540 K.Fang and W.-J.Li we can conclude that the proposed DMNet can utilize unlabeled data to improve the segmentation performance,especially when the amount of labeled data is lim- ited.When only 10%of labeled data is available,DMNet can improve the mloU from 67.0%to 78.7%.When all labeled data is available,in which case the amount of unlabeled data is almost the same as that of labeled data,DMNet can also improve the mIoU from 84.2%to 87.0%. Table 3.Comparison between DMNet and its variants Method Amount of labeled data 10% 30% 50% 100% Supervised DMNet without adv 59.3% 75.8% 79.4% 84.5% Supervised DMNet with adv 67.0% 76.9% 79.8% 84.2% Separate DMNet 76.1% 84.2% 84.4% 85.0% DMNet_wo_adv_wo_sharpen 75.8% 82.0% 82.5% 86.8% DMNet_wo_sharpen 76.9% 82.3% 83.9% 86.9% DMNet 78.7% 85.0% 85.4% 87.0% 5 Conclusion In this paper,we propose a novel semi-supervised method,called DMNet,for semantic segmentation in medical image analysis.DMNet can be trained with a limited amount of labeled data and a large amount of unlabeled data.Hence, DMNet can be used to solve the problem that it is typically difficult to collect a large amount of labeled data in medical image analysis.Experiments on a kidney tumor dataset and a brain tumor dataset show that DMNet can outperform other baselines,including both supervised ones and semi-supervised ones,to achieve the best performance. References 1.Badrinarayanan,V.,Kendall,A.,Cipolla,R.:SegNet:a deep convolutional encoder-decoder architecture for image segmentation.IEEE Trans.Pattern Anal. Mach.Intell..39(12),2481-2495(2017) 2.Bai,W.,et al.:Semi-supervised learning for network-based cardiac MR image segmentation.In:Descoteaux,M.,Maier-Hein,L.,Franz,A.,Jannin,P.,Collins, D.L.,Duchesne,S.(eds.)MICCAI 2017.LNCS,vol.10434,pp.253-260.Springer, Cham(2017).https:/doi.org/10.1007/978-3-319-66185-829 3.Berthelot,D.,Carlini,N.,Goodfellow,I.J.,Papernot,N.,Oliver,A.,Raffel,C.: MixMatch:a holistic approach to semi-supervised learning.CoRR (2019) 4.Blum,A.,Mitchell,T.M.:Combining labeled and unlabeled data with co-training. In:Proceedings of Annual Conference on Computational Learning Theory (COLT) (1998)540 K. Fang and W.-J. Li we can conclude that the proposed DMNet can utilize unlabeled data to improve the segmentation performance, especially when the amount of labeled data is lim￾ited. When only 10% of labeled data is available, DMNet can improve the mIoU from 67.0% to 78.7%.When all labeled data is available, in which case the amount of unlabeled data is almost the same as that of labeled data, DMNet can also improve the mIoU from 84.2% to 87.0%. Table 3. Comparison between DMNet and its variants Method Amount of labeled data 10% 30% 50% 100% Supervised DMNet without adv 59.3% 75.8% 79.4% 84.5% Supervised DMNet with adv 67.0% 76.9% 79.8% 84.2% Separate DMNet 76.1% 84.2% 84.4% 85.0% DMNet wo adv wo sharpen 75.8% 82.0% 82.5% 86.8% DMNet wo sharpen 76.9% 82.3% 83.9% 86.9% DMNet 78.7% 85.0% 85.4% 87.0% 5 Conclusion In this paper, we propose a novel semi-supervised method, called DMNet, for semantic segmentation in medical image analysis. DMNet can be trained with a limited amount of labeled data and a large amount of unlabeled data. Hence, DMNet can be used to solve the problem that it is typically difficult to collect a large amount of labeled data in medical image analysis. Experiments on a kidney tumor dataset and a brain tumor dataset show that DMNet can outperform other baselines, including both supervised ones and semi-supervised ones, to achieve the best performance. References 1. Badrinarayanan, V., Kendall, A., Cipolla, R.: SegNet: a deep convolutional encoder-decoder architecture for image segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 39(12), 2481–2495 (2017) 2. Bai, W., et al.: Semi-supervised learning for network-based cardiac MR image segmentation. In: Descoteaux, M., Maier-Hein, L., Franz, A., Jannin, P., Collins, D.L., Duchesne, S. (eds.) MICCAI 2017. LNCS, vol. 10434, pp. 253–260. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-66185-8 29 3. Berthelot, D., Carlini, N., Goodfellow, I.J., Papernot, N., Oliver, A., Raffel, C.: MixMatch: a holistic approach to semi-supervised learning. CoRR (2019) 4. Blum, A., Mitchell, T.M.: Combining labeled and unlabeled data with co-training. In: Proceedings of Annual Conference on Computational Learning Theory (COLT) (1998)