運用影像處理方法於輔助醫學影像的分析與判別病灶已經行之有年，近來經由深度學習方法其處理效能更加提升。醫學影像計算與電腦輔助干預協會(Medical Image Computing and Computer Assisted Intervention, MICCAI)每年舉辦腦部影像病灶區分割比賽，期望透過自動判別的方式判斷及分割磁振成像(MRI)之腦神經膠質瘤所在位置。文獻上已有許多應用深度學習方法於腦腫瘤分割的系統架構，本論文提出三種基於Unet架構之深度學習方法: (1)第一為在每一層中運用Inception Module來擷取特徵，加寬深度學習網路架構，使其納入更多不同褶積萃取的特徵提供選擇; (2)第二方法是基於第一個架構，為了在加深網路的同時可以提升準確率，在每一層中加入了殘差連接(Skip Connection)來維持網路之穩定性; (3)第三方法基於第二個架構，在解碼端(Decoder)加入深監督(Deep Supervision)機制，藉由在上取樣時添加分支網路，來提升準確率。我們所使用的資料庫為Brats2018之Training Data，其包含了210筆高級別膠質瘤以及75筆低級別膠質瘤，而預測結果之評估可分為三個部分，第一個是全腫瘤(Whole Tumor, WT)，第二個是腫瘤核心(Tumor Core, TC)，最後一個則是增強腫瘤(Enhancing Tumor, ET)。實驗結果顯示，我們所提出的第二個方法與架構在Dice係數上的平均預測率為Dice_WT=0.8691、 Dice_TC=0.7845以及 Dice_ET = 0.7002，整體而言分割效能優於其他方法。

Image segmentation methods have been utilized to analyze medical image contents and detect lesions. In recent years, image segmentation methods using deep learning help much in improving performances. A Medical Image Computing and Computer Assisted Intervention (MICCAI) society holds the brain tumor segmentation competition every year. The main purpose of this competition is to automatically detect and segment gliomas from images through Magnetic Resonance Imaging (MRI). Many deep learning methods have been proposed for brain tumor segmentation. The aim of this research is to identify lesions in brain MRI images using deep learning methods. We proposed three deep learning based brain tumor segmentation methods for MRI images, all of which are developed based on a Unet framework: (1) For the first method, we utilized inception modules to extract features in each layer. It helps to enable multi-level feature extraction for learning; (2) For the second method, we added skip connection in each layer to mitigate the gradient vanishing/exploding problem; (3) For the third one, we used a deep supervision method for the decoder, in which it adds one branch at the up-sampling stage to improve the segmentation accuracy. The dataset, Brats2018 Training Data, comprises 210 high-grade gliomas and 75 low-grade gliomas. Performance evaluations comprise three aspects: whole tumor (WT), tumor core (TC), and enhancing tumor (ET). Experiments showed that the second method outperforms the others in brain tumor image segmentation accuracy in all aspects, e.g., the average Dice coefficients of these three evaluations for the second method are Dice_WT=0.8691, Dice_TC=0.7845 and Dice_ET=0.7002.

[1] B. H. Menze et al., The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS). In IEEE Transactions on Medical Imaging, vol.34, pp. 1993-2024, 2015.
[2] O. Ronneberger, P. Fischer, and T. Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation. In MICCAI, LNCS, vol. 9351, pp. 234–241, Springer, Cham, 2015.
[3] A. Bansal et al., PixelNet: Representation of the Pixels, by the Pixels, and for the Pixels. arXiv:1702.06506, 2017.
[4] G. Pleiss et al., Memory-Efficient Implementation of DenseNets. arXiv:1707.06990, 2017.
[5] S. Ioffe and C. Szegedy, Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. In Proceedings of The 32nd International Conference on Machine Learning, pp. 448–456, 2015.
[6] D. P. Kingma and J. L. Ba, Adam: a method f or stochastic optimization. arXiv:1412.6980, 2014.
[7] Y. LeCun et al., Gradient-based Learning Applied to Document Recognition. In Proceedings of the IEEE, pp. 2278–2324, 1998.
[8] A. Krizhevsky, I. Sutskever and G. E. Hinton, ImageNet Classification with Deep Convolutional Neural Networks. In NIPS, pp. 1106–1114, 2012.
[9] K. Simonyan et al., Very Deep Convolutional Networks for Large Scale Image Recognition. arXiv:1409.1556, 2014.
[10] C. Szegedy et al., Going Deeper with Convolutions. In CVPR, pp. 1 – 9, 2015.
[11] C. Szegedy et al., Rethinking the Inception Architecture for Computer Vision. arXiv:1512.00567, 2015.
[12] K. He et al., Deep Residual Learning for Image Recognition. arXiv:1512.03385, 2015.
[13] J. Long, E. Shelhamer and T. Darrell, Fully Convolutional Networks for Semantic Segmentation. In CVPR, pp. 3431-3440, 2015.
[14] E. Puybareau et al., Segmentation of Gliomas and Prediction of Patient Overall Survival: A Simple and Fast Procedure. In MICCAI, LNCS, vol. 11384. Springer, Cham, 2018.
[15] A. Kermi, I. Mahmoudi and M. T. Khadir, Deep Convolutional Neural Networks Using U-Net for Automatic Brain Tumor Segmentation in Multimodal MRI Volumes. In MICCAI, LNCS, vol. 11384. Springer, Cham, 2018.
[16] D. E. Cahall et al., Inception Modules Enhance Brain Tumor Segmentation. Front. Comput. Neurosci., vol. 13, article 44, 2019.
[17] Q. Dou et al., 3D Deeply Supervised Network for Automatic Liver Segmentation from CT Volumes. arXiv:1607.00582, 2016.
[18] T. Zhou et al., Deep Learning Model Integrating Dilated Convolution and Deep Supervision for Brain Tumor Segmentation in Multi-Parametric MRI. In MLMI, LNCS, vol. 11861. Springer, Cham, 2019.
[19] S. Bakas et al., Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features, Nature Scientific Data, 2017.
[20] S. Bakas et al., Identifying the Best Machine Learning Algorithms for Brain Tumor Segmentation, Progression Assessment, and Overall Survival Prediction in the BRATS Challenge. arXiv:1811.02629, 2018.
[21] K. He et al., Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. arXiv:1502.01852, 2015.
[22] F. Milletari et al., V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation. arXiv:1606.04797, 2016.
[23] W. Crum, O. Camara, D. Hill, Generalized Overlap Measures for Evaluation and Validation in Medical Image Analysis. In IEEE, TMI 25(11), pp. 1451–1461, 2006.
[24] P. A. Yushkevich et al., User-guided 3D Active Contour Segmentation of Anatomical Structures: Significantly Improved Efficiency and Reliability. Neuroimage, pp. 1116 – 1128, 2006.
[25] A. Myronenko, 3D MRI Brain Tumor Segmentation Using Autoencoder Regularization. In MICCAI, LNCS, vol. 11384. Springer, Cham, 2018.