根據全球衛生組織於2020年的報告顯示，肺結核是全球十大死亡原因之一。2019年時有超過一千萬新增病例，120萬例死亡病例，大部分的死亡病例發生在開發中國家，其中醫療照顧與檢測能量的不足是根本原因。因此，一個快速，準確，並且能夠負擔的起的檢測方法是必要的。
目前主要診斷結核病的方式有兩種：1) 醫師或醫檢師使用胸部X光進行診斷；2) 結核菌培養測試，通常需要4到6個禮拜。由於胸部X光診斷所需較少，因此在早期結核病判斷時常被使用。但是結核病在胸部X光上的特徵並不固定，同一個病人也有可能有多種表現形式或同時感染其他疾病，經常發生不同觀察者給出不同診斷的情況。而早期結核病在胸部X光影像上也很難辨認。本論文針對以上問題設計一套胸部X光結核病協助偵測系統，以輔助醫師與醫檢師進行更準確的判讀。
本論文提出一個以卷積類神經網路為基礎的結核病分類與檢測系統，用以輔助醫師與放射師判讀胸部X光影像。系統在讀入一張胸部X光影像後會輸出兩個輸出：1) 此胸部X光影像含有結核病的機率；2) 肺結核病灶的可能位置。大部分用於分類的類神經網路僅輸出一個機率用來表示含有肺結核的機率，但是如同黑箱(Black Box)一樣只有輸入輸出關係的系統對於醫師與放射師來說是意義不大的，我們的系統額外增加了一條分支輸出病灶位置，提供額外的資訊來輔助判斷。
本系統可以分為兩個部分，一個部分是傳統的分類分支，另一個是區域檢測分支，兩個分支是同時訓練的。首先將已知結果的胸部X光影像輸入系統，透過深度學習的方式學習含有肺結核與不含有肺結核的影像特徵，這些特徵會先由分類分支學習，接著將影像輸入檢測分支學習，為了鼓勵網路學習含有肺結核的X光影像中肺結核的位置，並讓網路結合兩個分支的資訊，我們特別設計了一個損失函數。
此系統主要使用Python以及Pytorch框架來實作。我們的方法能在分類準確度上達到99%，檢測方面則能夠達到68%的召回率與59%的精確率。這些結果表示我們的系統可以提供醫師與放射師一個準確的分類參考，以及一個可靠的胸部X光結核病病灶檢測輔助判讀。一般來說訓練與使用深度學習的卷積類神經網路需要標準以上的電腦輔以大量時間，我們的系統設計較為精簡，能夠在一般的電腦上順利運行，並提供一個快速且精確的檢測輔助。

According to World Health Organization's report in 2020, tuberculosis (TB) is one of the top ten causes of death worldwide. There are more than 10 million people who fell ill with TB and 1.2 million TB deaths in 2019. Most of the deaths are from developing countries, insufficient health care and diagnosis are the fundamental reasons. Therefore, a fast, accurate, and relativity cheap diagnosis is needed.
There are two major methods for tuberculosis diagnosing: 1) The radiologist uses chest X-ray (CXR) images for diagnosis; 2) The patient takes a tuberculosis culture test. But the test takes about 4-6 weeks to get results. CXR diagnosing is preferred in the early stage due to the required time. However, symptoms vary substantially between patients. Radiologists sometimes provides contrastive diagnoses on one CXR image. Additional, early-stage TB infections are difficult to identify. Our target is to help radiologists and doctors to overcome these problems while making decisions.
We proposed a tuberculosis classification and detection system utilizing a convolutional neural network (CNN) to help radiologists to determine tuberculosis in chest X-rays. The system takes chest X-rays as input, then produces two outputs: 1) Confident of TB infection; 2) Suspicious regions of TB infection. Most CNN models only output a single probability as output. However, for radiologists, one confident score is usually useless for diagnosis. Our system has an additional output branch to produce extra information.
