Optics and Precision Engineering, Volume. 33, Issue 4, 591(2025)
Pseudo-label confidence regulates semi-supervised semantic segmentation of pathological images of colorectal cancer
Figures & Tables(17)
Fig. 6. Comparison of segmentation results （1/4 labeled data） by different methods
Fig. 8. Visualization of results on LUAD-HistoSeg （1/4 labeled data）
Table 1. Experimental data set partitioning
View tableView in ArticleTable 1. Experimental data set partitioning
Dataset Size/pixel Train Test Full 1/4 1/8 1/16 CR-SEG 224×224 4 640 1 160 580 290 1 160 BCSS 224×224 6 720 1 680 840 420 1 680 LUAD 224×224 480 120 60 30 120
Table 2. Comparative experiments of different backbone networks（DeepLabV3+）
View tableView in ArticleTable 2. Comparative experiments of different backbone networks（DeepLabV3+）
Backbone network Labeled data mIoU/% mDice/% mAcc/% mPre/% Resnet50
1/4 74.09 85.04 82.86 85.32 1/8 72.77 84.14 81.97 84.57 1/16 71.80 83.44 81.47 83.97 Resnet101
1/4 72.99 84.28 82.22 84.19 1/8 71.40 83.18 81.14 82.68 1/16 71.04 82.94 80.74 82.88
Table 3. Comparative experiments of different basic networks（resnet50）
View tableView in ArticleTable 3. Comparative experiments of different basic networks（resnet50）
Basic network Labeled data mIoU/% mDice/% mAcc/% mPre/% DeepLabV3+
1/4 74.09 85.04 82.86 85.32 1/8 72.77 84.14 81.97 84.57 1/16 71.80 83.44 81.47 83.97 PSPNet
1/4 72.50 83.99 81.54 83.68 1/8 70.87 82.82 80.62 83.22 1/16 68.83 81.30 79.63 80.49 UNet
1/4 71.55 83.29 81.18 82.95 1/8 69.75 82.11 78.97 82.12 1/16 68.05 80.90 77.35 80.80
Table 4. Experimental comparison of different semi-supervised network segmentation methods based on DeepLabV3+（resnet50）
View tableView in ArticleTable 4. Experimental comparison of different semi-supervised network segmentation methods based on DeepLabV3+（resnet50）
Method Labeled data mIoU mDice mAcc mPre ClassIoU ClassDice BACK STR NORM TUM BACK STR NORM TUM ST 1/4
66.47 79.58 77.76 78.89 71.72 76.86 58.03 59.26 83.53 86.93 73.44 74.42 ST++ 67.66 80.50 78.31 80.49 71.73 76.93 61.61 60.35 83.54 86.96 76.24 75.27 CCT 68.87 81.41 79.08 81.92 70.86 78.12 64.19 62.30 82.94 87.72 78.19 76.77 ELN 72.91 84.23 82.10 84.60 73.51 81.16 69.00 67.97 84.73 89.60 81.66 80.93 U2PL 73.48 84.61 82.56 84.68 74.13 81.53 70.02 68.22 85.15 89.83 82.37 81.11 UniMatch 73.68 84.75 82.70 85.07 74.93 81.25 70.61 67.92 85.67 89.66 82.78 80.90 Ours 74.09 85.04 82.86 85.32 74.50 81.24 70.97 69.66 85.38 89.65 83.02 82.12 ST 1/8
66.82 79.90 77.73 79.20 71.84 75.57 60.84 59.06 83.61 86.08 75.65 74.26 ST++ 67.37 80.28 78.20 80.10 72.11 76.76 60.97 59.65 83.79 86.85 75.76 74.73 CCT 66.56 79.75 77.08 80.72 68.23 76.17 62.13 59.70 81.12 86.47 76.65 74.76 ELN 71.44 83.18 81.28 83.71 74.42 80.47 66.84 64.03 85.34 89.18 80.13 78.07 U2PL 71.70 83.40 81.27 83.54 73.63 79.89 67.02 66.25 84.81 88.82 80.26 79.70 UniMatch 72.62 84.01 82.21 83.30 74.72 80.96 68.98 65.80 85.53 89.48 81.64 79.37 Ours 72.77 84.14 81.97 84.57 73.41 80.84 69.82 67.01 84.67 89.40 82.23 80.25 ST 1/16
