Laser & Optoelectronics Progress, Volume. 60, Issue 6, 0610011(2023)
Multiscale Dense Attention Network for Retinal Vessel Segmentation
Figures & Tables(13)
Fig. 6. Segmentation results of different algorithms. (a) Original image; (b) label; (c) proposed MSDA-Net; (d) U-Net; (e) Dense U-Net; (f) MFI-Net
Fig. 7. Detail comparison of segmentation results. (a) Original image; (b) details of original image; (c) details of label; (d) proposed MSDA-Net; (e) U-Net; (f) Dense U-Net; (g) MFI-Net
Fig. 8. Comparison of ROC curve and PR curve of different algorithms on DRIVE dataset. (a) ROC; (b) PR
Fig. 9. Comparison of ROC curve and PR curve of different algorithms on CHASE_DB1 dataset. (a) ROC; (b) PR
Table 1. Average performance index evaluation results on different datasets
View tableTable 1. Average performance index evaluation results on different datasets
Dataset Method F A Se Sp AUC（ROC） AUC（PR） DRIVE U-Net 0.8461 0.9630 0.8302 0.9815 0.9852 0.9305 Dense U-Net 0.8504 0.9639 0.8366 0.9817 0.9861 0.9336 MFI-Net 0.8487 0.9638 0.8291 0.9826 0.9856 0.9325 MSDA-Net 0.8549 0.9650 0.8417 0.9822 0.9873 0.9374 CHASE_DB1 U-Net 0.8113 0.9656 0.8161 0.9805 0.9850 0.8945 Dense U-Net 0.8127 0.9657 0.8209 0.9801 0.9853 0.8962 MFI-Net 0.8147 0.9661 0.8237 0.9802 0.9813 0.8842 MSDA-Net 0.8172 0.9662 0.8334 0.9795 0.9867 0.9021 STARE U-Net 0.8195 0.9648 0.7525 0.9900 0.9846 0.9189 Dense U-Net 0.8188 0.9645 0.7563 0.9892 0.9841 0.9171 MFI-Net 0.8173 0.9641 0.7559 0.9889 0.9852 0.9162 MSDA-Net 0.8338 0.9675 0.8039 0.9867 0.9888 0.9281
Table 2. Performance index comparison between the proposed algorithm and other advanced algorithms
View tableTable 2. Performance index comparison between the proposed algorithm and other advanced algorithms
Method DRIVE CHASE_DB1 STARE Se Sp A Se Sp A Se Sp A Method in Ref.［ 12 ］0.7861 0.9712 0.9466 0.7644 0.9716 0.9502 0.7882 0.9729 0.9547 Method in Ref.［ 13 ］0.7653 0.9818 0.9542 0.7633 0.9809 0.9610 0.7581 0.9846 0.9612 Method in Ref.［ 14 ］0.7632 0.9536 0.7815 0.9587 0.7423 0.9603 Method in Ref.［ 15 ］0.7631 0.9820 0.9538 0.7641 0.9806 0.9607 0.7735 0.9857 0.9638 Method in Ref.［ 16 ］0.7800 0.9806 0.9551 0.7888 0.9801 0.9627 0.8201 0.9843 0.9674 Method in Ref.［ 17 ］0.8062 0.9769 0.9547 0.8135 0.9762 0.9617 0.8308 0.9784 0.9593 Method in Ref.［ 18 ］0.7941 0.9798 0.9558 0.8176 0.9704 0.9608 0.7598 0.9878 0.9640 Method in Ref.［ 19 ］0.7352 0.9775 0.9480 0.7279 0.9658 0.9452 0.7265 0.9759 0.9548 Method in Ref.［ 20 ］0.8353 0.9751 0.9579 0.8176 0.9776 0.9632 0.7946 0.9821 0.9626 MSDA-Net 0.8417 0.9822 0.9650 0.8334 0.9795 0.9662 0.8039 0.9867 0.9675
Table 3. Comparison of results before and after algorithm improvement
View tableTable 3. Comparison of results before and after algorithm improvement
No. DCM scSE CDCM F A Se Sp AUC（ROC） AUC（PR） 1 0.8461 0.9630 0.8302 0.9815 0.9852 0.9305 2 √ 0.8533 0.9645 0.8410 0.9818 0.9867 0.9362 3 √ √ 0.8539 0.9647 0.8424 0.9817 0.9869 0.9364 4 √ √ 0.8537 0.9646 0.8414 0.9818 0.9867 0.9361 5 √ 0.8507 0.9641 0.8338 0.9823 0.9857 0.9332 6 √ √ 0.8511 0.9642 0.8332 0.9826 0.9862 0.9342 7 √ √ √ 0.8549 0.9650 0.8416 0.9822 0.9873 0.9374
Tools
Get Citation
Copy Citation Text
Liming Liang, Jie Yu, Longsong Zhou, Xin Chen, Jian Wu. Multiscale Dense Attention Network for Retinal Vessel Segmentation[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0610011
Paper Information
Category: Image Processing
Received: Nov. 30, 2021
Accepted: Jan. 21, 2022
Published Online: Mar. 16, 2023
The Author Email: Jian Wu (wujian@jxust.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20