超音波(ultrasound)是目前應用於腹部急症常見的檢查工具之一，在臨床實務操作上具有非侵入性、無輻射線、低成本、操作靈活且能提供即時影像的優點。但由於不同操作者之間對影像品質的要求、造影參數設定、操作經驗及受檢者的配合程度皆不盡相同，造成超音波影像之對比度與明暗度不一致，使診斷鑑別困難。因此，運用合適的輔助工具以快速取得標的正確的超音波影像，可提升臨床檢查之效率並降低操作門檻。本研究利用錄製30位受檢者之動態超音波影像(MPEG-4格式)轉換成單張靜態影像(JPG格式)，挑選出共885張腹部超音波靜態影像，使用卷積神經網路(convolutional neural network, CNN)模型：(i) visual geometry group-19 (VGG-19)、(ii) VGG-19結合support vector machine (SVM)、(iii) VGG-19轉移學習模型及(iv) ResNet50轉移學習模型共四種類型，進行自動辨識腹部超音波靜態影像視野(field of view, FOV)內之膽囊影像，並比較各模型分類結果，找出最佳分類模型。模型效能評比項目包含準確度、靈敏度、特異性、陽性預測值(positive predicted value, PPV)、陰性預測值(negative predicted value, NPV)及一致性係數(Kappa)。實驗結果證明經轉移學習之VGG-19與ResNet50模型最佳，其準確度、靈敏度、特異性、PPV、NPV及Kappa值皆顯著提升(皆大於98%)。本研究利用錄製人體腹部超音波之動態影像，經由人工智慧(artificial intelligence, AI)建構影像自動辨識及分類模型，以輔助經驗不足的操作者能在較短時間內獲得所需之影像，並滿足無法配合閉氣者之造影需求，研究結果可應用於救護車上裝備做為病患到院前之診斷。

Ultrasound is one of the common inspection tools commonly used in abdominal emergencies. It has the advantages of non-invasive, no radiation, low cost, flexible operation and can provide real-time images in clinical practice. However, due to complexity of parameter setting, experience of the operator, and uncooperative of the patient usually resulting in bad quality of images. In this study, I recorded the dynamic ultrasound images (MPEG-4 format) from 30 subjects and converted into a single static image (JPG format) to obtain 885 ultrasound images from the abdomen. Four convolutional neural network (CNN) models including (i) Visual Geometry Group-19 (VGG-19), (ii) VGG-19 in conjunction with Support Vector Machine (SVM), (iii) transfer learning from VGG-19, and (iv) transfer learning from ResNet50 were applied to automatically identify the ultrasound static images of gallbladder in the field of view (FOV). The accuracy, sensitivity, specificity, positive predicted value (PPV), negative predicted value (NPV), and Kappa value among four models were compared respectively. Transfer learning from VGG-19 and ResNet50 models revealed the best outcome. These results indicated that combination of VGG-19 and ResNet50 models could be used to establish deep learning algorithm for acquiring dynamic images of ultrasound. The disadvantages of ultrasound can be solved by using recording dynamic images of the abdomen through artificial intelligence (AI) to construct automatic image recognition and classification models. These models can be used to assist inexperienced operators and to obtain high quality images in a shorter time. This method could also be further expanded on ambulances so that emergency physicians can perform pre-medical treatment before patients arrive at the hospital.

第一章 緒論 1
第一節 前言 1
第二節 超音波造影原理 3
第三節 影像辨識原理 6
第四節 研究動機與目的 9
第二章 文獻探討 11
第一節 影像辨識於醫學影像之應用 11
第二節 動態造影於臨床影像之應用 13
第三節 卷積神經網路(convolution neural network, CNN) 15
第四節 CNN模型 17
第三章 研究方法與步驟 21
第一節 研究流程 21
第二節 造影條件 21
第三節 影像前處理 23
第四節 VGG-19模型 23
第五節 SVM分類器 25
第六節 轉移學習 26
第七節 評估模型之方法 28
第四章 結果 33
第一節 MPEG超音波影像擷取暨分類結果 33
第二節 轉移學習模型分類結果 35
第三節 轉移學習模型訓練分析 37
第四節 多次訓練模型之結果 47
第五節 ResNet50轉移學習全數據集實驗結果 49
第六節 結論 50
第五章 討論 51
第一節 討論 51
第一節 研究限制 56
第一節 未來研究方向 57
參考文獻 59
附錄一 VGG-19層數設計及其參數 67
附錄二 轉移學習之ResNet50層數設計及其參數 69
附錄三 VGG-19、VGG-19+SVM及ResNet50轉移學習模型100次訓練結果 77

[1] Louis J. Acierno, and L. Timothy Worrell. 2002. Father of Echocardiography. Clinical Cardiology 25:197-99.
[2] G. E. Hinton, and R. R. Salakhutdinov. 2006. Reducing the Dimensionality of Data with Neural Networks. Science 313:504-7.
[3] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. HintonImagenet. 2012. Classification with deep convolutional neural networks. Advances in neural information processing systems 25.
[4] Kunihiko Fukushima. 1980. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological Cybernetics 36:193-202.
