Abstract | ||
---|---|---|
•We propose an automatic end-to-end topology and width aware network for artery/vein classification, which, for the first time, integrates topology and vessel width information into the deep learning framework to boost the A/V classification performance.•A topology-aware module is proposed to increase the topological connectivity of the segmented artery and vein maps, which contains a topology-ranking discriminator and a topology preserving regularization module.•A width-aware module is designed to extract features related to the vessel width via predicting the width maps for the dilated/non-dilated ground truth A/V masks to enhance the model’s perception of the vessel width, which is regularized by a width-aware loss.•The proposed framework is validated quantitatively and qualitatively on three publicly available datasets, including AV-DRIVE, INSPIRE-AVR and high-resolution fundus (HRF). In addition, we have manually annotated the pixel-wise artery and vein classification labels for the HRF dataset, which will be released for public access.•Source code of the proposed framework will be made publicly available. |
Year | DOI | Venue |
---|---|---|
2022 | 10.1016/j.media.2021.102340 | Medical Image Analysis |
Keywords | DocType | Volume |
Retinal images,Artery/vein classification,Deep learning,Topological connectivity,Generative adversarial network | Journal | 77 |
ISSN | Citations | PageRank |
1361-8415 | 2 | 0.39 |
References | Authors | |
0 | 8 |
Name | Order | Citations | PageRank |
---|---|---|---|
Wenting Chen | 1 | 3 | 0.74 |
Shuang Yu | 2 | 7 | 3.15 |
Kai Ma | 3 | 49 | 18.48 |
Wei Ji | 4 | 2 | 1.41 |
Cheng Bian | 5 | 4 | 2.44 |
Chunyan Chu | 6 | 2 | 0.39 |
Linlin Shen | 7 | 1351 | 90.25 |
Yefeng Zheng | 8 | 1391 | 114.67 |