Neovascularization Segmentation via a Multilateral Interaction-Enhanced Graph Convolutional Network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Choroidal neovascularization (CNV) is a key feature of wet age-related macular degeneration (AMD), a leading cause of blindness.
• Optical coherence tomography angiography (OCTA) is vital for analyzing CNV but faces segmentation challenges due to image artifacts and limited datasets.
• Accurate segmentation of CNV regions and vessels is critical for clinical assessment of wet AMD.

Purpose of the Study:

• To address the lack of public datasets for CNV analysis.
• To develop a novel network for accurate segmentation of CNV regions and vessels in OCTA images.
• To improve the clinical assessment of wet AMD through enhanced image analysis.

Main Methods:

• Construction of the first publicly accessible CNV dataset (CNVSeg).
• Proposal of a multilateral graph convolutional interaction-enhanced CNV segmentation network (MTG-Net).
• Integration of region and vessel morphology using graph-based cross-task modules (MIGR and MRGR) and an uncertainty-weighted loss function.

Main Results:

• MTG-Net achieved superior performance compared to existing methods.
• Achieved a Dice score of 87.21% for CNV region segmentation.
• Achieved a Dice score of 88.12% for CNV vessel segmentation.

Conclusions:

• MTG-Net effectively segments CNV regions and vessels by integrating morphological information and utilizing graph reasoning.
• The developed CNVSeg dataset and MTG-Net offer significant advancements for wet AMD research and clinical practice.
• The proposed method demonstrates robustness against imaging artifacts and noise, enhancing segmentation accuracy.