Optics and Precision Engineering, Volume. 30, Issue 16, 2021(2022)
Skin lesion segmentation based on high-resolution composite network
Figures & Tables(18)
Fig. 8. Different network segmentation results on ISBI2016 and ISBI2017 datasets
Fig. 9. Different network segmentation results on PH2 and ISIC2018 datasets
Table 1. Results of different networks on ISBI2016 and ISBI2017 datasets
View tableView in ArticleTable 1. Results of different networks on ISBI2016 and ISBI2017 datasets
Method ISBI 2016 ISBI 2017 Accuracy Sensitivity Specificity Dice Jaccard Accuracy Sensitivity Specificity Dice Jaccard Para/M Time/s U-Net 94.92% 93.06% 95.66% 91.18% 83.80% 92.26% 75.29% 97.50% 82.11% 69.65% 28.21 383 HRNet 94.98% 93.36% 95.62% 91.30% 83.99% 92.70% 76.35% 97.74% 83.18% 71.15% 30.79 1064 DEDNet 95.49% 93.17% 96.40% 92.10% 85.35% 93.00% 77.69% 97.73% 83.97% 72.37% 44.54 623 CANet 95.61% 94.32 %96.12% 92.37% 85.83% 93.17% 78.59% 98.47% 84.44% 73.07% 45.55 435 CHNet 95.93% 93.46% 96.60% 92.78% 86.13% 93.56% 81.66% 98.17% 87.23% 76.26% 48.24 668 HCNet 96.14% 93.62% 97.24 %93.16 %87.01 %93.72 %83.54 %98.52 %88.56 %77.19 %33.96 497
Table 2. Results of different networks on PH2 and ISIC2018 datasets
View tableView in ArticleTable 2. Results of different networks on PH2 and ISIC2018 datasets
Method PH2 ISIC2018 Accuracy Sensitivity Specificity Dice Jaccard Accuracy Sensitivity Specificity Dice Jaccard Para/M Time/s U-Net 91.68% 92.33% 92.85% 89.05% 80.26% 93.56% 89.32% 95.16% 88.38% 80.18% 28.21 343 HRNet 91.73% 92.69% 90.70% 86.62% 78.34% 93.70% 89.90% 93.76% 87.10% 81.14% 30.79 978 DEDNet 92.93% 91.21% 93.75 %89.27% 80.61% 94.74% 88.88% 95.57% 88.61% 81.55% 44.54 564 CANet 92.86% 93.32% 92.64% 89.39% 80.81% 94.87% 91.20% 94.77% 88.95% 82.09% 45.55 403 CHNet 92.94% 94.28% 92.31% 89.60% 81.16% 95.15% 90.92% 95.06% 90.59% 84.52% 48.24 594 HCNet 94.31 %96.06 %93.48% 91.59 %84.48 %95.73 %91.67 %97.57 %92.00 %85.19 %33.96 455
Table 3. Objectivity comparison of different networks on ISBI2016 datasets
View tableView in ArticleTable 3. Objectivity comparison of different networks on ISBI2016 datasets
Method Accuracy Sensitivity Specificity Dice Jaccard EXB 95.30% 91.00% 96.50% 91.00% 84.30% CUMED 94.90% 91.10% 95.70% 89.70% 82.90% Mahudr 95.20% 88.00% 96.90% 89.50% 82.20% SFU-mial 94.40% 91.50% 95.50% 88.50% 81.10% TMUteam 94.60% 83.20% 98.70 %88.80% 81.00% Ref.［ 29 ］95.02% 90.15% 97.00% 90.11% 83.30% Ref.［ 30 ］95.4% 92.7% 96.1% 90.8% 84.5% Ref.［ 31 ］95.7% 92.8% 96.3% 91.9% 85.5% Ref.［ 32 ］95.8% 91.4% 97.1% 91.8% 85.8% Ref.［ 33 ］95.87% 92.67% 96.42% 91.47% 85.34% Ref.［ 28 ］96.09% 92.50% 97.43% 92.51% 86.23% HCNet 96.14 %93.62 %97.24% 93.16 %87.01 %
Table 4. Objectivity comparison of different networks on ISBI2017 datasets
View tableView in ArticleTable 4. Objectivity comparison of different networks on ISBI2017 datasets
Method Accuracy Sensitivity Specificity Dice Jaccard Ref.［ 11 ］93.29% 87.34% 93.14% 87.86% 80.45 %Ref.［ 30 ］92.6% 82.5% 96.5% 83.0% 74.2% Ref.［ 32 ］93.8% 87.0% 96.4% 86.2% 78.3% Ref.［ 28 ］93.63% 81.06% 97.43% 84.91% 74.27% Ref.［ 34 ］93.2% 93.0 %90.5% 85.1% 76.7% Ref.［ 35 ］93.39% 90.82% 92.68% 84.26% 74.81% Ref.［ 36 ］93.5% 83.5% 97.6% 85.9% 77.1% Ref.［ 12 ］93.6% 81.6% 98.3% 87.8% 78.2% HCNet 93.72 %83.54% 98.52 %88.56 %77.19%
Table 5. Objectivity comparison of different networks on ISIC2018 datasets
View tableView in ArticleTable 5. Objectivity comparison of different networks on ISIC2018 datasets
Method Accuracy Sensitivity Specificity Dice Jaccard Ref.［ 37 ］93.4% 96.7 %90.4% 87.7% 79.4% Ref.［ 38 ］94.58% 91.57% 95.85% 90.96% 83.42% Ref.［ 39 ］94.7% 94.2% 94.1% 90.8% 84.4% Ref.［ 40 ］95.68% 93.06% 94.69% 89.55% 83.09% HCNet 95.73 %91.67% 97.57 %92.00 %85.19 %
Table 6. Ablation Experiment of each module
View tableView in ArticleTable 6. Ablation Experiment of each module
Method ISBI2016 ISBI2017 Accuracy Sensitivity Specificity Dice Jaccard Accuracy Sensitivity Specificity Dice Jaccard U-Net 94.92% 93.06% 95.66% 91.18% 83.80% 92.26% 75.29% 97.50% 82.11% 69.65% HCNet_1 95.02% 93.46% 95.43% 91.78% 84.19% 92.96% 80.57% 97.42% 83.09% 72.63% HCNet_2 95.88% 93.30% 97.57 %92.42% 86.42% 93.55% 81.69% 98.73 %86.97% 75.37% HCNet_3 95.41% 95.50 %95.35% 92.10% 85.37% 93.12% 84.95 %97.14% 84.63% 74.02% HCNet_4 95.63% 93.98% 96.14% 92.22% 85.57% 93.41% 81.72% 97.62% 87.00% 75.26% HCNet_5 96.02% 93.42% 96.81% 93.03% 86.99% 93.68% 82.56% 98.39% 88.12% 77.07% HCNet 96.14 %93.62% 97.24% 93.16 %87.01 %93.72 %83.54% 98.52% 88.56 %77.19 %
Tools
Get Citation
Copy Citation Text
Liming LIANG, Longsong ZHOU, Jun FENG, Xiaoqi SHENG, Jian WU. Skin lesion segmentation based on high-resolution composite network[J]. Optics and Precision Engineering, 2022, 30(16): 2021
Paper Information
Category: Information Sciences
Received: Mar. 13, 2022
Accepted: --
Published Online: Sep. 22, 2022
The Author Email: WU Jian (wujian@jxust.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20