[5] LeCun Yann, Bottou Leon, Bengio Yoshua, and Haffner Patrick. 1998. Gradient-based learning applied to document recognition. Proceedings of the IEEE 86: 2278-324.
[6] LeCun Yann, Bengio Yoshua, and Hinton Geoffrey. 2015. Deep learning. Nature 521:436-44.
[7] Marie Ploquin, Adrian Basarab, and Denis Kouamé. 2015. Resolution enhancement in medical ultrasound imaging. Journal of Medical Imaging 2.
[8] Robert W. Cootney. 2001. Ultrasound Imaging Principles and Applications in Rodent Research. ILAR Journal 42:233-47.
[9] Szegedy Christian, Liu Wei, Jia Yang, Sermanet Pierre, Reed Scott, Anguelov Dramogir, Erhan Dumitru, Vanhoucke Vincent, and Rabinovich Andrew. 2015. Going deeper with convolutions. IEEE Conference on Computer Vision and Pattern Recognition:1-9.
[10] Simonyan Karen, and Zisserman Andrew. 2014. Very deep convolutional networks for large-scale image recognition. Computer Science arXiv:1409.1556.
[11] Matthew D. Zeiler, and Rob Fergus. 2014. Visualizing and understanding convolutional networks. European Conference on Computer Vision:818-33.
[12] Eigen David, Rolfe Jason, Fergus Rob, and LeCun Yann. 2013. Understanding deep architectures using a recursive convolutional network. Machine Learning arXiv:1312.1847.
[13] Sergey Ioffe, Christian Szegedy. 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. Machine Learning arXiv:1502.03167.
[14] Yichuan Tang. 2013. Deep Learning using Linear Support Vector Machines. Machine Learning arXiv:1306.0239.
[15] Ren Wu, Shengen Yan, Yi Shan, Qingqing Dang, and Gang Sun. 2015. Deep image: Scaling up image recognition. Machine Learning arXiv:1501.02876.
[16] Abien Fred Agarap. 2017. Towards Building an Intelligent Anti-Malware System: A Deep Learning Approach using Support Vector Machine (SVM) for Malware. Classification Neural and Evolutionary Computing arXiv:1801.00318.
[17] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. 2012. Imagenet classification with deep convolutional neural networks. Communications of the ACM: 1097-105.
[18] Tsung-Hsien Wen, Milica Gašić, Nikola Mrkšić, Pei-Hao Su, David Vandyke, and Steve Young. 2015. Semantically conditioned lstm-based natural language generation for spoken dialogue systems. Conference on Empirical Methods in Natural Language Processing:1711-21.
[19] Jan Chorowski, Dzmitry Bahdanau, Dmitriy Serdyuk, Kyunghyun Cho, and Yoshua Bengio. 2015. Attention-based models for speech recognition. Proceedings of the 28th International Conference on Neural Information Processing Systems 1: 577-85.
[20] Zichao Yang, Diyi Yang, Chris Dyer, Xiaodong He, Alex Smola, and Eduard Hovy. 2016. Hierarchical Attention Networks for Document Classification. Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies:1480-9.
[21] Agarap A.F.. 2018. A Neural Network Architecture Combining Gated Recurrent Unit (GRU)and Support Vector Machine (SVM)for Intrusion Detection in Network Traffic Data. 10th International Conference on Machine Learning and Computing: 26-30.
[22] Abdulrahman Alalshekmubarak, and Leslie S. Smith. 2013. A novel approach combining recurrent neural network and support vector machines for time series classification. 9th International Conference on Innovations in Information Technology:42-7.
[23] Tarek M. Hassan, Mohammed Elmogy, and El-Sayed Sallam. 2017. Diagnosis of focal liver diseases based on deep learning technique for ultrasound images. Arabian Journal for Science and Engineering 42:3127-40.
[24] Nima Tajbakhsh, Jae Y. Shin, Suryakanth R. Gurudu, R. Todd Hurst, Christopher B. Kendall, Michael Gotway, Jianming Liang. 2017. On the necessity of fine-tuned convolutional neural networks for medical imaging. Deep Learning and Convolutional Neural Networks for Medical Image Computing: Precision Medicine, High Performance and Large-Scale Datasets:181-93.
[25] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. IEEE Conference on Computer Vision and Pattern Recognition:770-8.
[26] Xiang Chun Li, Sheng Zhang, Qiang Zhang, Xi Wei, Yi Pan, and Jing Zhao. 2019. Diagnosis of thyroid cancer using deep convolutional neural network models applied to sonographic images: a retrospective, multicohort, diagnostic study. The Lancet Oncology 20:193-201.
[27] Lingyun Wu, Jie-Zhi Cheng, Shengli Li, Baiying Lei, Tianfu Wang, and Dong Ni. 2017. FUIQA: fetal ultrasound image quality assessment with deep convolutional networks. IEEE Transactions on Cybernetics 47:1336-49.
[28] Tomoko Yamakawa, Shin-Ichi Toyabe, Pengyu Cao, and Kouhei Akazawa. 2004. Web-based delivery of medical multimedia contents using an MPEG-4 system. Computer Methods and Programs in Biomedicine 75:259-64.
