基于最优传输特征选择的医学图像分割迁移学习

doi:10.13229/j.cnki.jdxbgxb20210652

Abstract

Abstract:

In the unsupervised domain adaptive transfer learning process， domain-independent features lead to the degradation of model segmentation performance， but there is no effective feature selection method for transfer learning segmentation model at present. To solve this problem， a general feature selection module for transfer learning was proposed based on optimal transport， which can be applied to various unsupervised domain adaptive image segmentation models. In this module， the optimal sample subsets of two domains are selected by weighted optimal transport of segmentation accuracy， and then the features of sample subsets are subjected to entropy regularized optimal transport， so as to obtain a descending list of similarity between two domains to remove domain-independent features. The universal feature selection module is applied to three unsupervised domain adaptive models to solve the problem of Covid-19 image segmentation， which improves the model performance to a certain extent.

Key words: artificial intelligence, transfer learning, unsupervised domain adaptation, optimal transport, feature selection, image segmentation

CLC Number:

TP18

Sheng-sheng WANG,Lin-yan JIANG,Yong-bo YANG. Transfer learning of medical image segmentation based on optimal transport feature selection[J].Journal of Jilin University(Engineering and Technology Edition), 2022, 52(7): 1626-1638.

Figures/Tables 12

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Table 1

Table 2

Fig.6

Fig.7

Table 3

Table 4

Fig. 8

References 27

1	Huang J, Smola A J, Gretton A, et al.Correcting sample selection bias by unlabeled data[J]. Advances in Neural Information Processing Systems, 2007, 19: 601-608.
2	Ganin Y, Ustinova E, Ajakan H, et al. Domain-adversarial training of neural networks[J]. Advances in Computer Vision and Pattern Recognition, 2017, 17(1): 2096-2130.
3	许新征, 丁世飞, 史忠植,等. 图像分割的新理论和新方法[J]. 电子学报, 2010, 38(): 76-82.
	Xu Xin-zheng, Ding Shi-fei, Shi Zhong-zhi, et al. New theories and methods of image segmentation[J]. Acta Electronica Sinica, 2010, 38(Sup.): 76-82.
4	He B S, Zhu F, Shi Y G. Medical image segmentation[J]. Advanced Materials Research, 2013, 760-762(4): 1590-1593.
5	Courty N, Flamary R, Tuia D, et al. Optimal transport for domain adaptation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(9): 1853-1865.
6	Courty N, Flamary R, Habrard A, et al. Joint distribution optimal transportation for domain adaptation[J]. Advances in Neural Information Processing Systems, 2017, 30: 3731-3740.
7	Tzeng E, Hoffman J, Zhang N, et al. Deep domain confusion: maximizing for domain invariance[J/OL]. [2021-06-10].
8	Ruan C, Wang W, Hu H, et al. Category-level adversaries for semantic domain adaptation[J]. IEEE Access, 2019, 7: 83198-83208.
9	Hoffman J, Tzeng E, Park T, et al. CyCADA: cycle-consistent adversarial domain adaptation[C]∥The 35th International Conference on Machine Learning, Stockholm, Sweden, 2018: 3162-3174.
10	Dou Q, Ouyang C, Chen C, et al. Unsupervised cross-modality domain adaptation of convnets for biomedical image segmentations with adversarial loss[C]∥Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, New York, USA, 2018: 691-697.
11	Redko I, Habrard A, Sebban M. Theoretical analysis of domain adaptation with optimal transport[C]∥Joint European Conference on Machine Learning and Knowledge Discovery in Databases, London, England, 2017: 737-753.
12	Balaji Y, Chellappa R, Feizi S. Normalized wasserstein for mixture distributions with applications in adversarial learning and domain adaptation[C]∥Proceedings of the IEEE International Conference on Computer Vision, Seoul, South Korea, 2019: 6509-6517.
13	Damodaran B B, Kellenberger B, Flamary R, et al. DeepJDOT: Deep joint distribution optimal transport for unsupervised domain adaptation[C]∥Proceedings of the European Conference on Computer Vision (ECCV), New York, USA, 2018: 447-463.
14	Ackaouy A, Courty N, Vallée E, et al. Unsupervised domain adaptation with optimal transport in multi-site segmentation of multiple sclerosis lesions from MRI data[J]. Frontiers in Computational Neuroscience, 2020, 14: 19.
15	Xu R, Liu P, Wang L,et al. Reliable weighted optimal transport for unsupervised domain adaptation[C]∥Proceedings of the Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 4394-4403.
16	Li M, Zhai Y M, Luo Y W, et al. Enhanced transport distance for unsupervised domain adaptation[C]∥Proceedings of the Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 13936-13944.
17	Li J, Zhao J, Lu K. Joint feature selection and structure preservation for domain adaptation[C]∥Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence, New York, USA, 2016: 1697-1703.
18	Sun F, Wu H, Luo Z, et al. Informative feature selection for domain adaptation[J]. IEEE Access, 2019, 7: 142551-142563.
19	Gautheron L, Redko I, Lartizien C. Feature selection for unsupervised domain adaptation using optimal transport[C]∥Joint European Conference on Machine Learning and Knowledge Discovery in Databases, London, England, 2018: 759-776.
20	Villani C. Optimal transport old and new[J]. Springer Science & Business Media, 2007, 338(23/24): 480-488.
21	Cuturi M. Sinkhorn distances: lightspeed computation of optimal transport[C]∥Neural Information Processing Systems, Lake Tahoe, Spain, 2013: 2292-2300.
22	Knight P A. The Sinkhorn-knopp algorithm: convergence and applications[J]. SIAM Journal on Matrix Analysis and Applications, 2008, 30(1): 261-275.
23	Li H, Loehr T, Sekuboyina A, et al. e-UDA: efficient unsupervised domain adaptation for cross-site medical image segmentation[J/OL]. [2021-6-10]. .
24	Perone C S, Ballester P, Barros R C, et al. Unsupervised domain adaptation for medical imaging segmentation with self-ensembling[J]. NeuroImage, 2019, 194: 1-11.
25	Laradji I, Rodriguez P, Manas O, et al. A weakly supervised consistency-based learning method for COVID-19 segmentation in CT images[C]∥Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, New York, USA, 2021: 2453-2462.
26	Zhou T, Canu S, Ruan S. An automatic COVID-19 CT segmentation network using spatial and channel attention mechanism[J]. International Journal of Imaging Systems and Technology, 2020, 21(1): 16-27.
27	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]∥International Conference on Learning Representations, New York, USA, 2015: 1-14.

