臺灣博碩士論文加值系統

近年來，在電腦視覺識別的研究中，大多數模型都基於充足的數據量，並且利用大量影像進行訓練，得到良好的模型表現。然而在醫學領域中，醫學影像的資料量往往遠少於其他應用領域。因此，在醫學影像分析中若使用一般電腦視覺識別模型，其效能會相對較差。在本研究中，我們提出了新的影像分析方法，在醫學領域資料有限的條件下，有效提升模型的效能。我們所提出的方法中，首先會進行病徵偵測，接著整合病徵ROI與原始影像，最後進行病症分類。在本研究中，也利用傳統影像處理演算法，將影像進行前處理，提取不同的影像特徵，提升病徵偵測與病症分類之效能。本研究使⽤了來⾃台灣中部某醫學中⼼所提供的腹部 X 光影像進行實驗，其中包含健康樣本與患有僵直性脊椎炎之樣本。在病徵ROI偵測的部分，使用Patch-Level分類器來將影像中含有病徵特徵的Patch偵測出來。結合傳統影像處理演算法進行特徵提取，病徵偵測模型的AUC能達到0.925。接著，利用此病徵偵測模型來產生病徵特徵熱圖。病症分類效能方面，在結合病徵特徵熱圖後，病症分類模型的accuracy能達到82%，相較於未整合病徵偵測之模型效能，獲得大幅度的提升。

In recent years, most of the models in computer vision recognition were based on sufficient data volume and trained over a large number of images to obtain good performance. However, in the medical field, the volume of images is often much smaller than other applications. Therefore, if a general computer visual recognition model is used in medical image analysis, its performance will be relatively poor. In our research, we propose a method of image analysis to effectively improve the performance of the model with the limited data of the medical image. In our method, the symptom detection is first performed, and then the symptom feature maps is integrated with the original image for disease classification. Moreover, traditional image processing algorithms were also used to extract different image features. For experimental evaluations, we used an abdominal X-ray dataset provided by a medical center in the middle of Taiwan, which included normal samples and ankylosing spondylitis samples. In symptom detection, we obtained 0.925 AUC on the symptom detection model. After integrating with symptom detection, we obtain a great improvement(82% versus 73%) on disease classification.

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Research Aims and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Related Work 4
2.1 Image Classification with Deep Learning . . . . . . . . . . . . . . . . . . . 4
2.2 Object Detection with Deep Learning . . . . . . . . . . . . . . . . . . . . . 6
2.3 Medical Image Analysis with Deep Learning . . . . . . . . . . . . . . . . . 7
3 Proposed Method: Symptom-Aided Disease Recognition Framework 9
3.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Proposed Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.1 Texture Feature Extraction . . . . . . . . . . . . . . . . . . . . . . 10
3.2.2 Symptom Feature Extraction . . . . . . . . . . . . . . . . . . . . . 16
3.2.3 Disease Classification . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4 Experimental Evaluation 21
4.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Experimental Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.1 Texture Feature Extraction . . . . . . . . . . . . . . . . . . . . . . 25
4.2.2 Symptom Feature Extraction . . . . . . . . . . . . . . . . . . . . . 27
4.2.3 Disease Classification . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Evaluation Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
iii
4.4 Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5.1 Effect on Texture Feature Extraction . . . . . . . . . . . . . . . . . 33
4.5.2 Effect on Symptom Feature Extraction . . . . . . . . . . . . . . . . 36
4.5.3 Performance of Automatic Framework . . . . . . . . . . . . . . . . 37
5 Conclusion and Future Works 42
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
References 44
Appendices 50

[1] RongshengZhu,RongZhang,andDixiuXue.“Lesiondetectionofendoscopyimages based on convolutional neural network features”. In: Image and Signal Processing (CISP), 2015 8th International Congress on. IEEE. 2015, pp. 372–376. [2] Nima Tajbakhsh, Suryakanth R Gurudu, and Jianming Liang. “Automatic polyp detectionincolonoscopyvideosusinganensembleofconvolutionalneuralnetworks”. In:BiomedicalImaging(ISBI),2015IEEE12thInternationalSymposiumon.IEEE. 2015, pp. 79–83. [3] SpirosVGeorgakopoulos,DimitrisKIakovidis,MichaelVasilakakis,VassilisPPlagianakos, and Anastasios Koulaouzidis. “Weakly-supervised convolutional learning for detection of inflammatory gastrointestinal lesions”. In: Imaging Systems and Techniques (IST), 2016 IEEE International Conference on. IEEE. 2016, pp. 510– 514. [4] Yaniv Bar, Idit Diamant, Lior Wolf, and Hayit Greenspan. “Deep learning with non-medical training used for chest pathology identification”. In: Medical Imaging 2015: Computer-Aided Diagnosis. Vol. 9414. International Society for Optics and Photonics. 2015, p. 94140V. [5] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. “Imagenet classification with deep convolutional neural networks”. In: Advances in neural information processing systems. 2012, pp. 1097–1105. [6] Karen Simonyan and Andrew Zisserman. “Very deep convolutional networks for large-scale image recognition”. In: arXiv preprint arXiv:1409.1556 (2014). [7] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich, et al. “Going deeper with convolutions”. In: Cvpr. 2015.
