Deep Convolutional Neural Networks Using U-Net for Automatic Brain Tumor Segmentation in Multimodal MRI Volumes
Precise 3D computerized segmentation of brain tumors remains, until nowadays, a challenging process due to the variety of the possible shapes, locations and image intensities of various tumors types. This paper presents a fully automated and efficient brain tumor segmentation method based on 2D Deep Convolutional Neural Networks (DNNs) which automatically extracts the whole tumor and intra-tumor regions, including enhancing tumor, edema and necrosis, from pre-operative multimodal 3D-MRI. The network architecture was inspired by U-net and has been modified to increase brain tumor segmentation performance. Among applied modifications, Weighted Cross Entropy (WCE) and Generalized Dice Loss (GDL) were employed as a loss function to address the class imbalance problem in the brain tumor data. The proposed segmentation system has been tested and evaluated on both, BraTS’2018 training and validation datasets, which include a total of 351 multimodal MRI volumes of different patients with HGG and LGG tumors representing different shapes, giving promising and objective results close to manual segmentation performances obtained by experienced neuro-radiologists. On the challenge validation dataset, our system achieved a mean enhancing tumor, whole tumor, and tumor core dice score of 0.783, 0.868 and 0.805 respectively. Other quantitative and qualitative evaluations are presented and discussed along the paper.
KeywordsBrain tumor segmentation 3D-MRI Machine learning Deep learning Convolutional Neural Networks U-net BraTS’2018 challenge
This work was in part financially supported by an Algerian research project (CNEPRU) funded by the Ministry of Higher Education and Scientific Research (Project title and number: “PERFORM”, B*04120140014). We would like to thank Bakas, S., Ph.D. and Postdoctoral researcher at SBIA of Perelman School of Medicine University of Pennsylvania – USA for providing the entire BraTS’2018 datasets employed in this study. We also gratefully acknowledge the support of CERIST, the Algerian Research Center for Scientific and Technical Information, for allowing us the use of “IBNBADIS” Cluster, without which we could not have performed these tests.
- 1.Dass, R., Priyanka, Devi, S.: Image segmentation techniques. Int. J. Electron. Commun. Technol. (IJECT) 3, 66–70 (2012)Google Scholar
- 8.Yanyun, L., Zhijian, S.: Automated brain tumor segmentation in magnetic resonance imaging based on sliding-window technique and symmetry analysis. Chin. Med. J. 127, 462–468 (2014)Google Scholar
- 10.Ronneberger, O., Fischer, P., Brox, T.: U-Net: convolutional networks for biomedical image segmentation. In: Navab, N., Hornegger, J., Wells, W.M., Frangi, A.F. (eds.) MICCAI 2015. LNCS, vol. 9351, pp. 234–241. Springer, Cham (2015). https://doi.org/10.1007/978-3-319-24574-4_28CrossRefGoogle Scholar
- 12.Milletari, F., Navab, N., Ahmadi, S.A.: V-Net: fully convolutional neural networks for volumetric medical image segmentation. In: 2016 Fourth International Conference on 3D Vision (3DV), pp. 565–571 (2016)Google Scholar
- 14.Bakas, S., et al.: Segmentation labels and radiomic features for the pre-operative scans of the TCGA-GBM collection. The Cancer Imaging Archive (2017)Google Scholar
- 15.Bakas, S., et al.: Segmentation labels and radiomic features for the pre-operative scans of the TCGA-LGG collection. The Cancer Imaging Archive (2017)Google Scholar
- 18.He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778 (2016)Google Scholar
- 20.He, K., Zhang, X., Ren, S., Sun, J.: Delving deep into rectifiers: surpassing human-level performance on ImageNet classification. In: Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV), pp. 1026–1034. IEEE Computer Society (2015)Google Scholar
- 21.Sudre, C.H., Li, W., Vercauteren, T., Ourselin, S., Jorge Cardoso, M.: Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations. In: Cardoso, M.J., et al. (eds.) DLMIA/ML-CDS -2017. LNCS, vol. 10553, pp. 240–248. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-67558-9_28CrossRefGoogle Scholar
- 22.Kermi, A., Mahmoudi, I., Khadir, M.T.: Brain tumor segmentation in multimodal 3D-MRI of BraTS’2018 datasets using Deep Convolutional Neural Networks. In: Pre-conference Proceedings of the 7th International MICCAI BraTS’2018 Challenge, Granada, Spain, pp. 252–263 (2018)Google Scholar
- 23.Bakas, S., et al.: Identifying the Best Machine Learning Algorithms for Brain Tumor Segmentation, Progression Assessment, and Overall Survival Prediction in the BRATS Challenge. arXiv preprint arXiv:1811.02629 (2018)
- 24.Castillo, L.S., Daza, L.A., Rivera, L.C., Arbelàez, P.: Volumetric multimodality neural network for brain tumor segmentation. In: Proceedings of the 6th MICCAI BraTS Challenge, Quebec, Canada, pp. 34–41 (2017)Google Scholar