TPOT: TOPOLOGY PRESERVING OPTIMAL TRANSPORT IN RETINAL FUNDUS IMAGE ENHANCEMENT

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Topology

Background:

• Retinal fundus photography enhancement is vital for diagnosing and monitoring retinal diseases.
• Existing methods, including Generative Adversarial Networks (GANs), often fail to preserve complex blood vessel topology, leading to structural inaccuracies.

Purpose of the Study:

• To develop a novel topology-preserving training paradigm for retinal image enhancement.
• To improve the preservation of blood vessel structures in enhanced retinal images.

Main Methods:

• Proposed a training paradigm that regularizes blood vessel structures by minimizing differences in persistence diagrams.
• Introduced the Topology Preserving Optimal Transport (TPOT) framework.
• Utilized persistence diagrams to capture and preserve topological features.

Main Results:

• TPOT demonstrated superior performance in enhancing retinal image quality compared to state-of-the-art methods.
• The method significantly improved the accuracy of downstream blood vessel segmentation tasks.
• Experimental results on a large-scale dataset validated the effectiveness of TPOT.

Conclusions:

• The proposed TPOT framework effectively preserves topological information in retinal blood vessels.
• TPOT offers a significant advancement in retinal image enhancement for improved disease diagnosis and monitoring.
• The method shows promise for clinical applications in ophthalmology.