Opto-Electronic Engineering, Volume. 50, Issue 1, 220116(2023)
Boundary attention assisted dynamic graph convolution for retinal vascular segmentation
Figures & Tables(11)
Fig. 1. Boundary attention assisted dynamic graph convolution U-shaped network
Fig. 3. Dynamic graph convolution calculation process and Feature fusion network. (a) Dynamic graph convolution calculation process; (b) Feature fusion network
Fig. 5. Data preprocessing results. (a) Pre-processed image slices; (b) Ground truth slices
Fig. 6. Comparison of ablation results. (a) Original image and details; (b) Ground truth; (c) U-Net; (d) DGU-Net; (e) BU-Net; (f) BDGU-Net
Fig. 7. Comparison of ablation results. (a) Original image and details; (b) Ground truth; (c) Iternet; (d) MLA-DU-Net; (e) Res2Unet; (f) BDGU-Net
Table 1. Evaluation indexes
Table 2. Ablation experiments results
View tableView in ArticleTable 2. Ablation experiments results
Dataset Network F1 SE SP ACC AUC Time/s DRIVE U-Net 0.8353 0.8330 0.9777 0.9601 0.9848 1.00 DGU-Net 0.8357 0.8275 0.9789 0.9605 0.9850 1.22 BU-Net 0.8355 0.8352 0.9773 0.9600 0.9849 1.00 BDGU-Net 0.8359 0.8300 0.9785 0.9604 0.9851 1.24 CHASEDB1 U-Net 0.8108 0.8052 0.9820 0.9660 0.9854 1.31 DGU-Net 0.8053 0.7812 0.9842 0.9658 0.9846 1.73 BU-Net 0.8133 0.8210 0.9803 0.9659 0.9858 1.31 BDGU-Net 0.8136 0.8146 0.9813 0.9662 0.9856 1.76 STARE U-Net 0.7845 0.7380 0.9863 0.9648 0.9810 2.21 DGU-Net 0.7824 0.7045 0.9908 0.9660 0.9825 2.85 BU-Net 0.7845 0.7352 0.9868 0.9650 0.9806 2.21 BDGU-Net 0.7934 0.7273 0.9900 0.9672 0.9834 2.85
Table 3. Value analysis of different α and β coefficients
View tableView in ArticleTable 3. Value analysis of different α and β coefficients
α β F1 SE SP ACC AUC 0.8 0.2 0.8348 0.8252 0.9790 0.9604 0.9848 0.7 0.3 0.8351 0.8230 0.9795 0.9605 0.9848 0.6 0.4 0.8359 0.8300 0.9785 0.9604 0.9851 0.5 0.5 0.8351 0.8284 0.9785 0.9603 0.9849
Table 4. Index comparison of different networks in DRIVE, CHASEDB1 and STARE datasets
View tableView in ArticleTable 4. Index comparison of different networks in DRIVE, CHASEDB1 and STARE datasets
Dataset Network Year F1 SE SP ACC AUC Time/s DRIVE 2nd Human Observer[ 20 ]— — 0.7760 0.9724 0.9472 — — Iternet[ 21 ]2020 0.8353 0.8370 0.9770 0.9600 0.9847 1.37 Yang[ 22 ]2021 0.8297 0.8353 0.9751 0.9579 — — MLA-DU-Net[ 23 ]2021 0.8352 0.8269 0.9788 0.9603 0.9849 1.64 Res2Unet[ 24 ]2022 0.8292 0.8332 0.9756 0.9583 0.9782 1.21 Zhang[ 15 ]2022 0.8349 0.8345 0.9773 0.9600 0.9850 1.20 BDGU-Net 2022 0.8359 0.8300 0.9785 0.9604 0.9851 1.24 CHASEDB1 2nd Human Observer[ 20 ]— — 0.8105 0.9711 0.9545 — — Iternet[ 21 ]2020 0.8120 0.8144 0.9809 0.9658 0.9856 1.48 Yang[ 22 ]2021 0.7997 0.8176 0.9776 0.9632 — — MLA-DU-Net[ 23 ]2021 0.8063 0.8137 0.9796 0.9646 0.9841 2.13 Res2Unet[ 24 ]2022 0.8071 0.8444 0.9753 0.9634 0.9794 1.47 Zhang[ 15 ]2022 0.8050 0.8141 0.9792 0.9643 0.9841 1.72 BDGU-Net 2022 0.8136 0.8146 0.9813 0.9662 0.9856 1.76 STARE 2nd Human Observer[ 20 ]— — 0.8952 0.9384 0.9349 — — Iternet[ 21 ]2020 0.7962 0.7490 0.9874 0.9667 0.9838 2.76 Yang[ 22 ]2021 0.8155 0.7946 0.9821 0.9626 — — MLA-DU-Net[ 23 ]2021 0.7965 0.7925 0.9812 0.9649 0.9826 3.95 Res2Unet[ 24 ]2022 0.7858 0.7392 0.9865 0.9650 0.9709 2.77 Zhang[ 15 ]2022 0.7878 0.7381 0.9871 0.9655 0.9823 3.07 BDGU-Net 2022 0.7934 0.7273 0.9900 0.9672 0.9834 2.85
Tools
Get Citation
Copy Citation Text
Jia Lv, Zeyu Wang, Haocheng Liang. Boundary attention assisted dynamic graph convolution for retinal vascular segmentation[J]. Opto-Electronic Engineering, 2023, 50(1): 220116
Paper Information
Category: Article
Received: Jun. 8, 2022
Accepted: Sep. 27, 2022
Published Online: Feb. 27, 2023
The Author Email: Lv Jia (lvjia@cqnu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20