The system can be divided into two parts. One is the traditional classification branch, and the other is the region detection branch. Both parts are trained simultaneously. First, chest X-ray images with labels are fed into our system, so that our system can learn the characteristic of TB and non-TB images. When the classification branch is done, chest X-ray images are then fed into the region detection branch. To encourage the consistency of the two parts, we thus design a loss function.
The system is developed using Python and Pytorch framework. The classification performance reaches 99% accuracy. The detection performance reaches 68% on recall and 59% on precision. These results show our proposed method can provide accurate classifications and relatively reliable detections of tuberculosis in chest X-rays images. Deep learning models usually need decent computing resources and are time-consuming. Most countries that suffer from tuberculosis may not have enough hardware to run a regular deep learning model. Our method is designed to be as simple as possible so that these countries can still run the system on an ordinary hardware.

Reference
[1] "Global tuberculosis report 2020," World Health Organization (WHO), 2020.
[2] S. Jaeger, A. Karargyris, S. Antani and G. Thoma, "Detecting tuberculosis in radiographs using combined lung masks," Proc. Int. IEEE Eng. Med. Biol. Soc., pp. 4978-4981, 2012.
[3] I. Gabriella, K. S. A. and S. A. W., "Early Detection of Tuberculosis using Chest X-Ray (CXR) with Computer-Aided Diagnosis," 2nd International Conference on Biomedical Engineering (IBIOMED), pp. 76-79, 2018.
[4] L. G. C. Evalgelista and E. B. Guedes, "Computer-Aided Tuberculosis Detection from Chest X-Ray Images with Convolutional Neural Networks," XV Encontro Nacional de Inteligência Artificial e Computacional, pp. 518-527, 2018.
[5] E. J. Hwang, S. Park and K.-N. Jin, "Development and Validation of a Deep Learning–based Automatic Detection Algorithm for Active Pulmonary Tuberculosis on Chest Radiographs," Clinical and Technical Considerations. Clin. Infect. Dis., 2018.
[6] C.-C. Chuang, Tuberculosis Detection Using Convolutional Neural Network in Chest Radiographs, National Cheng Kung University, 2019.
[7] Y. LeCun, L. Bottou, Y. Bengio and P. Haffner, "Gradient-based learning applied to document recognition.," Proc. IEEE 86, p. 2278–2324, 1998.
[8] I. Sutskever, A. Krizhevsky and G. E. Hinton, "ImageNet Classification with Deep Convolutional," International Conference on Neural Information Processing Systems (NIPS), pp. 1097-1105, 2012.
[9] Deng Jia; Dong Wei; Socher Richard; Li Li-Jia ; Li Kai ; Fei-Fei Li, "Imagenet: A large-scale hierarchical image database," CVPR, 2009.
[10] K. Simonyan and A. Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," arXiv preprint arXiv:1409.1556, 2015.
[11] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke and A. Rabinovich, "Going deeper with convolutions.," IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1-9, 2015.
[12] K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. pp770-778, 2016.
[13] G. Huang, Z. Liu, L. v. d. Maaten and K. Q. Weinberger, "Densely Connected Convolutional Networks," The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4700-4708, 2017.
[14] Redmon, Joseph;Farhadi, Ali, "YOLOv3: An Incremental Improvement," arXiv, 2018.
[15] Wei Liu;Dragomir Anguelov;Dumitru Erhan;Christian Szegedy;Scott Reed;Cheng-Yang Fu;Alexander C. Berg, "SSD: Single Shot MultiBox Detector," 2016 European Conference on Computer Vision (ECCV), 2016.
[16] Lin, Tsung-Yi ; Goyal, Priya ; Girshick, Ross ; He, Kaiming ; Dollár, Piotr, "Focal Loss for Dense Object Detection," 2017 IEEE International Conference on Computer Vision (ICCV), pp. 2999-3007, 2017.
[17] R. Girshick, J. Donahue, T. Darrell and J. Malik, "Rich feature hierarchies for accurate object detection and semantic segmentation," in IEEE Conference on Computer Vision and Pattern Recognition, pp. 580-587, 2014.
