跳到主要內容

臺灣博碩士論文加值系統

(44.222.64.76) 您好!臺灣時間:2024/06/15 05:16
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:王寬
研究生(外文):WANG, KUAN
論文名稱:應用增強版YOLOv4於乳房腫瘤偵測
論文名稱(外文):Enhanced YOLOv4 for Breast Cancer Detection from X-ray Images
指導教授:張文中馬尚智
指導教授(外文):CHANG, WEN-CHUNGMA, SHANG-CHIH
口試委員:張文中馬尚智沐海王怡鈞葉家承
口試委員(外文):CHANG, WEN-CHUNGMA, SHANG-CHIHMOHAMMAD BASSAM MOHAMMAD ALKHALEEFAHWang, YI CHUNYEH, CHIA-CHENG
口試日期:2021-07-21
學位類別:碩士
校院名稱:國立臺北科技大學
系所名稱:電機工程系
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2021
畢業學年度:109
語文別:中文
論文頁數:49
中文關鍵詞:深度學習醫學影像物件偵測電腦輔助偵測
外文關鍵詞:Deep learningMedical imagingObject detectionComputer-aided detection
相關次數:
  • 被引用被引用:0
  • 點閱點閱:263
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
本研究利用增強版YOLOv4模型於乳房腫瘤偵測。近年來,深度學習正在醫學影像的領域上蓬勃發展,主要是利用電腦視覺或深度學習相關演算法去幫助醫師進行臨床輔助判斷或教學的用途,其優點在於可在短時間內進行大量圖像資料的分類、圈選及去背等,且不需耗費醫護人員端的人力資源,故近年來常被用於醫學資料分析及臨床醫學輔助診斷等用途。
本研究旨在將深度學習應用在為數不多的乳房腫瘤樣本上,設計一end-to-end (端對端) 系統去提取這些樣本的Region of interest (ROI) ,以利醫師進行病灶診斷,此也為Computer-aided detection (CAD,電腦輔助偵測) 系統的宗旨;本研究額外探究了高加索人種及亞洲人的乳房腫瘤特徵樣本差異,即使用其一的預訓練模型去訓 並預測另一個資料集的實驗方式。本研究採用2個資料集,分別是來自葡萄牙的公開資料集:Inbreast 及來自臺灣振興醫院的私有資料集-振興醫院資料集,將其利用有效前處理方法 (包含去標籤、大津法及gamma函數等) 結合後提升平均精度,再經過深度學習模型架構的改良: You Only Look Once version 4 (YOLOv4) 的加強版,這邊為改良YOLOv4的特徵萃取部分架構 (backbone) ,此架構改變了其中稱為殘差模塊 (Resblock) 在不同大小的特徵圖上的數量針對較大及較小的腫瘤作特化處理,使其對於乳房腫瘤的特徵萃取有更好的適性,進一步提升平均精度。本實驗主要貢獻有:提供一套乳腫瘤CAD系統,優化其對於乳房腫瘤資料的適性及泛化性,以利輔助醫師進行臨床診斷或做為教學之用,以及提供一套有效的擴充資料集,合Inbreast及振興資料集,資料集的大小與有效性對於深度學習的成敗至關重要。

In recent years, deep learning is widely adopted in the field of medical imaging. The recent development in Computer-aided detection (CAD) systems which utilize computer vision and deep learning algorithms, can help doctors to identify breast lesions quickly, conveniently, and accurately. Therefore, deep learning algorithms are widely adopted in medical image analysis and CAD systems.
In this research, I designed an end-to-end deep learning algorithm to detect breast tumors from two x-ray datasets. The first dataset is the Inbreast public dataset from Portugal. The second one is the Chen Hsin dataset from Taiwan. By combining both datasets (Inbreast and Chen Hsin) for training, it shows 69.25% Mean Accuracy Precision (mAP) from You Only Look Once version4 (YOLOv4) which is the best-fit model for this dataset. After effective preprocessing method, including getting rid of the labels, OTSU segmentation and gamma function, the mAP has been improved from 69.25% to 74.68%. Finaly, the mAP of the enhanced-YOLOv4 which I improved the backbone of YOLOv4 by changing the number of Resblocks. Originally 1, 2, 8, 8, 4 in different size of feature maps. The revised one is 1, 2, 12, 8, 8. The result shows mAP of 76.91%.
