近年來，將深度學習模型應用於醫學影像的研究發展十分蓬勃，本論文將延續過往實驗室對於乳房超音波影像分割的研究，使用DRA-UNet、Trans-UNet、U2-Net這三個深度學習模型來分割乳房超音波影像，並引入集成學習的方法，將各種深度學習模型透過集成學習當中Stacking組合在一起，以更有效率的幫助醫生判斷腫瘤的大小及位置。
本論文將使用DRA-UNet、Trans-UNet、U2-Net這三個深度學習模型來分割乳房超音波影像並使用客觀的錯誤評估指標來進行比較，實驗結果顯示，U2-Net有四項評估指標為三者中最高;IOU來到77.67%、DSC來到87.4%、Sensitivity來到87.59%、ACC來到96.77%，DRA-Unet有兩項評估指標為三者中最高; Precision來到88.29%，Specificity來到98.36%。之後再使用集成學習當中的Stacking將這三個深度學習模型進行兩兩與全部的不同組合方式組合在一起，觀察使用Stacking進行組合之後的模型是否比未進行組合前的模型還要優秀，實驗結果顯示，使用全部三個模型組合的Stacking在重要的指標IOU、DSC、Sensitivity上表現都比只使用任兩個模型組合的Stacking效果要好，也比原來的三個個別模型為佳;IOU來到79.62%、DSC來到88.64%、Sensitivity來到89.26%。

In recent years, the research on applying deep learning models to medical imaging has developed very vigorously both abroad and domestic. This thesis will continue previous research on breast ultrasound image segmentation in our laboratory by using three deep learning models, namely the DRA-UNet, Trans-UNet, and U2-Net to segment breast ultrasound images. And introduce a method in ensemble learning called Stacking to combine various deep learning models in order to more efficiently help doctors determine the size and location of tumors.
Several objective error evaluation metrics are used to evaluate the aforementioned deep learning models. From the experimental results, U2-Net has the highest IOU of 77.67%, DSC of 87.4%, and Sensitivity of 87.59% and DRA-UNet has the highest Precision of 88.29% and Specificity of 98.36%. Then Stacking is used to combine these three models in four different combinations to improve the performance. From the experimental results, it is found that the Stacking of all three models outperforms the original three non-Stacking models as well as any combinations of Stacking of two models. Using Stacking to combine three model has the highest IOU of 79.62%%, DSC of 88.64%, and Sensitivity of 89.26%.

[1] 衛生福利部國民健康署; 109年死因統計結果分析(含附表).pdf from: https://www.mohw.gov.tw/cp-5017-61533-1.html
[2] 東元醫療財團法人 東元綜合醫院 高級健檢中心/林洋鈺醫師; from: https://www.tyh.com.tw/health/f10.html
[3] Wells, P. N. T., and M. Halliwell. "Speckle in ultrasonic imaging." Ultrasonics 19.5 (1981): 225-229.
[4] Liu, Bo, et al. "Probability density difference-based active contour for ultrasound image segmentation." Pattern Recognition 43.6 (2010): 2028-2042.
[5] Huang, Qinghua, Yaozhong Luo, and Qiangzhi Zhang. "Breast ultrasound image segmentation: a survey." International journal of computer assisted radiology and surgery 12.3 (2017): 493-507.
[6] Horsch, Karla, et al. "Automatic segmentation of breast lesions on ultrasound." Medical physics 28.8 (2001): 1652-1659.
[7] Horsch, Karla, et al. "Computerized diagnosis of breast lesions on ultrasound." Medical physics 29.2 (2002): 157-164.
[8] Horsch, Karla, et al. "Performance of computer-aided diagnosis in the interpretation of lesions on breast sonography." Academic radiology 11.3 (2004): 272-280.
[9] Rodrigues, Paulo S., and Gilson A. Giraldi. "Improving the non-extensive medical image segmentation based on Tsallis entropy." Pattern Analysis and Applications 14.4 (2011): 369-379.
[10] Xian, Min, Yingtao Zhang, and Heng-Da Cheng. "Fully automatic segmentation of breast ultrasound images based on breast characteristics in space and frequency domains." Pattern Recognition 48.2 (2015): 485-497.
