Reproduce “SPNet: A novel deep neural network for retinal vessel segmentation based on shared decoder and pyramid-like loss” on OpenResearch

About this paper

Geng-Xin Xu, Chuan-Xian Ren

SPNet, a deep neural network, improves retinal vessel segmentation by combining a shared decoder and a pyramid-like loss function to accurately delineate capillaries and vessel edges. The system exhibits high computational efficiency and strong generalization across various retinal imaging datasets.

Key insights

• Weight-sharing in decoder modules efficiently captures multi-scale features by promoting knowledge sharing between global and local structures, optimizing parameter usage and preventing overfitting.
• Explicitly decomposing segmentation targets into residual pyramid maps guides the network to progressively identify and compensate for errors, particularly for fine vessels and blurred boundaries.
• Applying a multi-stage loss function that targets both global segmentation and specific residual information from hidden layers significantly enhances feature extraction and model robustness.

Summary by alphaXiv.