摘要 i
ABSTRACT iii
誌謝 iv
目錄 v
表目錄 vii
圖目錄 viii
1 緒論 1
1.1 醫學影像簡介 1
1.2 研究動機與目的 1
1.3 乳房攝影簡介 2
1.4 乳房攝影暨乳房超音波檢查分級 3
1.5 論文內容大綱 5
2 研究背景 6
2.1 Convolutional neural network (CNN) 模型發展簡介 6
2.2 物件偵測模型簡介與發展史 6
2.3 Regional convolutional neural network (R-CNN) 物件偵測模型系列 7
2.3.1 R-CNN系列模型總體比較 7
2.3.2 Regional convolutional neural network (R-CNN) 8
2.3.3 Fast regional convolutional neural network (Fast RCNN) 10
2.3.4 Faster regional convolutional neural network (Faster RCNN) 11
2.3.5 Mask regional convolutional neural network (Mask RCNN) 13
2.4 Single Shot multibox Detector (SSD) 物件偵測模型 14
2.5 You Only Look Once (YOLO) 物件偵測模型系列 17
2.5.1 YOLO系列模型總體比較 17
2.5.2 You Only Look Once version 1 (YOLOv1) 17
2.5.3 You Only Look Once version 2 (YOLOv2) 18
2.5.4 You Only Look Once version 3 (YOLOv3) 20
2.5.5 You Only Look Once version 4 (YOLOv4) 21
2.6 物件偵測模型評估指標 28
3 研究方法 30
3.1 實驗總流程 30
3.2 資料前處理 31
3.2.1 去標籤 31
3.2.2 OTSU segmentation (大津法) 31
3.2.3 上下去邊 34
3.3 確立基本模型樣板 35
3.4 YOLOv4改良 36
4 實驗結果 37
4.1 資料集取得與簡介及模型超參數設置 37
4.2 模型比較數據 40
4.3 跨資料集預訓練模型引入 41
4.4 混合資料集於不同輸入圖片大小下結果 42
4.5 資料增強暨模型修改後結果 43
4.6 各模型Precision to recall (P-R) 曲線 44
4.7 實驗結果圖示例 45
5 結論與未來展望 46
5.1 結論 46
5.2 未來展望 46
參考文獻 47
[1] A. Lagree, M. Mohebpour, N. Meti, K. Saednia, F.­I. Lu, E. Slodkowska, S. Gandhi, E.Rakovitch, A. Shenfield, A. Sadeghi­Naini,et al., “A review and comparison of breasttumor cell nuclei segmentation performances using deep convolutional neural networks,”Scientific Reports, vol. 11, no. 1, pp. 1–11, 2021.
[2] V. K. Singh, H. A. Rashwan, S. Romani, F. Akram, N. Pandey, M. M. K. Sarker, A. Saleh,M. Arenas, M. Arquez, D. Puig,et al., “Breast tumor segmentation and shape classifi­cation in mammograms using generative adversarial and convolutional neural network,”Expert Systems with Applications, vol. 139, p. 112855, 2020.
[3] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep con­volutional neural networks,”Advances in neural information processing systems, vol. 25,pp. 1097–1105, 2012.
[4] J. Redmon and A. Farhadi, “Yolo9000: Better, faster, stronger,” inProceedings of theIEEE conference on computer vision and pattern recognition, 2017, pp. 7263–7271.
[5] A. Bochkovskiy, C.­Y. Wang, and H.­Y. M. Liao, “Yolov4: Optimal speed and accuracyof object detection,”arXiv preprint arXiv:2004.10934, 2020.
[6] T.­Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramidnetworks for object detection,” inProceedings of the IEEE conference on computer visionand pattern recognition, 2017, pp. 2117–2125.
[7] J. R. Uijlings, K. E. Van De Sande, T. Gevers, and A. W. Smeulders, “Selective search forobject recognition,”International journal of computer vision, vol. 104, no. 2, pp. 154–171, 2013.
