Affinity Feature Strengthening for Accurate, Complete and Robust Vessel Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Image Analysis

Background:

• Vessel segmentation is vital for diagnosing conditions like coronary stenoses, retinal diseases, and brain aneurysms.
• Current methods struggle to achieve high accuracy, complete topology, and contrast robustness simultaneously.

Purpose of the Study:

• To present a novel approach, the Affinity Feature Strengthening Network (AFN), for robust and accurate vessel segmentation.
• To jointly model geometry and refine pixel-wise segmentation features using a contrast-insensitive, multiscale affinity approach.

Main Methods:

• Developed the Affinity Feature Strengthening Network (AFN) utilizing a multiscale affinity approach.
• Computed a multiscale affinity field for each pixel to capture semantic relationships and local geometry.
• Learned spatial- and scale-aware adaptive weights to enhance vessel features.

Main Results:

• AFN demonstrated superior performance on four diverse vascular datasets (XCAD, PV, DSA, DRIVE).
• Achieved higher accuracy and improved topological metrics compared to state-of-the-art methods.
• Exhibited enhanced robustness to variations in image contrast.

Conclusions:

• The AFN effectively addresses the challenges in medical vessel segmentation.
• The proposed method offers a promising solution for accurate and robust vessel analysis in various medical imaging applications.