STSRNet: Self-Texture Transfer Super-Resolution and Refocusing Network.

Area of Science:

• Biomedical Imaging
• Computational Microscopy
• Artificial Intelligence in Medicine

Background:

• High-resolution (HR) biomedical images with axial information are crucial for diagnosis but are time-consuming and expensive to acquire.
• Existing methods often require specialized equipment or scanning, limiting practical applications.

Purpose of the Study:

• To develop a novel network, STSRNet, for reconstructing HR multi-focal plane (MFP) images from a single 2D low-resolution (LR) wide-field image.
• To enable SR and refocusing without scanning or special devices, making advanced imaging more accessible.

Main Methods:

• Proposed STSRNet with three modules: backbone for feature extraction, self-texture transfer for feature fusion, and flexible reconstruction for SR and refocusing.
• The self-texture transfer module leverages image self-similarity, particularly in cytological images.
• The reconstruction module uses pluggable components for efficient, simultaneous SR and refocusing across multiple focal planes.

Main Results:

• Extensive experiments on cytological images demonstrated that STSRNet-reconstructed MFP images exhibit richer axial and horizontal details compared to input LR images.
• Reconstructed MFP images showed improved performance on high-level tasks compared to single 2D wide-field images.
• The method successfully generates high-quality MFP images, expanding the utility of LR wide-field images.

Conclusions:

• STSRNet offers a practical solution for obtaining high-quality HR MFP images when direct acquisition is infeasible.
• The developed method significantly enhances the application potential of LR wide-field biomedical images.
• A new cytology dataset, RSDC, has been released to facilitate further research in this area.