MG-UNet: multi-scale convolutional gate-based network for background-removal imaging in fluorescence microscopy.

Area of Science:

• * Biological imaging
• * Medical research
• * Computational microscopy

Background:

• * Wide-field fluorescence microscopy is crucial for visualizing biological structures.
• * Out-of-focus blur and background noise degrade image quality and axial resolution in wide-field microscopy.
• * Enhancing image clarity and resolution is essential for accurate analysis.

Purpose of the Study:

• * To introduce a novel deep neural network, multiscale convolutional gated UNet (MG-UNet), for fluorescence image enhancement.
• * To improve the contrast and sharpness of wide-field fluorescence microscopy images.
• * To provide a computationally efficient solution for high-quality biological imaging.

Main Methods:

• * Developed MG-UNet, a deep neural network utilizing multi-scale convolutional gate modules for coordinate encoding.
• * Implemented a combination of convolutional filters at different scales to preserve spatial information and enhance efficiency.
• * Adapted the 2D image restoration model for lightweight 3D applications using spatial-channel transformation operators.

Main Results:

• * MG-UNet demonstrated superior performance over state-of-the-art models in both 2D and 3D fluorescence microscopy image restoration.
• * Achieved enhanced image quality, characterized by higher contrast and sharpness.
• * Showcased lower computational costs compared to standard UNet architectures.

Conclusions:

• * MG-UNet effectively enhances wide-field fluorescence microscopy images, overcoming limitations of traditional methods.
• * The model offers a significant advancement in biological and medical imaging, providing clearer and more resolved visualizations.
• * MG-UNet presents a promising tool for researchers requiring high-fidelity imaging with improved computational efficiency.