Reference-Based OCT Angiogram Super-Resolution With Learnable Texture Generation.

Area of Science:

• Ophthalmology
• Biomedical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography angiography (OCTA) is crucial for visualizing retinal microvasculature and identifying disease biomarkers.
• Increasing the field-of-view (FOV) in OCTA scans typically reduces resolution due to fixed acquisition times.
• Existing methods face limitations in maintaining resolution across larger scanning areas.

Purpose of the Study:

• To develop a novel reference-based super-resolution (RefSR) framework to enhance OCTA image resolution while expanding the field-of-view.
• To introduce a learnable texture generator (LTG) that generates textures for super-resolution, overcoming limitations of traditional RefSR models.
• To create a robust OCTA super-resolution method that does not rely on a reference image during inference.

Main Methods:

• Proposed a novel reference-based super-resolution (RefSR) framework utilizing a learnable texture generator (LTG).
• Trained the LTG using textures from a normal RefSR pipeline to generate textures dynamically.
• Developed LTGNet, which generates textures internally, eliminating the need for a reference image during inference.

Main Results:

• The proposed LTGNet demonstrated competitive performance and robustness compared to state-of-the-art methods.
• Experimental and visual results confirmed the framework's ability to preserve resolution while increasing the scanning area.
• The method proved invulnerable to the selection of a reference image, enhancing reliability.

Conclusions:

• The novel LTGNet framework offers a promising solution for high-resolution OCTA imaging across larger fields-of-view.
• This approach enhances the reliability and potential for real-life deployment in diagnosing retinal diseases.
• The developed method expands the texture space for super-resolution beyond single reference images.