Optics and Precision Engineering, Volume. 32, Issue 4, 565(2024)
Brain tumor image segmentation based on Semantic Flow Guided Sampling and Attention Mechanism
Figures & Tables(15)
Fig. 2. Structural diagram of Multi-Fiber unit and Dilated Multi-Fiber unit
Fig. 5. Brain tumor image and real segmentation tags with different modalities
Table 1. Datasets composition
View tableView in ArticleTable 1. Datasets composition
Datasets Train Valid BraTS2018 285 66 BraTS2019 335 125
Table 2. Comparison of effects of different models using FA
View tableView in ArticleTable 2. Comparison of effects of different models using FA
Methods Dice/% Sensitivity/% Specificity/% Hausdorff95/mm ET WT TC ET WT TC ET WT TC ET WT TC DMFNet 76.68 88.47 80.19 77.79 88.31 79.14 99.81 99.42 99.73 3.341 5.407 6.138 FA+DMFNet 76.78 88.54 80.98 77.53 88.70 79.49 99.82 99.52 99.75 3.134 5.385 6.148 3D UNet 72.68 87.18 70.88 70.15 86.72 65.81 99.84 99.49 99.89 5.341 9.172 9.461 FA+UNet 74.65 88.21 72.62 74.28 87.78 69.04 99.88 99.47 99.83 5.422 5.430 9.652 HDCNet 76.98 89.15 81.51 77.63 92.26 80.61 99.83 99.26 99.75 4.159 6.433 8.009 FA+HDCNet 77.17 89.56 81.55 78.94 92.22 80.78 99.81 99.28 99.71 3.079 5.769 7.133
Table 3. Ablation experiment of DAFANet model
View tableView in ArticleTable 3. Ablation experiment of DAFANet model
Methods Dice/% Hausdorff95/mm ET WT TC ET WT TC DMFNet 76.68 88.47 80.19 3.341 5.407 6.138 DMFNet+EMA 77.14 89.69 81.25 3.960 5.181 7.091 FA+DMFNet 76.78 88.54 80.98 3.134 5.385 6.148 FA+DMFNet+EMA（up） 76.81 89.78 81.61 3.742 4.863 6.954 FA+DMFNet+EMA（down） 77.09 89.59 81.54 2.677 4.959 6.125 FA+DMFNet+2EMA 78.11 90.10 82.21 3.072 4.360 6.229
Table 4. Ablation experiment of EMA position
View tableView in ArticleTable 4. Ablation experiment of EMA position
Methods Dice/% Hausdorff95/mm Params/M FLOPs/G ET WT TC ET WT TC Layer 1 76.72 88.35 78.67 4.316 5.972 6.871 4.11 30.51 Layer 2 74.85 87.66 76.25 3.241 8.451 7.628 4.13 31.04 Layer 3 78.11 90.10 82.21 3.072 4.360 6.229 4.23 30.50
Table 5. Comparative experiment of superparameter
k View tableView in ArticleTable 5. Comparative experiment of superparameter
k k Dice/% Hausdorff95/mm ET WT TC ET WT TC 64 78.11 90.10 82.21 3.072 4.306 6.229 32 76.01 89.38 80.21 3.246 5.027 6.071 16 78.16 90.03 80.73 3.022 4.682 6.326
Table 6. Comparison of segmentation results with classical models
View tableView in ArticleTable 6. Comparison of segmentation results with classical models
Methods Dice/% Hausdorff95/mm Params/M FLOPs/G ET WT TC ET WT TC TransBTS［ 13 ］78.93 90.00 81.94 3.736 5.644 6.049 32.99 333.00 3D UNet［ 10 ］72.68 87.18 70.88 5.341 9.172 9.461 13.12 176.61 3D ESPNet［ 25 ］77.35 89.30 81.83 4.190 6.740 7.980 3.63 76.51 Attention UNet［ 12 ］75.96 88.81 77.20 5.020 7.756 8.258 34.90 51.30 DMFNet［ 20 ］76.68 88.47 80.19 3.341 5.407 6.138 3.88 27.04 HDCNet［ 21 ］76.98 89.15 81.51 4.159 6.433 8.009 0.29 24.00 DAFANet 78.11 90.10 82.21 3.072 4.360 6.229 4.23 30.50
Table 7. Comparison of Dice coefficient and Hausdorff95 distance with other models under different datasets
View tableView in ArticleTable 7. Comparison of Dice coefficient and Hausdorff95 distance with other models under different datasets
Dataset Methods Dice/% Hausdorff95/mm ET WT TC ET WT TC BraTS2019 Nuechterlein［ 25 ］77.35 89.30 81.83 4.19 6.74 7.98 Chen［ 20 ］76.68 88.47 80.19 3.34 5.41 6.14 Zhang［ 26 ］70.90 87.00 70.90 N/A N/A N/A Sheng［ 27 ］72.40 87.50 78.80 5.79 9.35 11.47 Akbar［ 28 ］74.20 88.48 80.98 6.67 10.25 10.83 Liu［ 29 ］77.91 89.94 83.89 4.03 5.45 6.56 Chang［ 30 ］78.20 89.00 81.20 3.82 8.53 7.43 Ours 78.11 90.10 82.21 3.07 4.36 6.23 BraTS2018 Nuechterlein［ 25 ］73.70 88.30 81.40 5.30 5.46 7.85 Chen［ 20 ］80.11 90.61 84.54 3.06 4.66 6.44 Zhang［ 26 ］77.20 87.20 80.80 5.69 12.60 9.62 Sheng［ 27 ］77.10 87.60 81.10 3.22 8.42 10.56 Akbar［ 28 ］77.71 89.59 79.77 3.90 9.13 8.67 Liu［ 29 ］80.41 89.85 85.44 2.24 4.05 5.76 Chang［ 30 ］79.50 90.00 83.90 2.92 6.51 5.71 Ours 80.44 90.07 84.57 2.75 4.70 5.49
Tools
Get Citation
Copy Citation Text
Jianli SONG, Xiaoqi LÜ, Yu GU. Brain tumor image segmentation based on Semantic Flow Guided Sampling and Attention Mechanism[J]. Optics and Precision Engineering, 2024, 32(4): 565
Paper Information
Category:
Received: Apr. 5, 2023
Accepted: --
Published Online: Apr. 2, 2024
The Author Email: LÜ Xiaoqi (lxiaoqi@imut.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20