Vessel-Width-Based Metrics and Weight Masks for Retinal Blood Vessel Segmentation

• João Paulo M. Linaris USP
• Nina S. T. Hirata USP

Resumo

Visual inspection of fundus images is widely used to diagnose various eye diseases, such as diabetic retinopathy and glaucoma. Alterations in the blood vessel structure are known to be related to these and other diseases. Therefore, automatic segmentation of vessels is of great interest. Existing methods, in general, struggle in segmenting thin vessels accurately. Motivated by this, we propose weight masks built based on the thickness of the vessels to be used with the well-known binary-cross entropy (BCE) loss. We also propose segmentation performance metrics conditioned on the vessel thickness. Then, we experiment different compositions of the loss function with the U-Net architecture to better understand the impact of each weight mask, especially on the segmentation of thin vessels. The experimental results indicate that the proposed weight masks present better performance regarding thin vessel segmentation, while, at the same time, maintaining competitive overall performance.

Referências

L. Wei, X. Sun, C. Fan, R. Li, S. Zhou, and H. Yu, “The patho-physiological mechanisms underlying diabetic retinopathy,” Frontiers in Cell and Developmental Biology, vol. 10, 2022. [Online]. Available: [link]

O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2015, pp. 234–241.

C. Chen, J. H. Chuah, R. Ali, and Y. Wang, “Retinal vessel segmentation using deep learning: A review,” IEEE Access, vol. PP, pp. 1–1, 08 2021.

E. R. Dougherty and R. A. Lotufo, Hands-on Morphological Image Processing. SPIE Press, 2003.

PyTorch Contributors, “PyTorch: Tensors and Dynamic Neural Networks in Python with Strong GPU Acceleration,” [link], 2024, accessed: 2025-07-27.

——, “ReduceLROnPlateau,” [link], 2025, online documentation; accessed 2025-07-27.

NumPy Developers, “NumPy: The fundamental package for scientific computing with Python,” [link], 2024, accessed: 2025-07-27.

Scikit-learn Developers, “scikit-learn: Machine Learning in Python,” [link], 2024, accessed: 2025-07-28.

Developers of DSE-Skeleton-Pruning, “DSE,” [link], 2022, accessed: 2025-07-31.

X. Bai and L. J. Latecki, “Discrete skeleton evolution,” in Proceedings of the 6th International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition. Berlin, Heidelberg: Springer-Verlag, 2007, p. 362–374.