[8] Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. “Gradient-based learningappliedtodocumentrecognition”.In:ProceedingsoftheIEEE 86.11(1998), pp. 2278–2324. [9] David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning internal representations by error propagation. Tech. rep. California Univ San Diego La Jolla Inst for Cognitive Science, 1985. [10] MinLin,QiangChen,andShuichengYan.“Networkinnetwork”.In:arXivpreprint arXiv:1312.4400 (2013). [11] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. “Deep residual learning for image recognition”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 770–778. [12] Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. “Densely Connected Convolutional Networks.” In: CVPR. Vol. 1. 2. 2017, p. 3. [13] Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. “Aggregated residual transformations for deep neural networks”. In: Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on. IEEE. 2017, pp. 5987– 5995. [14] FrançoisChollet.“Xception:Deeplearningwithdepthwiseseparableconvolutions”. In: arXiv preprint (2017), pp. 1610–02357. [15] Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. “Region-based convolutional networks for accurate object detection and segmentation”. In: IEEE transactions on pattern analysis and machine intelligence 38.1 (2016), pp. 142–158. [16] Pierre Sermanet, David Eigen, Xiang Zhang, Michaël Mathieu, Rob Fergus, and Yann LeCun. “Overfeat: Integrated recognition, localization and detection using convolutional networks”. In: arXiv preprint arXiv:1312.6229 (2013). [17] Dumitru Erhan, Christian Szegedy, Alexander Toshev, and Dragomir Anguelov. “Scalableobjectdetectionusingdeepneuralnetworks”.In:Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2014, pp. 2147–2154.
[18] JasperRRUijlings,KoenEAVanDeSande,TheoGevers,andArnoldWMSmeulders.“Selectivesearchforobjectrecognition”.In:International journal of computer vision 104.2 (2013), pp. 154–171. [19] CorinnaCortesandVladimirVapnik.“Support-vectornetworks”.In:Machinelearning 20.3 (1995), pp. 273–297. [20] KaimingHe,XiangyuZhang,ShaoqingRen,andJianSun.“Spatialpyramidpooling in deep convolutional networks for visual recognition”. In: european conference on computer vision. Springer. 2014, pp. 346–361. [21] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. “Faster r-cnn: Towards real-time object detection with region proposal networks”. In: Advances in neural information processing systems. 2015, pp. 91–99. [22] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. “You only look once: Unified, real-time object detection”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 779–788. [23] WeiLiu,DragomirAnguelov,DumitruErhan,ChristianSzegedy,ScottReed,ChengYangFu,andAlexanderCBerg.“Ssd:Singleshotmultiboxdetector”.In:European conference on computer vision. Springer. 2016, pp. 21–37. [24] Jifeng Dai, Yi Li, Kaiming He, and Jian Sun. “R-fcn: Object detection via regionbased fully convolutional networks”. In: Advances in neural information processing systems. 2016, pp. 379–387. [25] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. “Mask r-cnn”. In: Computer Vision (ICCV), 2017 IEEE International Conference on. IEEE. 2017, pp. 2980–2988. [26] Joseph Redmon and Ali Farhadi. “YOLO9000: better, faster, stronger”. In: arXiv preprint (2017). [27] Hyeonwoo Noh, Seunghoon Hong, and Bohyung Han. “Learning deconvolution network for semantic segmentation”. In: Proceedings of the IEEE international conference on computer vision. 2015, pp. 1520–1528.
[28] JonathanLong,EvanShelhamer,andTrevorDarrell.“Fullyconvolutionalnetworks for semantic segmentation”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2015, pp. 3431–3440. [29] Olaf Ronneberger, Philipp Fischer, and Thomas Brox. “U-net: Convolutional networksforbiomedical imagesegmentation”.In: International Conference on Medical image computing and computer-assisted intervention. Springer. 2015, pp. 234–241. [30] VijayBadrinarayanan,AlexKendall,andRobertoCipolla.“Segnet:Adeepconvolutionalencoder-decoderarchitectureforimagesegmentation”.In:IEEE transactions on pattern analysis and machine intelligence 39.12 (2017), pp. 2481–2495. [31] Qing Li, Weidong Cai, Xiaogang Wang, Yun Zhou, David Dagan Feng, and Mei Chen.“Medicalimageclassificationwithconvolutionalneuralnetwork”.In:Control Automation Robotics & Vision (ICARCV), 2014 13th International Conference on. IEEE. 2014, pp. 844–848. [32] Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mohammadhadi Bagheri, and Ronald M Summers. “Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases”. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE. 2017, pp. 3462–3471. [33] Alimpyeva Anastasia. “Deep learning for abnormality detection in chest x-ray images”. In: (2018). [34] Andre Esteva, Brett Kuprel, Roberto A Novoa, Justin Ko, Susan M Swetter, Helen MBlau,andSebastianThrun.“Dermatologist-levelclassificationofskincancerwith deep neural networks”. In: Nature 542.7639 (2017), p. 115. [35] Pranav Rajpurkar et al. “Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning”. In: arXiv preprint arXiv:1711.05225 (2017). [36] HolgerRRoth,ChristopherTLee,Hoo-ChangShin,AriSeff,LaurenKim,Jianhua Yao, Le Lu, and Ronald M Summers. “Anatomy-specific classification of medical images using deep convolutional nets”. In: arXiv preprint arXiv:1504.04003 (2015).
