臺灣博碩士論文加值系統

近年來，隨著人工智慧（Artificial Intelligence，簡稱AI）和相關技術的飛速發展，醫療影像AI浪潮席捲世界各地，研發深度學習於醫療影像的應用，藉由AI來輔助醫生診斷、提升流程效能，賦予醫師更大能力以提供患者更好的醫療照護，是積極前進的目標。
氣胸(Pneumothorax)在急診部門並不罕見。在嚴重的情況下，如果不緊急處理，將存在致命的危險。建立一個急診醫師通過使用人工智慧科技加快診斷過程的氣胸檢測系統，將有極大助益。
本研究基於生成對抗網絡(Generative Adversarial Network，簡稱GAN)深度學習演算法，包括生成器和鑑別器，由本實驗室團隊進行修改和開發一個自動標記醫學影像的程式系統，放射科技生成對抗網路(Radiologic Technology Generative Adversarial Network，簡稱RTGAN)，能在短時間內自動偵測氣胸，藉由不斷學習，RTGAN可以利用訓練少量影像數據來提高訓練後影像質量。同時本實驗中，也使用邊界平衡生成對抗網絡（Boundary Equilibrium Generative Adversarial Networks ，簡稱BEGAN）來增加影像數據量。
使用RTGAN生成對抗網絡進行迭代訓練，大約3秒鐘內即可在每個影像中標記異常位置，準確性可達到近九成。系統立即快速篩選並自動標記氣胸位置，可以輔助急診部門醫師迅速制定治療指南，為患者提供優質高效的治療。

In recent years, with the rapid development of artificial intelligence (AI) and related technologies, the medical imaging AI wave swept the world, research and development of deep learning in medical imaging applications, using AI to assist doctors in diagnosis, improve process efficiency, and empower physicians Ability to provide patients with better medical care.
Pneumothorax is not uncommon in the emergency department. In severe cases, there is a fatal danger if not handled urgently. To establish a pneumothorax detection system that can assist emergency doctor speed up the diagnosis process by using artificial intelligence techniques.
This study modifies and developed by a GAN-based deep learning algorithm, which includes both generator and discriminator. Established a system RTGAN for labeled medical images, which immediately displayed to the physician for reference in the shortest time. Through continuous learning, it improves the image quality with training a small amount of data.
The boundary balance generation adversarial network (BEGAN) was used to increase the amount of image data, and evaluate the effectiveness of improving the quality of the generated labeled image.
Iterative training is performed using generative adversarial networks, labeling abnormal locations in each image in approximately 3 seconds. The accuracy of automatically labeling abnormal locations can reach nearly 90%. The system quickly screens and automatically labeled the pneumothorax. It can assist emergency physicians to quickly develop treatment guidelines and provide patients with high-quality and effective treatment.

目錄
中文摘要 I
英文摘要 II
致謝 III
目錄 V
圖目錄 VIII
表目錄 X
公式目錄 XI
第一章 緒論 1
第一節 研究動機 1
第二節 研究背景 2
壹、氣胸(Pneumothorax) 2
貳、氣胸影像學 3
第三節 研究目的 6
第四節 章節安排 8
第二章 文獻回顧與探討 9
第一節 深度學習(Deep Learning) 9
第二節 深度學習在氣胸上的運用 9
第三節 GAN的架構 11
第四節 GAN在醫學影像領域的運用 13
第三章 研究方法 15
基於BEGAN和RTGAN之氣胸影像的擴增與氣胸區域的偵測 15
第一節 數據集收集 15
第二節 計算環境設置 17
一、搭建環境Setting up the environment 17
二、軟體設置 18
三、硬體設置 18
第三節 RTGAN與BEGAN的研究架構 19
一、RTGAN 19
二、BEGAN 26
第四章 研究實驗與觀察 27
第一節 改變數據集大小 27
第二節 Iteration 觀察 32
第三節 Batch_Size 調整 35
第四節 Training Dataset 調整 38
第五節 BEGAN 的運用 40
第六節 住院醫師標記影像 42
第七節 統計基礎 42
第五章 研究結果與探討 47
第一節 Resident test 48
第二節 110 all PI train Model test 49
第三節 209 all PI train Model test 50
第四節 240train Model test (PI：NI= 7：1) 51
第五節 280train Model test (PI：NI= 3：1) 52
第六節 354train Model test (PI：NI= 3：2) 53
第七節 418train Model test (PI：NI= 1：1) 54
第八節 PI+BI train Model test 56
第九節 389 PI+NI+BI train Model test 57
第十節 資料分布整理 58
第六章 結論與未來展望 62
參考文獻 64

[1]. LUH, S.P., Diagnosis and treatment of primary spontaneous pneumothorax. 中華民國急救加護醫學會雜誌, 2010. 21: p. 9-20
[2]. Wang, Y., L.L. Sun, and Q. Jin, Enhanced Diagnosis of Pneumothorax with an Improved Real-time Augmentation for Imbalanced Chest X-rays Data Based on DCNN. IEEE/ACM Trans Comput Biol Bioinform, 2019.
[3]. 衛生福利部統計處, 107年度全民健康保險醫療統計年報, 衛生福利部統計處, Editor. 2020.
[4]. Fukushima, K., Neocognitron: a self organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biol Cybern, 1980. 36(4): p. 193-202.
[5]. Krizhevsky, A.a.S., Ilya and Hinton, Geoffrey E., ImageNet Classification with Deep Convolutional Neural Networks. Commun. ACM, 2017. 60(June 2017).
[6]. Taylor, A.G., C. Mielke, and J. Mongan, Automated detection of moderate and large pneumothorax on frontal chest X-rays using deep convolutional neural networks: A retrospective study. PLoS Med, 2018. 15(11): p. e1002697.
[7]. Chan, Y.H., et al., Effective Pneumothorax Detection for Chest X-Ray Images Using Local Binary Pattern and Support Vector Machine. J Healthc Eng, 2018. 2018: p. 2908517.
[8].Li, X., et al., Deep learning-enabled system for rapid pneumothorax screening on chest CT. European journal of radiology, 2019. 120: p. 108692-108692.
[9]. Nam, J.G., et al., Development and Validation of Deep Learning-based Automatic Detection Algorithm for Malignant Pulmonary Nodules on Chest Radiographs. Radiology, 2019. 290(1): p. 218-228.
[10]. Goodfellow I., P.-A.J., Mirza M., et al, Generative adversarial nets. Adv Neural Inf Process Syst, 2014. 27: p. 2672-2680.
[11]. Sorin, V., et al., Creating Artificial Images for Radiology Applications Using Generative Adversarial Networks (GANs) - A Systematic Review. Acad Radiol, 2020.
[12]. David Berthelot, T.S., Luke Metz, BEGAN: Boundary Equilibrium Generative
Adversarial Networks. 2017.
[13]. Hah, J., et al., Information-Based Boundary Equilibrium Generative Adversarial Networks with Interpretable Representation Learning. Comput Intell Neurosci, 2018. 2018: p. 6465949.
[14]. Armand Joulin, L.v.d.M., Allan Jabri, Nicolas Vasilache, Learning Visual Features from Large Weakly Supervised Data. 2015: arXiv.org.
[15]. Suhua Lei, H.Z., Ke Wang, Zhendong Su, How Training Data Affect the Accuracy and Robustness of Neural Networks for Image Classification, in International Conference on Learning Representations. 2019.