Retinal Vessel Segmentation via Self-Adaptive Compensation Network

Fig. 6. Visualization results on different datasets. (a) Original images; (b) ground truth by expert 1; (c) ground truth by expert 2; (d) segmentation results of U-Net; (e) segmentation results of SACom

Download full size

Fig. 7. Detail comparison of DRIVE dataset segmentation results. (a) Original images; (b) local images; (c) local images of ground truth 1; (d) local images of ground truth 2; (e) local images of SACom segmentation results

Download full size

Fig. 8. Visualization of AUC for SACom and other advanced methods

Download full size

Table 1. Performance comparison under different weights
View table
Table 1. Performance comparison under different weights
Weight A_cc S_e S_p F₁ AUC
1.0 0. 9703 0.8002 0.9866 0.8252 0.9878
1.5 0.9700 0.8156 0.9848 0.8263 0.9875
2.0 0.9695 0.8403 0.9819 0.8285 0.9880
2.5 0.9681 0.8549 0.9790 0.8243 0.9874
3.0 0.9669 0.8723 0.9759 0.8218 0.9878

Table 2. Comparison of different supervision and fusion strategies
View table
Table 2. Comparison of different supervision and fusion strategies
Strategy A_cc S_e S_p F₁ AUC
1 0.9688 0.8374 0.9814 0.8246 0.9871
2 0.9684 0.8369 0.9810 0.8227 0.9868
3 0.9692 0.8397 0.9816 0.8269 0.9875
4 0.9695 0.8403 0.9819 0.8285 0.9880

Table 3. Comparison of ablation experimental results

View table

Table 3. Comparison of ablation experimental results

Model	A_cc	S_e	S_p	F₁	AUC	Params /M
U-Net₆₄	0.9551	0.7841	0.9800	0.8167	0.9758	17.26
U-Net₃₂（Baseline）	0.9554	0.7849	0.9802	0.8175	0.9761	4.32
Baseline+DC₁₂₃	0.9642	0.7883	0.9810	0.7939	0.9789	4.20
Baeline+DC₂₃₄	0.9647	0.7914	0.9814	0.7972	0.9802	3.79
Baeline+DC₃₄₅	0.9612	0.7869	0.9779	0.7802	0.9765	3.87
Baeline+DC₂₃₄+AMAC	0.9667	0.8277	0.9801	0.8134	0.9849	4.17
Baeline+DC₂₃₄+AMAC+CCB	0.9695	0.8403	0.9819	0.8285	0.9880	6.01

Table 4. Performance comparison of SACom with different expert annotations on DRIVE test set
View table
Table 4. Performance comparison of SACom with different expert annotations on DRIVE test set
Label A_cc S_e S_p F₁ AUC
1 0.9695 0.8403 0.9819 0.8285 0.9800
2 0.9737 0.8713 0.9832 0.8484 0.9910

Table 5. Segmentation performance comparison of different methods

View table

Table 5. Segmentation performance comparison of different methods

Dataset	Method	A_cc	S_e	S_p	F₁	AUC
DRIVE	U-Net	0.9554	0.7849	0.9802	0.8175	0.9761
	DUNet	0.9566	0.7963	0.9800	0.8237	0.9802
	AA-UNet	0.9558	0.7941	0.9798	0.8216	0.9847
	CSU-Net	0.9565	0.8071	0.9782	0.8251	0.9801
	CMNet	0.9664	0.8069	0.9818	0.8064	0.9840
	SACom	0.9695	0.8403	0.9819	0.8285	0.9880
CHASE_DB1	U-Net	0.9576	0.7304	0.9717	0.7793	0.9782
	DUNet	0.9610	0.8155	0.9752	0.7883	0.9804
	AA-UNet	0.9608	0.8176	0.9704	0.7892	0.9865
	CSU-Net	0.9706	0.8427	0.9836	0.8105	0.9824
	CMNet	0.9742	0.8461	0.9836	0.8097	0.9879
	SACom	0.9763	0.8748	0.9831	0.8229	0.9917
STARE	U-Net	0.9636	0.7638	0.9867	0.8135	0.9789
	DUNet	0.9641	0.7595	0.9878	0.8143	0.9832
	AA-UNet	0.9640	0.7598	0.9878	0.8142	0.9824
	CSU-Net	0.9702	0.8432	0.9845	0.8516	0.9825
	SACom	0.9753	0.8506	0.9856	0.8394	0.9919

Tools

Get Citation

Copy Citation Text

Lin Zhang, Chuang Wu, Xinyu Fan, Chaoju Gong, Suyan Li, Hui Liu. Retinal Vessel Segmentation via Self-Adaptive Compensation Network[J]. Acta Optica Sinica, 2023, 43(14): 1418001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites