根據衛生福利部公布最新國人十大死因，癌症連續35 年位居十大死因之首，其中又以肺癌的死亡率最高。根據多項研究結果顯示，針對罹患肺癌的高風險的特定族群，使用低劑量電腦斷層篩檢肺癌，比起使用傳統的胸部X光篩檢，能降低20%的肺癌死亡率。隨著深度學習在醫療領域愈來愈成熟，U-Net神經網絡已被廣泛應用在分析醫學影像分析，而肺葉分割是肺部疾病的一項重要任務。經典的肺葉分割方法依賴於成功檢測裂縫和其他解剖信息，例如血管和氣管的位置。本研究提出以深度學習的U-Net分割模型，用於分割電腦斷層影像的肺部區域，達到能準確分割肺部五個肺葉。藉此功能，協助醫師在肺癌篩檢防治工作上能預測肺結節位置，並且提供後續的臨床建議，以此作為醫師在影像判讀上的參考，進而減低醫師的工作負擔、降低人為疏失的發生機率。

According to the latest top ten causes of death in Taiwan (ROC) announced by the Ministry of Health and Welfare, cancer has ranked first among the top ten causes of death for 35 consecutive years, among which lung cancer has the highest mortality rate. Low-dose computed tomography screening for lung cancer can reduce lung cancer mortality by 20% compared to traditional chest X-ray screening. As deep learning becomes more and more mature in the medical field, U-Net neural network has been widely used to analyze medical images. Because lobe segmentation is an important task in lung disease, classic lobe segmentation methods rely on successful detection of cracks and other anatomical information such as the location of blood vessels and trachea. In this study, an enhanced deep learning U-Net segmentation model was proposed to segment the lung region of computed tomography images, so as to accurately segment the five lung lobes of the lung. This function can help physicians to predict the location of pulmonary nodules in lung cancer screening and prevention work. In addition, it can provide follow-up clinical advice as a reference for physicians in image interpretation, thereby reducing the workload of physicians and the probability of human error.

中文摘要	ii
Abstract	iii
目錄	iv
表目錄	vi
圖目錄	vii
第 1 章 緒論	1
1.1	研究背景	1
1.2	研究動機與目的	2
1.3	論文架構	3
第 2 章 文獻回顧	4
2.1	MCP神經元模型	4
2.2	感知機(Perceptron)	4
2.3	適應線性神經元(Adaline)	8
2.4	倒傳遞網路(Backpropagation)	11
2.5	自編碼器(AutoEncoder)	16
2.6	LeNet	19
2.7	AlexNet	24
2.8	VGG (Visual Geometry Group)	29
2.9	GoogLeNet (Inception)	34
2.10	ResNet	41
2.11	FCN (Fully Connected Network)	43
2.12	U-Net	48
第 3 章 增強U-Net分割肺葉	50
3.1	增強U-Net分割肺部區域	51
3.2	增強U-Net分割五個肺葉	55
第 4 章 實驗實作與結果討論	56
4.1	實驗設備與樣本環境	56
4.2	肺部解剖位置介紹	56
4.3	效能衡量指標	58
4.4	實驗結果	59
4.5	輸出成果呈現	61
4.6	容易出現的錯誤	62
第 5 章 結論及未來展望	64

表目錄
表 1 LeNet-5卷積神經網路各層功能說明	23
表 2 AlexNet卷積神經網路各層功能說明	28
表 3 VGGNet多種不同版本 	30
表 4 VGG各種不同模型的比較	33
表 5 GoogLeNet模型的參數量與計算量 	40
表 6各種不同深度的ResNet網路與所使用的Residual Block 	43
表 7分割肺部U-Net網路模型說明	52
表 8混淆矩陣	58
表 9 Unet預測結果分類報告	60
表 10增強U-Net預測結果分類報告	60
表 11常見預測錯誤原因	62

圖目錄
圖 1神經細胞(神經元)	5
圖 2 感知機(Perceptron) 	6
圖 3. 線性可分割 	7
圖 4梯度下降法示意圖 	9
圖 5 Adaline 	10
圖 6 Sigmoid函數	12
圖 7三層的神經網路架構	12
圖 8 原始自編碼器	18
圖 9 多層自編碼器	19
圖 10 卷積運算	20
圖 11 具有padding的卷積運算	21
圖 12 池化運算	22
圖 13 LeNet-5卷積神經網路 	23
圖 14 AlexNet卷積神經網路(2個GPU)	24
圖 15 ReLU函數圖	25
圖 16 AlexNet神經網路架構簡化版(1個GPU)	26
圖 17 LeNet與AlexNet的網路架構對照圖	28
圖 18 ImageNet 歷屆ILSVRC的方法與辨識錯誤率	29
圖 19 VGG-16神經網路模型	31
圖 20 兩個3x3卷積堆疊的感受野大小相當於一個5x5卷積	32
圖 21 VGG的訓練與測試	33
圖 22 Inception模組原始(naive)版本	35
圖 23 Inception模組原始(naive)版本	36
圖 24 Inception原始模組5x5卷積核所需的參數數量與計算量	36
圖 25 Inception-V1模組	37
圖 26 Inception-V1模組	37
圖 27 Inception-V1模組5x5卷積核所需的參數數量與計算量	38
圖 28 GoogLeNet網路模型架構圖 	39
圖 29 20層與56層卷積網路在CIFAR-10資料集的分類錯誤率	41
圖 30 Residual Block 	42
圖 31 ResNet-18與ResNet-34在ImageNet資料集的分類錯誤率	42
圖 32 兩種不同Residual Block [15]	43
圖 33 物件分類、定位、偵測、語意分割、與實例分割的區別	44
圖 34 語意分割	45
圖 35 FCN將傳統卷積神經網路的全連接層卷積化 	46
圖 36 FCN將反卷積的21張熱圖的像素逐一進行預測 	46
圖 37 FCN的上採樣(以VGG16模型為範例) 	47
圖 38 FCN不同上採樣方式(32s, 16s, 8s)預測的結果	48
圖 39 U-Net卷積神經網路模型 	49
圖 40 3x3 Convolution + ReLU	50
圖 41 2x2 Up-Convolution	50
圖 42分割肺部區域的增強U-Net網路模型	51
圖 43分割五個肺葉的增強U-Net網路模型	55
圖 44實驗設備與設置的軟體環境	56
圖 45肺部五個肺葉示意圖	57
圖 46肺部周圍器官示意圖	57
圖 47 預測五個肺葉之示意圖	58
圖 48低劑量電腦斷層預測之影像	61
圖 49電腦斷層預測之影像	61
圖 50細切電腦斷層預測之影像	62