[18] S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," Pattern Analysis and Machine Intelligence, pp. 1137 - 1149, 2017.
[19] Kaiming He; Georgia Gkioxari; Piotr Dollár; Ross Girshick, "Mask R-CNN," 2017 IEEE International Conference on Computer Vision (ICCV), pp. 2980-2988, 2017.
[20] Han, Yan ; Chen, Chongyan ; Tewfik, Ahmed ; Ding, Ying ; Peng, Yifan, "Pneumonia Detection On Chest X-Ray Using Radiomic Features And Contrastive Learning," 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI), pp. 247-251, 2021.
[21] Robert J Gillies; Paul E Kinahan; Hedvig Hricak, "Radiomics: images are more than pictures, they are data," Radiology, vol. 278, no. 2, pp. 563-577, 2016.
[22] Sergey Ioffe; Christian Szegedy, "Batch normalization: accelerating deep network training by reducing internal covariate shift," ICML, 2015.
[23] J. P. J.-Y. L. I. S. K. Sanghyun Woo, "CBAM: Convolutional Block Attention Module," in in The European Conference on Computer Vision (ECCV), pp. 3-19, 2018.
[24] Jie Hu; Li Shen; Samuel Albanie; Gang Sun; Enhua Wu, "Squeeze-and-Excitation Networks," CVPR, 2018.
[25] Xingang Pan; Ping Luo; Jianping Shi; Xiaoou Tang, "Two at Once: Enhancing Learning and Generalization Capacities via IBN-Net," ECCV, 2018.
[26] Dmitry Ulyanov; Andrea Vedaldi; Victor Lempitsky, "Improved Texture Networks: Maximizing Quality and Diversity in Feed-forward Stylization and Texture Synthesis," CVPR, 2017.
[27] Robert M. Haralick; K. Shanmugam; Its'Hak Dinstein, "Textural Features for Image Classification," IEEE Transactions on Systems, Man, and Cybernetics, Vols. SMC-3, no. 6, pp. 610-621, 1973.
[28] van Griethuysen, J. J. M., Fedorov, A., Parmar, C., Hosny, A., Aucoin, N., Narayan, V., Beets-Tan, R. G. H., Fillon-Robin, J. C., Pieper, S., Aerts, H. J. W. L., "Computational Radiomics System to Decode the Radiographic Phenotype," Cancer Research, vol. 77, no. 21, pp. 104-107, 2017.
[29] M. M.Galloway, "Texture analysis using gray level run lengths," Computer Graphics and Image Processing, vol. 4, no. 2, pp. 172-179, 6 1975.
[30] T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie, "Feature pyramid networks for object detection.," IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2117-2125, 2017.
[31] T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollr and C. L. Zitnick, "Microsoft coco: Common objects in context," in ECCV, 2014.
[32] Bernard A. Galler;Michael J. Fisher, "An improved equivalence algorithm," Communications of the ACM, vol. 7, pp. 301-303, 1964.
[33] R. Haralick, "Statistical and structural approaches to texture," Proceedings of the IEEE, pp. 786-804, 1979.
[34] S. Liu, L. Qi, H. Qin, J. Shi and J. Jia, "Path Aggregation Network for Instance Segmentation," in The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp.8759-8768, 2017.
[35] Chen Yue ; Bai Yalong ; Zhang Wei; Mei Tao, "Destruction and Construction Learning for Fine-Grained Image Recognition," in The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019.
[36] "RSNA International COVID-19 Open Radiology Database," [Online]. Available: https://www.rsna.org/covid-19/covid-19-ricord. [Accessed 07 2021].
[37] "covid-chestxray-dataset," [Online]. Available: https://github.com/ieee8023/covid-chestxray-dataset. [Accessed 7 2021].
[38] Ramprasaath R. Selvaraju; Michael Cogswell; Abhishek Das; Ramakrishna Vedantam; Devi Parikh; Dhruv Batra, "Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization," arXiv, 2019.