[29] Andrea Giordano, Fabio Comazzi, Francesco Scapellato, Ermanno Eleuteri, Pantaleo Giannuzzi, and Giuseppe Minuco. 2009. Web-based delivery of medical multimedia contents using an MPEG-4 system. Istituto Superiore di Sanità 45:372-7.
[30] P. Gabriel Peterson, Sung K. Pak, Binh Nguyen, Genevieve Jacobs, and Les Folio. 2012. Extreme compression for extreme conditions: pilot study to identify optimal compression of CT images using MPEG-4 video compression. Journal of Digital Imaging 25:764-70.
[31] Andreas S. Panayides, Marios S. Pattichis, Christos P. Loizou, Marios Pantziaris, Anthony G. Constantinides, Constantinos S. Pattichis. 2014. An Effective Ultrasound Video Communication System Using Despeckle Filtering and HEVC. Journal of Biomedical and Health Informatics 19:668-76.
[32] Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein. 2015. Imagenet large scale visual recognition challenge. International Journal of Computer Vision 115:211-52.
[33] Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. 1998. Gradient-Based Learning Applied to Document Recognition. Proceedings of the IEEE: 86:2278-324.
[34] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. 2012. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems:1097-105.
[35] Karen Simonyan, Andrew Zisserman. 2014. Very deep convolutional networks for large-scale image recognition. Computer Science arXiv:1409.1556.
[36] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. conference on computer vision and pattern recognition:770-8.
[37] Corinna Cortes, and Vladimir Vapnik. 1995. Support-vector networks. Machine Learning 20:273-97.
[38] J.A.K. Suykens, and J. Vandewalle. 1999. Least squares support vector machine classifiers. Neural Processing Letters 9:293-300.
[39] Mohamed Elleucha, Rania Maalejb, Monji Kherallahc. 2016. A new design based-SVM of the CNN classifier architecture with dropout for offline Arabic handwritten recognition. Procedia Computer Science 80:1712-23.
[40] Chuanqi Tan, Fuchun Sun, Tao Kong, Wenchang Zhang, Chao Yang, Chunfang Liu. 2018. Survey on Deep Transfer Learning. International Conference on Artificial Neural Networks 11141:270-9.
[41] Hoo-Chang Shin, Holger R. Roth, Mingchen Gao, Le Lu, Ziyue Xu, Isabella Nogues, Jianhua Yao, Daniel Mollura, Ronald M. Summers. 2016. Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning. IEEE Trans Med Imaging 35:1285-98.
[42] Daniel S. Kermany, Michael Goldbaum, Wenjia Cai, Carolina C.S. Valentim, Huiying Liang, Sally L. Baxter, Alex McKeown, Ge Yang, Xiaokang Wu, Fangbing Yan, Justin Dong, Made K. Prasadha, Jacqueline Pei1, Magdalene Y. L. Ting, Jie Zhu, Christina Li, Sierra Hewett, Jason Dong, and KangZhang. 2018. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell 172:1122-31.
[43] Michal Uricár, Radu Timofte, Rasmus Rothe, Jirí Matas, and Luc Van Gool. 2016. Structured Output SVM Prediction of Apparent Age, Gender and Smile From Deep Features. IEEE CVPRW.
[44] Di-Xiu Xue, Rong Zhang, Hui Feng, and Ya-Lei Wang. 2016. CNN-SVM for Microvascular Morphological Type Recognition with Data Augmentation. Journal of Medical and Biological Engineering vol 36:755-64.
[45] Ladislav Rampasek, and Anna Goldenberg. 2018. Learning From Everyday Images Enables Expert-like Diagnosis of Retinal Diseases. Cell 172(5):893-95.
[46] Varun Gulshan, Lily Peng, Marc Coram, Martin C. Stumpe, Derek Wu, Arunachalam Narayanaswamy, Subhashini Venugopalan, Kasumi Widner, Tom Madams, Jorge Cuadros, Ramasamy Kim, Rajiv Raman, Philip C. Nelson, Jessica L. Mega, and Dale R. Webster. 2016. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. JAMA 316(22):2402-10.
[47] Andre Esteva, Brett Kuprel, Roberto A. Novoa, Justin Ko, Susan M. Swetter, Helen M. Blau, and Sebastian Thrun. 2017. Dermatologist-level classification of skin cancer with deep neural networks. Nature vol 542:115-18.
[48] Priya Goyal, Piotr Dollár, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, Andrew Tulloch, Yangqing Jia, and Kaiming He. 2017. Large Minibatch SGD: Training ImageNet in 1 Hour. arXiv:1706.02677.
[49] Nitish Shirish Keskar, Dheevatsa Mudigere, Jorge Nocedal, Mikhail Smelyanskiy, and Ping Tak Peter Tang. 2017. On large-batch training for deep learning: Generalization gap and sharp minima. arXiv:1609.04836.
[50] Elad Hoffer, Itay Hubara, and Daniel Soudry. 2017. Train longer, generalize better: closing the generalization gap in large batch training of neural networks. arXiv:1705.08741.