• The proposed CNN algorithm demonstrated substantial performance improvements in 2PM image denoising.
• The multi-scale, coarse-to-fine approach effectively addressed noise challenges inherent in 2PM.
• The method outperformed existing baseline denoising techniques.

Conclusions:

• The developed deep learning algorithm significantly enhances the quality of 2PM 3D neural images.
• This approach offers a robust solution for noise reduction in deep tissue neuroimaging.
• The end-to-end learning capability simplifies the processing pipeline for researchers.