TP-Net: Two-Path Network for Retinal Vessel Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Ophthalmology

Background:

• Accurate retinal vessel segmentation is vital for early retinopathy detection.
• Current methods struggle with thin and low-contrast vessels, leading to mis-segmentations.

Purpose of the Study:

• To develop an advanced two-path retinal vessel segmentation network (TP-Net).
• To enhance segmentation accuracy, particularly for challenging thin and low-contrast vessels.

Main Methods:

• Proposed TP-Net with a main-path for trunk detection and a sub-path for edge information.
• Incorporated a global feature selection mechanism (GFSM) for low-contrast vessel enhancement.
• Introduced an edge feature extraction and loss function for thin vessel segmentation.
• Utilized a multi-scale feature aggregation module (MFAM) to fuse path predictions.

Main Results:

• TP-Net demonstrated superior performance and generalization across DRIVE, STARE, and CHASE DB1 datasets.
• Achieved refined segmentation by effectively combining trunk and edge information.
• Showcased improved accuracy for thin and low-contrast retinal vessels compared to state-of-the-art methods.

Conclusions:

• TP-Net offers a robust solution for automatic retinal vessel segmentation.
• The network effectively addresses limitations of existing methods in detecting subtle vascular structures.
• TP-Net presents a promising tool for computer-aided diagnosis of retinopathy with fewer parameters.