[11] Kekre, H. B., and Pravin Shrinath. "Tumour delineation using statistical properties of the breast us images and vector quantization based clustering algorithms." International Journal of Image, Graphics and Signal Processing 5.11 (2013): 1.
[12] Moon, Woo Kyung, et al. "Tumor detection in automated breast ultrasound images using quantitative tissue clustering." Medical physics 41.4 (2014): 042901.
[13] Shan, Juan, H. D. Cheng, and Yuxuan Wang. "A novel segmentation method for breast ultrasound images based on neutrosophic l‐means clustering." Medical physics 39.9 (2012): 5669-5682.
[14] Lo, Chiao, et al. "Computer-aided multiview tumor detection for automated whole breast ultrasound." Ultrasonic imaging 36.1 (2014): 3-17.
[15] Huang, Yu-Len, and Dar-Ren Chen. "Watershed segmentation for breast tumor in 2-D sonography." Ultrasound in medicine & biology 30.5 (2004): 625-632.
[16] Huang, Yu-Len, and Dar-Ren Chen. "Automatic contouring for breast tumors in 2-D sonography." 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference. IEEE, 2006.
[17] Zhang, Ling, and Ming Zhang. "A fully automatic image segmentation using an extended fuzzy set." International Workshop on Computer Science for Environmental Engineering and EcoInformatics. Springer, Berlin, Heidelberg, 2011.
[18] Lo, C. M., Chen, R. T., Chang, Y. C., Yang, Y. W., Hung, M. J., Huang, C. S., & Chang, R. F.: Multi-dimensional tumor detection in automated whole breast ultrasound using topographic watershed. 2014, IEEE transactions on medical imaging, 33(7), 1503-1511.
[19] 徐詠璿, 結合影像特徵分析的距離正規化水平集演算法-用於乳房超音波影像的腫瘤範圍圈選, 碩士論文,電機工程學系, 國立清華大學. 2015.
[20] 鈕采紋, 以基於形態學得到初始輪廓之距離正規化水平集演化法做超音波乳房影像腫瘤區塊自動化切割, 碩士論文,電機工程學系, 國立清華大學. 2016.
[21] 謝洵, 以利用引導影像濾波器及L0梯度最小化平滑前處理與形態特徵得到初始輪廓之距離正規化水平集演算法做乳房超音波影像腫瘤區塊自動化切割, 碩士論文,電機工程學系, 國立清華大學. 2017.
[22] 林鉉博, 利用基於深度學習之Mask-RCNN於乳房超音波影像腫瘤區塊檢測以及輪廓切割, 碩士論文,電機工程學系, 國立清華大學. 2018.
[23] 何岳庭, 利用基於深度學習之DRA-UNet於乳房超音波影像腫瘤區塊切割, 碩士論文,電機工程學系, 國立清華大學. 2021.
[24] Fujioka, Tomoyuki, et al. "Distinction between benign and malignant breast masses at breast ultrasound using deep learning method with convolutional neural network." Japanese journal of radiology 37.6 (2019): 466-472.
[25] Fujioka, Tomoyuki, et al. "Classification of breast masses on ultrasound shear wave elastography using convolutional neural networks." Ultrasonic Imaging 42.4-5 (2020): 213-220.
[26] Yap, Moi Hoon, et al. "Automated breast ultrasound lesions detection using convolutional neural networks." , 2017, IEEE journal of biomedical and health informatics 22.4: 1218-1226.
[27] LeCun, Yann, et al. "Gradient-based learning applied to document recognition." Proceedings of the IEEE 86.11 (1998): 2278-2324.
[28] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in neural information processing systems 25 (2012).
[29] Long, Jonathan, Evan Shelhamer, and Trevor Darrell. "Fully convolutional networks for semantic segmentation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2015.
[30] Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015.
[31] Arnab, Anurag, et al. "Conditional random fields meet deep neural networks for semantic segmentation: Combining probabilistic graphical models with deep learning for structured prediction." IEEE Signal Processing Magazine 35.1 (2018): 37-52.
