Context-Aware Optimal Transport Learning for Retinal Fundus Image Enhancement.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Retinal fundus photography is vital for diagnosing eye diseases but susceptible to image quality issues.
• High-quality images are essential for accurate diagnosis and automated analysis.
• Existing enhancement methods often struggle with preserving local structures and minimizing artifacts.

Purpose of the Study:

• To propose a novel context-informed optimal transport (OT) learning framework for unpaired fundus image enhancement.
• To address limitations of standard generative methods in handling contextual information.

Main Methods:

• Developed a context-aware OT learning paradigm utilizing deep contextual features.
• Derived the context-aware OT using the earth mover's distance, providing theoretical guarantees.
• Formulated enhancement as a distribution alignment problem between low- and high-quality images.

Main Results:

• The proposed method significantly outperforms state-of-the-art supervised and unsupervised techniques.
• Demonstrated superiority in signal-to-noise ratio and structural similarity index.
• Showcased improved performance in two downstream diagnostic tasks.

Conclusions:

• The context-aware OT learning framework offers superior performance for unpaired fundus image enhancement.
• This approach effectively preserves local structures and minimizes artifacts, crucial for clinical applications.
• The method provides a robust solution for improving retinal image quality for disease diagnosis.