Masked Vascular Structure Segmentation and Completion in Retinal Images.

Area of Science:

• Ophthalmology and Medical Imaging
• Computer Vision and Machine Learning

Background:

• Diabetic retinopathy and other diseases cause early microscopic retinal vascular changes.
• Accurate micro-level evaluation of retinal vasculature is crucial for understanding angiopathology and aiding clinical assessment.
• Current imaging modalities struggle to fully extract fragmented microvascular networks due to limited contrast and resolution.

Purpose of the Study:

• To develop a robust method for reconstructing complete retinal vascular networks from fragmented imaging data.
• To improve the segmentation accuracy and completeness of retinal microvascular analysis.
• To reduce the need for extensive manual data labeling in retinal image analysis.

Main Methods:

• Proposed a backbone-agnostic Masked Vascular Structure Segmentation and Completion (MaskVSC) method.
• Simulated missing vessel segments to train a model for predicting and connecting these gaps.
• Introduced a connectivity loss function to penalize interruptions in the vascular network.

Main Results:

• MaskVSC effectively reconstructs interconnected vascular networks, with optimal performance achieved when masking 40% of segments.
• The method demonstrated superior performance over state-of-the-art techniques in maintaining vascular completeness and segmentation accuracy across diverse datasets.
• MaskVSC improved performance when integrated with various segmentation backbones.

Conclusions:

• MaskVSC offers a significant advancement in reconstructing retinal vascular networks, addressing limitations of current imaging techniques.
• The method enhances the potential for improved disease assessment and management in ophthalmology through more accurate vascular analysis.
• The developed approach is versatile and applicable across different retinal imaging types and segmentation architectures.