65.83 79.09 77.22 78.64 71.58 76.22 59.06 56.45 83.43 86.50 74.26 72.12 ST++ 66.26 79.42 77.49 79.28 71.94 76.56 58.77 57.79 83.68 86.73 74.03 73.25 CCT 64.70 78.29 75.85 79.00 69.65 75.04 57.88 56.22 82.11 85.74 73.32 71.97 ELN 70.70 82.66 80.86 82.32 73.75 80.04 65.11 63.88 84.89 88.91 78.87 77.96 U2PL 70.80 82.77 80.66 82.92 72.95 79.49 66.42 64.35 84.36 88.57 79.82 78.31 UniMatch 71.40 83.17 81.21 82.90 72.46 80.51 68.38 64.25 84.03 89.20 81.22 78.23 Ours 71.80 83.44 81.47 83.97 74.34 80.49 67.28 65.08 85.28 89.19 80.44 78.84
Table 5. Ablation results of different PCNet modules （Resnet50）
View tableView in ArticleTable 5. Ablation results of different PCNet modules （Resnet50）
Method mIoU mDice mAcc mPre ClassIoU ClassDice CCR PLR C-SDA BACK STR NORM TUM BACK STR NORM TUM √ √ √ 74.09 85.04 82.86 85.32 74.50 81.24 70.97 69.66 85.38 89.65 83.02 82.12 √ √ 72.07 83.66 81.45 83.66 73.73 79.85 67.63 67.07 84.88 88.80 80.69 80.29 √ √ 72.68 84.11 81.78 84.53 73.53 79.92 69.66 67.61 84.75 88.90 82.12 80.68 √ √ 71.30 83.12 81.11 83.43 72.93 80.43 67.22 64.62 84.34 89.22 80.40 78.51 √ 70.33 82.46 80.03 83.26 71.39 78.93 66.37 64.63 83.31 88.22 79.79 78.52 √ 68.96 81.44 79.33 81.28 72.10 78.31 63.96 61.47 83.79 87.80 78.02 76.14 √ 69.74 82.00 79.89 82.39 73.45 78.58 63.66 63.26 84.69 88.00 77.79 77.50 67.28 80.24 77.88 80.71 71.64 76.18 60.87 60.45 83.48 86.48 75.67 75.35
Table 6. Influence of different confidence thresholds on the training teacher model
View tableView in ArticleTable 6. Influence of different confidence thresholds on the training teacher model
Confidence threshold mIoU mDice mAcc mPre 0.3 54.59 67.64 70.05 69.23 0.4 69.96 82.20 79.97 81.63 0.5 70.33 82.46 80.03 83.26 0.6 69.15 81.59 79.36 82.30 0.7 68.42 81.12 78.27 82.79
Table 7. Comparative experimental data on BCSS-WSSS
View tableView in ArticleTable 7. Comparative experimental data on BCSS-WSSS
Method Labeled data mIoU mDice mAcc mPre ClassIoU ClassDice OTR TUM NEC STR LYM OTR TUM NEC STR LYM ST 1/8
46.72 61.32 70.84 57.17 25.93 71.78 23.92 61.56 50.38 41.19 83.57 38.61 76.21 67.00 ST++ 46.98 62.12 69.16 59.43 30.93 69.11 25.72 61.44 47.67 47.25 81.73 40.92 76.12 64.57 CCT 58.05 72.48 75.07 72.73 34.95 74.87 55.60 66.83 58.00 51.80 85.63 71.46 80.12 73.42 ELN 59.30 73.37 76.52 72.07 34.73 77.49 60.09 67.93 56.27 51.55 87.32 75.07 80.91 72.02 U2PL 60.58 74.47 77.80 74.90 37.16 78.64 58.11 69.45 59.53 54.19 88.04 73.51 81.97 74.63 UniMatch 61.98 75.71 78.56 75.61 41.12 79.99 59.23 69.87 59.68 58.28 88.88 74.39 82.26 74.75 Ours 63.09 76.68 78.51 77.20 43.26 79.72 62.71 69.71 60.06 60.40 88.72 77.08 82.15 75.05 ST 1/16
43.86 58.71 66.51 56.12 31.90 67.31 16.99 57.53 45.57 48.37 80.46 29.04 73.04 62.61 ST++ 46.97 61.75 69.86 58.90 28.61 70.70 23.64 61.48 50.42 44.49 82.84 38.24 76.15 67.04 CCT 51.51 66.05 72.63 62.27 22.40 74.53 47.08 64.26 49.26 36.60 85.41 64.02 78.24 66.01 ELN 55.95 70.58 74.48 68.78 32.17 75.18 52.79 66.10 53.51 48.68 85.83 69.10 79.59 69.71 U2PL 58.92 73.26 75.35 75.42 37.63 76.91 54.95 66.98 58.15 54.68 86.95 70.93 80.23 73.54 UniMatch 60.10 74.28 77.57 71.38 41.38 78.70 57.16 68.73 54.54 58.54 88.08 72.74 81.47 70.58 Ours 60.20 74.33 77.20 72.57 39.75 78.08 58.53 68.37 56.27 56.89 87.69 73.84 81.21 72.02
Table 7. Comparative experimental data on BCSS-WSSS
View tableView in ArticleTable 7. Comparative experimental data on BCSS-WSSS
Method Labeled data mIoU mDice mAcc mPre ClassIoU ClassDice OTR TUM NEC STR LYM OTR TUM NEC STR LYM ST 1/4