[8] R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurateobject detection and semantic segmentation,” inProceedings of the IEEE conference oncomputer vision and pattern recognition, 2014, pp. 580–587.
[9] P. F. Felzenszwalb and D. P. Huttenlocher, “Efficient graph­based image segmentation,”International journal of computer vision, vol. 59, no. 2, pp. 167–181, 2004.
[10] R. Girshick, “Fast r­cnn,” inProceedings of the IEEE international conference on com­puter vision, 2015, pp. 1440–1448.47
[11] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large­scale imagerecognition,”arXiv preprint arXiv:1409.1556, 2014.
[12] S. Ren, K. He, R. Girshick, and J. Sun, “Faster r­cnn: Towards real­time object detectionwith region proposal networks,”arXiv preprint arXiv:1506.01497, 2015.
[13] K. He, G. Gkioxari, P. Dollár, and R. Girshick, “Mask r­cnn,” inProceedings of the IEEEinternational conference on computer vision, 2017, pp. 2961–2969.
[14] J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic seg­mentation,” inProceedings of the IEEE Conference on Computer Vision and PatternRecognition (CVPR), Jun. 2015.
[15] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.­Y. Fu, and A. C. Berg, “Ssd:Single shot multibox detector,” inEuropean conference on computer vision, Springer,2016, pp. 21–37.
[16] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke,and A. Rabinovich, “Going deeper with convolutions,” inProceedings of the IEEE con­ference on computer vision and pattern recognition, 2015, pp. 1–9.
[17] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016,pp. 770–778.
[18] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real­time object detection,” inProceedings of the IEEE conference on computer vision andpattern recognition, 2016, pp. 779–788.
[19] J.RedmonandA.Farhadi,“Yolov3:Anincrementalimprovement,”arXiv preprint arXiv:1804.02767,2018.
[20] C.­Y. Wang, H.­Y. M. Liao, Y.­H. Wu, P.­Y. Chen, J.­W. Hsieh, and I.­H. Yeh, “Cspnet:A new backbone that can enhance learning capability of cnn,” inProceedings of theIEEE/CVF conference on computer vision and pattern recognition workshops, 2020,pp. 390–391.
[21] S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, and Y. Yoo, “Cutmix: Regularization strategyto train strong classifiers with localizable features,” inProceedings of the IEEE/CVFInternational Conference on Computer Vision, 2019, pp. 6023–6032.48
[22] G.Ghiasi, T.­Y.Lin, and Q.V. Le, “Dropblock: Aregularization methodfor convolutionalnetworks,”arXiv preprint arXiv:1810.12890, 2018.
[23] D. Misra, “Mish: A self regularized non­monotonic neural activation function,”arXivpreprint arXiv:1908.08681, vol. 4, 2019.
[24] M. Tan, R. Pang, and Q. V. Le, “Efficientdet: Scalable and efficient object detection,”inProceedings of the IEEE/CVF conference on computer vision and pattern recognition,2020, pp. 10781–10790.
[25] S. Woo, J. Park, J.­Y. Lee, and I. S. Kweon, “Cbam: Convolutional block attention mod­ule,” inProceedings of the European conference on computer vision (ECCV), 2018, pp. 3–19.
[26] T. Fawcett, “An introduction to roc analysis,”Pattern recognition letters, vol. 27, no. 8,pp. 861–874, 2006.
[27] N. Otsu, “A threshold selection method from gray­level histograms,”IEEE transactionson systems, man, and cybernetics, vol. 9, no. 1, pp. 62–66, 1979.
[28] I. C. Moreira, I. Amaral, I. Domingues, A. Cardoso, M. J. Cardoso, and J. S. Cardoso,“Inbreast: Toward a full­field digital mammographic database,”Academic radiology,vol. 19, no. 2, pp. 236–248, 2012.
[29] T.Y.Chang,“Chen­hsin hospital mammography database,”Department of Medical imag­ing, 2019

電子全文 電子全文(網際網路公開日期:20260814)
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