Author Affiliations
Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211, Chinashow less
Fig. 1. Overall framework of proposed method
Fig. 2. Input image and corresponding ground truth in training stage. (a) Input images; (b) ground truth for training WholeSegmentNet; (c) ground truth for training ThinSegmentNet (dark color area in picture)
Fig. 3. Residual network structure
Fig. 4. Results of database partitioning of DRIVE, STARE and CHASE_DB1. (a) Original images; (b) ground truth; (c) segmented whole vessel images; (d) segmented small vessel images; (e) fusion results; (f) results of proposed method
Fig. 5. ROC curves with different databases. (a) DRIVE; (b) STARE; (c) CHASE_DB1
Fig. 6. Effect of ThinSegmentNet segmentation and post-processing on segmentation results. (a) Ground truth; (b) vessel images of WholeSegmentNet predictions; (c) vessel images of WholeSegmentNet+ThinSegmentNet predictions; (d) results of proposed method
Fig. 7. Segmentation of vessels in different areas. (a) Segmentation of vessels in pathological areas; (b) segmentation of vessels in central line reflex areas; (c) segmentation of vessels in low contrast areas
Database | Method | Rse | Rsp | RAcc | AUC |
---|
| WholeSegmentNet | 0.7578 | 0.9816 | 0.9531 | 0.9732 | DRIVE | Whole+ThinSegmentNet | 0.7971 | 0.9736 | 0.9522 | 0.9702 | | Proposed method | 0.8062 | 0.9769 | 0.9547 | 0.9739 | | WholeSegmentNet | 0.7719 | 0.9828 | 0.9610 | 0.9766 | STARE | Whole+ThinSegmentNet | 0.8095 | 0.9757 | 0.9586 | 0.9737 | | Proposed method | 0.8308 | 0.9784 | 0.9593 | 0.9788 | | WholeSegmentNet | 0.7689 | 0.9801 | 0.9609 | 0.9778 | CHASE_DB1 | Whole+ThinSegmentNet | 0.7997 | 0.9731 | 0.9601 | 0.9764 | | Proposed method | 0.8135 | 0.9762 | 0.9617 | 0.9782 |
|
Table 1. Evaluation results of small vessel segmentation and post-processing methods in DRIVE, STARE and CHASE_DB1 test sets
Method | DRIVE | STARE | CHASE_DB1 | | |
---|
Rse | Rsp | | | RAcc | AUC | Rse | Rsp | RAcc | AUC | Rse | Rsp | RAcc | AUC |
---|
Ref. [22] | 0.7395 | 0.9782 | 0.9494 | 0.9672 | 0.7317 | 0.9842 | 0.956 | 0.9673 | 0.7615 | 0.9575 | 0.9467 | 0.9623 | Ref. [5] | 0.7655 | 0.9704 | 0.9442 | 0.9614 | 0.7716 | 0.9701 | 0.9497 | 0.9563 | 0.7585 | 0.9587 | 0.9387 | 0.9487 | Ref. [23] | 0.7420 | 0.9820 | 0.9540 | 0.8620 | 0.7800 | 0.9780 | 0.9560 | 0.8740 | | | | | Ref. [8] | 0.7743 | 0.9725 | 0.9476 | 0.9636 | 0.7791 | 0.9758 | 0.9554 | 0.9748 | 0.7626 | 0.9661 | 0.9452 | 0.9606 | Ref. [9] | 0.7569 | 0.9816 | 0.9527 | 0.9738 | 0.7726 | 0.9844 | 0.9628 | 0.9879 | 0.7507 | 0.9793 | 0.9581 | 0.9716 | Ref. [10] | 0.7897 | 0.9684 | | | 0.768 | 0.9738 | | | 0.7277 | 0.9712 | | | Ref. [24] | 0.7691 | 0.9801 | 0.9533 | 0.9744 | | | | | | | | | Ref. [14] | 0.7653 | 0.9818 | 0.9542 | 0.9752 | 0.7581 | 0.9846 | 0.9612 | 0.9801 | 0.7633 | 0.9809 | 0.9610 | 0.9781 | Proposed method | 0.8062 | 0.9769 | 0.9547 | 0.9739 | 0.8308 | 0.9784 | 0.9593 | 0.9788 | 0.8135 | 0.9762 | 0.9617 | 0.9782 |
|
Table 2. Comparison of retinal vessel segmentation results among different methods on DRIVE, STARE and CHASE_DB1 databases
Method | Rse | Rsp | RAcc |
---|
Ref. [18] | 0.6587 | 0.9565 | 0.9258 | Ref. [25] | 0.7262 | 0.9764 | 0.9511 | Ref. [9] | 0.7800 | 0.9805 | 0.9672 | Proposed method | 0.8137 | 0.9758 | 0.9609 |
|
Table 3. Evaluation results of 10 pathological images in STARE database