[37] Holger R Roth, Jianhua Yao, Le Lu, James Stieger, Joseph E Burns, and Ronald M Summers. “Detection of sclerotic spine metastases via random aggregation of deep convolutional neural network classifications”. In: Recent advances in computational methods and clinical applications for spine imaging. Springer, 2015, pp. 3–12. [38] Gabriel García, Jhair Gallardo, Antoni Mauricio, Jorge López, and Christian Del Carpio. “Detection of Diabetic Retinopathy Based on a Convolutional Neural NetworkUsingRetinalFundusImages”.In:InternationalConferenceonArtificialNeural Networks. Springer. 2017, pp. 635–642. [39] MarcoAlbanandTannerGilligan.“Automateddetectionofdiabeticretinopathyusing fluorescein angiography photographs”. In: Report of standford education (2016). [40] Harry Pratt, Frans Coenen, Deborah M Broadbent, Simon P Harding, and Yalin Zheng.“Convolutionalneuralnetworksfordiabeticretinopathy”.In:ProcediaComputer Science 90 (2016), pp. 200–205. [41] MojtabaAkbari,MajidMohrekesh,EbrahimNasr-Esfahani,SMSoroushmehr,Nader Karimi,ShadrokhSamavi,andKayvanNajarian.“PolypSegmentationinColonoscopy Images Using Fully Convolutional Network”. In: arXiv preprint arXiv:1802.00368 (2018). [42] Eduardo Ribeiro, Andreas Uhl, and Michael Häfner. “Colonic polyp classification withconvolutionalneuralnetworks”.In:Computer-BasedMedicalSystems(CBMS), 2016 IEEE 29th International Symposium on. IEEE. 2016, pp. 253–258. [43] Lorien Y Pratt. “Discriminability-based transfer between neural networks”. In: Advances in neural information processing systems. 1993, pp. 204–211. [44] Timo Ojala, Matti Pietikainen, and David Harwood. “Performance evaluation of texture measures with classification based on Kullback discrimination of distributions”.In:Pattern Recognition, 1994. Vol. 1-Conference A: Computer Vision & Image Processing., Proceedings of the 12th IAPR International Conference on. Vol. 1. IEEE. 1994, pp. 582–585. [45] Timo Ojala, Matti Pietikäinen, and David Harwood. “A comparative study of texturemeasureswithclassificationbasedonfeatureddistributions”.In:Patternrecognition 29.1 (1996), pp. 51–59.
[46] Ruikai Zhang, Yali Zheng, Tony Wing Chung Mak, Ruoxi Yu, Sunny H Wong, James YW Lau, and Carmen CY Poon. “Automatic detection and classification of colorectal polyps by transferring low-level CNN features from nonmedical domain”. In: IEEE journal of biomedical and health informatics 21.1 (2017), pp. 41–47. [47] Yan Xu, Zhipeng Jia, Yuqing Ai, Fang Zhang, Maode Lai, I Eric, and Chao Chang. “Deep convolutional activation features for large scale brain tumor histopathology image classification and segmentation”. In: Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on. IEEE. 2015, pp. 947–951. [48] Maxime Oquab, Léon Bottou, Ivan Laptev, and Josef Sivic. “Is object localization for free?-weakly-supervised learning with convolutional neural networks”. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015, pp. 685–694. [49] Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. “Learning deep features for discriminative localization”. In: Computer Vision and Pattern Recognition (CVPR), 2016 IEEE Conference on. IEEE. 2016, pp. 2921– 2929. [50] JohnCanny.“Acomputationalapproachtoedgedetection”.In:IEEETransactions on pattern analysis and machine intelligence 6 (1986), pp. 679–698. [51] JamesMacQueenetal.“Somemethodsforclassificationandanalysisofmultivariate observations”. In: Proceedings of the fifth Berkeley symposium on mathematical statistics and probability. Vol. 1. 14. Oakland, CA, USA. 1967, pp. 281–297. [52] Jie Hu, Li Shen, and Gang Sun. “Squeeze-and-excitation networks”. In: arXiv preprint arXiv:1709.01507 7 (2017).