[1]	W.S. McCulloch and W. Pitts, “A Logical Calculus of the Ideas Immanent in Nervous Activity,” Bulletin of Mathematical Biophysics, Vol. 5, No. 4, pp. 115-133, 1943.
[2]	F. Rosenblatt, “The Perceptron: A Perceiving and Recognizing Automaton,” Technical Report 85-460-1, Cornell Aeronautical Laboratory, 1957.
[3]	S. Raschka and V. Mirjalili, Python Machine Learning, 3rd Ed., Packt, 2020.
[4]	B. Widroff, “An Adaptive "Adaline" Neuron Using Chemical "Memistors",” Technical Report Number 1553-2, Standard Electronics Laboratories, 1960.
[5]	D.E. Rumelhart, G.E. Hinton, and R.J. Williams, “Learning Internal Representation by Error Propagation,” Parallel Distributed Processing, Vol. 1, MIT Press, 1986.
[6]	G.E. Hinton, S. Osindero, Y.W. The, “A Fast Learning Algorithm for Deep Belief Nets,” Neural Computation, Vol. 18, No. 7, pp. 1527-1554, 2006.
[7]	H. Bourlard and Y. Kamp, “Auto-Association by Multilayer Perceptrons and Singular Value Decomposition,” Biological Cybernetics, Vol. 59, pp. 291-294, 1988.
[8]	G. E. Hinton and R. S. Zemel, “Autoencoders, Minimum Description Length, and Helmholtz Free Energy,” in Advances in Neural Information Processing Systems 6 (NIPS’93), pp. 3-10, Morgan Kaufmann Publishers, Inc., 1994.
[9]	Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-Based Learning Applied to Document Recognition,” Proceedings of the IEEE, Vol. 86, No. 11, pp. 2278-2324, 1998.
[10]	C. Cortes and V. Vapnik, “Support-Vector Networks,” Machine Learning, Vol. 20, No. 3, pp. 273-297, 1995.
[11]	A. Krizhevsky, I. Sutskever, and G.E. Hinton, “Image Classification with Deep Convolutional Neural Networks,” Advances in Neural Information Processing Systems, pp. 1097-1105, 2012.
[12]	K. Simonyan and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” International Conference on Learning Representation (ICLR 2015), pp.1-14, 2015.
[13]	C. Szegedy et. al., “Going Deeper with Convolutions,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2015), pp. 1-9, 2015.
[14]	M. Lin, Q. Chen, and S. Yan, “Network in Network,” arXiv:1312.4400v3, 2014.
[15]	K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2016), pp. 770-778, 2016.
[16]	J. Long, E. Shelhamer, and T. Darrell, “Fully Convolutional Networks for Semantic Segmentation,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2015), pp. 3431-3440, 2015.
[17]	O. Ronneberger, P. Fisher, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation,” MICCAI 2015, LNCS 9351, pp. 234-241, 2015.
[18]	黃士嘉、林邑撰，人工智慧深度學習實作開發，第三版，博碩文化，2021。
[19]	K. He, X. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770-778, 2016.

[20]	S. Ren, K. He, R. Girshick, and J. Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. In NIPS, 2015.

[21]	W. Li, D. Eigen, and R. Fergus, “End-to-end integration of a Convolutional Network, Deformable Parts Model and non-maximum suppression,”  IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 851-859, 2015.

[22]	Ronneberger, O., Fischer, P., & Brox, T.  U-net: Convolutional networks for biomedical image segmentation. arXiv:1505.04597 [cs] , http://arxiv.org/abs/1505.04597,2015.

[23]	衛生福利部死因統計: https://dep.mohw.gov.tw/DOS/np-1776-113.html.

[24]	NSCLC: https://en.wikipedia.org/wiki/Non-small-cell_lung_carcinoma

[25]	肺腺癌，如何有效篩檢?只做胸部X光檢查是否足夠: https://www.tyh.com.tw/health/b_special_p29.html.

[26]	低劑量電腦斷層肺癌篩檢: https://www.mil.mohw.gov.tw/public/hygiene/222a6c0286f97139cd439c1945641509.pdf