46.23 60.88 69.59 58.55 28.80 70.79 20.75 59.88 50.93 44.72 82.90 34.37 74.90 67.49 ST++ 46.41 61.55 68.17 58.82 31.18 67.31 23.45 60.06 50.03 47.54 80.46 37.99 75.05 66.69 CCT 58.45 72.88 76.47 71.39 41.45 77.15 45.85 67.86 59.93 58.61 87.10 62.88 80.85 74.94 ELN 62.62 76.30 78.43 76.51 42.47 78.99 62.71 69.98 58.97 59.62 88.26 77.08 82.34 74.19 U2PL 62.80 76.34 78.10 77.12 41.06 80.82 63.77 68.00 60.33 58.21 89.39 77.87 80.96 75.26 UniMatch 63.39 76.84 79.30 77.52 43.02 81.05 63.10 70.56 59.20 60.16 89.54 77.38 82.74 74.37 Ours 64.26 77.55 79.25 78.21 44.32 80.93 65.76 70.79 59.50 61.42 89.46 79.35 82.90 74.61
Table 8. Comparative experimental data on LUAD-HistoSeg
View tableView in ArticleTable 8. Comparative experimental data on LUAD-HistoSeg
Method Labeled data mIoU mDice mAcc mPre ClassIoU ClassDice BACK TE TAS NEC LYM BACK TE TAS NEC LYM ST 1/4
49.69 65.90 64.36 64.99 58.49 57.61 53.73 35.96 42.64 73.81 73.10 69.90 52.90 59.79 ST++ 51.18 67.20 66.32 65.42 58..41 62.13 52.27 36.11 46.96 73.74 76.64 68.66 53.06 63.91 CCT 58.90 73.87 73.72 71.16 59.96 69.46 61.93 47.88 55.25 74.97 81.98 76.49 64.76 71.18 ELN 69.38 81.85 80.27 79.90 70.24 75.52 70.06 68.80 62.28 82.52 86.06 82.40 81.52 76.76 U2PL 69.45 81.95 79.50 81.07 67.76 73.44 69.44 69.98 66.60 80.78 84.69 81.96 82.34 79.95 UniMatch 69.37 81.88 79.85 81.47 67.80 74.12 70.07 69.09 65.76 80.81 85.14 82.40 81.72 79.35 Ours 69.45 81.93 79.79 81.13 64.74 73.88 71.16 70.41 67.04 78.60 84.97 83.15 82.64 80.27 ST 1/8
46.09 61.60 63.28 59.25 59.66 56.65 52.71 19.95 41.48 74.74 72.33 69.03 33.27 58.64 ST++ 47.92 64.26 61.42 64.32 60.04 54.14 50.89 37.15 37.36 75.03 70.25 67.45 54.17 54.40 CCT 48.12 62.80 66.49 63.11 54.12 65.80 51.86 53.82 15.01 70.23 79.37 68.30 69.98 26.11 ELN 66.19 79.53 77.59 79.41 69.79 73.01 66.11 65.25 56.80 82.21 84.40 79.60 78.97 72.45 U2PL 68.41 81.19 79.06 81.21 67.85 73.87 68.64 69.23 62.45 80.85 84.97 81.41 81.82 76.88 UniMatch 68.53 81.30 79.06 82.49 66.96 73.55 68.69 68.22 65.23 80.21 84.76 81.44 81.11 78.96 Ours 68.73 81.41 79.60 81.05 64.36 74.24 70.59 69.16 65.29 78.32 85.22 82.76 81.77 79.00 ST 1/16
43.77 58.35 62.29 56.73 57.88 56.63 51.09 9.91 43.35 73.32 72.31 67.63 18.03 60.48 ST++ 44.65 59.75 63.26 57.45 57.98 58.79 51.61 15.22 39.64 73.40 74.04 68.09 26.42 56.78 CCT 44.62 60.51 62.09 60.46 56.63 57.10 51.87 30.34 27.15 72.31 72.69 68.30 46.55 42.71 ELN 61.31 75.52 75.90 72.05 69.67 71.19 64.12 43.67 57.92 82.13 83.17 78.14 60.79 73.36 U2PL 65.92 79.43 75.76 81.08 66.99 70.14 62.01 66.53 63.95 80.23 82.45 76.55 79.90 78.01 UniMatch 68.20 81.05 78.28 77.97 66.29 71.52 68.56 71.35 63.30 79.73 83.39 81.35 83.28 77.53 Ours 68.92 81.55 79.27 80.40 62.97 72.87 70.95 71.12 66.71 77.28 84.30 83.00 83.12 80.03
Tools
Get Citation
Copy Citation Text
Hanhan XU, Yinhui ZHANG, Zifen HE, Jiacen LIU, Zhenhui LI, Lin WU, Benjie SHI. Pseudo-label confidence regulates semi-supervised semantic segmentation of pathological images of colorectal cancer[J]. Optics and Precision Engineering, 2025, 33(4): 591
Paper Information
Category:
Received: Apr. 1, 2024
Accepted: --
Published Online: May. 20, 2025
The Author Email: Yinhui ZHANG (zhangyinhui@kust.edu.cn), Zifen HE (zyhhzf1998@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20