Slicing Network for Wide-Field Fluorescence Image Based on the Improved U-Net Model.

Area of Science:

• Biomedical imaging
• Optical microscopy
• Deep learning applications

Background:

• Wide-field fluorescence microscopy generates out-of-focus background noise.
• Scattering in thick tissues further degrades image quality.
• Existing methods are insufficient for varying defocus levels.

Purpose of the Study:

• To develop and evaluate deep learning networks for deblurring fluorescence images.
• To improve image quality in 2D and 3D wide-field fluorescence microscopy.
• To offer a cost-effective alternative to traditional confocal microscopy.

Main Methods:

• Utilized upU-Net, 3D U-Net, and 3D upU-Net architectures.
• Trained networks on 2D and 3D wide-field fluorescence images.
• Assessed network performance in reducing background noise and enhancing image clarity.

Main Results:

• Demonstrated significant enhancements in fluorescence image quality.
• Successfully reduced out-of-focus background noise.
• Showcased the potential for more economical confocal microscopy.

Conclusions:

• UpU-Net and 3D U-Net models effectively address defocusing issues in fluorescence imaging.
• These deep learning approaches offer substantial benefits for biologists using wide-field microscopy.
• Advancements pave the way for more accessible high-quality biological imaging.