Related Articles 15

[1]	Hao-yu TIAN,Xin MA,Yi-bin LI. Skeleton-based abnormal gait recognition： a survey [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(4): 725-737.
[2]	Xue-zhi WANG,Qing-liang LI,Wen-hui LI. Spatio⁃temporal model of soil moisture prediction integrated with transfer learning [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(3): 675-683.
[3]	Xue WANG,Zhan-shan LI,Ying-da LYU. Medical image segmentation based on multi⁃scale context⁃aware and semantic adaptor [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(3): 640-647.
[4]	Yong LIU,Lei XU,Chu-han ZHANG. Deep reinforcement learning model for text games [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(3): 666-674.
[5]	Jing-pei LEI,Dan-tong OUYANG,Li-ming ZHANG. Relation domain and range completion method based on knowledge graph embedding [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(1): 154-161.
[6]	Zhi-hua LI,Ye-chao ZHANG,Guo-hua ZHAN. Realtime mosaic and visualization of 3D underwater acoustic seabed topography [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(1): 180-186.
[7]	Song-mei TANG. Key technology of face recognition system based on swarm intelligence in library [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(6): 2216-2224.
[8]	Yan-lei XU,Run HE,Yu-ting ZHAI,Bin ZHAO,Chen-xiao LI. Weed identification method based on deep transfer learning in field natural environment [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(6): 2304-2312.
[9]	Sheng-sheng WANG,Jing-yu CHEN,Yi-nan LU. COVID⁃19 chest CT image segmentation based on federated learning and blockchain [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(6): 2164-2173.
[10]	Yong YANG,Qiang CHEN,Fu-heng QU,Jun-jie LIU,Lei ZHANG. SP⁃k⁃means-+ algorithm based on simulated partition [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(5): 1808-1816.
[11]	Ya-hui ZHAO,Fei-yang YANG,Zhen-guo ZHANG,Rong-yi CUI. Korean text structure discovery based on reinforcement learning and attention mechanism [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1387-1395.
[12]	Jing-bin LI,Yu-kun YANG,Bao-qin WEN,Za KAN,Wen SUN,Shuo YANG. Method of extraction of navigation path of post-autumn residual film recovery based on stubble detection [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1528-1539.
[13]	Yan-hua DONG,Jing-wei LIU,Jing-hua ZHAO,Liang LI,Fang-xi XIE. Real-time torque tracking control based on BPNN online learning prediction model [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1405-1413.
[14]	Fu LIU,Yi-xin LIANG,Tao HOU,Yang SONG,Bing KANG,Yun LIU. Improvement of fuzzy c-harmonic mean algorithm on unbalanced data [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1447-1453.
[15]	Fu-hua SHANG,Mao-jun CAO,Cai-zhi WANG. Local outlier data mining based on artificial intelligence technology [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(2): 692-696.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