[32] Girshick, Ross, et al. "Rich feature hierarchies for accurate object detection and semantic segmentation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2014.
[33] Bishop, Christopher M., and Nasser M. Nasrabadi. Pattern recognition and machine learning. Vol. 4. No. 4. New York: springer, 2006.
[34] Nwankpa, Chigozie, et al. "Activation functions: Comparison of trends in practice and research for deep learning." arXiv preprint arXiv:1811.03378 (2018).
[35] Ganaie, M. A., and Minghui Hu. "Ensemble deep learning: A review." arXiv preprint arXiv:2104.02395 (2021).
[36] Birbil, S. Ilker, et al. "Learning with Subset Stacking." arXiv preprint arXiv:2112.06251 (2021).
[37] D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning representations by back-propagating errors,” Nature 323, 533–536 (1986).
[38] Lu, Lu, et al. "Dying relu and initialization: Theory and numerical examples." arXiv preprint arXiv:1903.06733 (2019).
[39] Basodi, Sunitha, et al. "Gradient amplification: An efficient way to train deep neural networks." Big Data Mining and Analytics 3.3 (2020): 196-207.
[40] Ruder, Sebastian. "An overview of gradient descent optimization algorithms." arXiv preprint arXiv:1609.04747 (2016).
[41] Koehrsen, Will. "Overfitting vs. underfitting: A complete example." Towards Data Science (2018).
[42] Srivastava, Nitish, et al. "Dropout: a simple way to prevent neural networks from overfitting." The journal of machine learning research 15.1 (2014): 1929-1958.
[43] Ioffe, Sergey, and Christian Szegedy. "Batch normalization: Accelerating deep network training by reducing internal covariate shift." International conference on machine learning. PMLR, 2015.
[44] Devlin, Jacob, et al. "Bert: Pre-training of deep bidirectional transformers for language understanding." arXiv preprint arXiv:1810.04805 (2018).
[45] Friederich, Simon. "Fine-tuning." , 2017, available from: https://seop.illc.uva.nl/entries/fine-tuning/
[46] Prechelt, Lutz. "Early stopping-but when?." Neural Networks: Tricks of the trade. Springer, Berlin, Heidelberg, 1998. 55-69.
[47] Chis. Using Constant Padding, Reflection Padding and Replication Padding with TensorFlow and Keras,Avalible from:
https://github.com/christianversloot/machine-learning-articles/blob/main/using-constant-padding-reflection-padding-and-replication-padding-with-keras.md,2020.
[48] Yu, Fisher, and Vladlen Koltun. "Multi-scale context aggregation by dilated convolutions." arXiv preprint arXiv:1511.07122 (2015).
[49] Dumoulin, Vincent, and Francesco Visin. "A guide to convolution arithmetic for deep learning." arXiv preprint arXiv:1603.07285 (2016).
[50] Kaiser, Lukasz, Aidan N. Gomez, and Francois Chollet. "Depthwise separable convolutions for neural machine translation." arXiv preprint arXiv:1706.03059 (2017).
[51] Yosinski, Jason, et al. "How transferable are features in deep neural networks?." Advances in neural information processing systems 27 (2014).
[52] Vaswani, Ashish, et al. "Attention is all you need." Advances in neural information processing systems 30 (2017).
[53] Dosovitskiy, Alexey, et al. "An image is worth 16x16 words: Transformers for image recognition at scale." arXiv preprint arXiv:2010.11929 (2020).
[54] Chen, Jieneng, et al. "Transunet: Transformers make strong encoders for medical image segmentation." arXiv preprint arXiv:2102.04306 (2021).
[55] Qin, Xuebin, et al. "U2-Net: Going deeper with nested U-structure for salient object detection." Pattern Recognition 106 (2020): 107404.
[56] Huang, Gao, et al. "Densely connected convolutional networks." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.
[57] Ibtehaz, Nabil, and M. Sohel Rahman. "MultiResUNet: Rethinking the U-Net architecture for multimodal biomedical image segmentation." Neural Networks 121 (2020): 74-87.
[58] Oktay, Ozan, et al. "Attention u-net: Learning where to look for the pancreas." arXiv preprint arXiv:1804.03999 (2018).