Comments

Recommended 0

No Suggested Reading articles found!

特征维度	评价指标	分割准确性加权OT	随机样本选择	OT1	OT2
512	Dice/%	67.3	54.9	64.1	65.6
	Sen/%	71.9	71.1	73.1	71.5
	Pre/%	65.7	46.0	66.2	65.1
2048	Dice/%	70.9	64.2	67.1	68.4
	Sen/%	74.8	72.3	71.2	73.1
	Pre/%	68.6	64.9	65.9	66.5
4096	Dice/%	69.1	69.1	69.1	69.1
	Sen/%	72.7	72.7	72.7	72.7
	Pre/%	65.9	65.9	65.9	65.9

域适应对	评价指标	未进行特征选择	本文方法	升序特征选择	随机特征选择
域适应对	评价指标	4096	↓512	↑512	→512
A→B	Dice/%	71.5	72.2	40.1	60.8
	Sen/%	73.9	75.6	61.6	72.2
	Pre/%	71.6	72.8	32.6	52.1
B→A	Dice/%	63.3	65.8	36.1	53.6
	Sen/%	72.5	71.3	60.3	70.1
	Pre/%	65.3	65.1	29.6	44.9

方法	评价指标	不同特征维度
方法	评价指标	↓512	↓1024	↓2048	3072
Baseline	Dice/%	67.1	69.2	70.5	71.2
	Sen/%	71.6	72.1	74.9	75.8
	Pre/%	65.2	65.3	68.2	69.3
seg‐jdot	Dice/%	74.9	74.8	76.9	78.1
	Sen/%	80.3	79.8	80.6	81.1
	Pre/%	73.1	75.3	75.3	76.8
e‐UDA	Dice/%	74.6	76.7	78.3	78.1
	Sen/%	78.1	80.1	79.9	79.6
	Pre/%	75.3	74.9	76.8	76.1
self‐ensembling	Dice/%	75.4	76.3	77.2	77.9
	Sen/%	79.2	79.4	79.1	80.2
	Pre/%	63.7	64.1	65.3	66.8

方法	评价指标	使用通用特征选择模块前	使用通用特征选择模块后
Baseline	Dice/%	68.9	71.2
	Sen/%	72.8	75.8
	Pre/%	65.3	69.3
seg‐jdot	Dice/%	75.1	78.1
	Sen/%	80.1	81.1
	Pre/%	74.8	76.8
e‐UDA	Dice/%	76.3	78.3
	Sen/%	79.5	79.9
	Pre/%	75.1	76.8
self‐ensembling	Dice/%	76.7	77.9
	Sen/%	79.4	80.2
	Pre/%	64.1	66.8

Transfer learning of medical image segmentation based on optimal transport feature selection

RICH HTML

PDF (PC)