對肺結節進行分類是早期肺癌診斷的關鍵步驟，也是後續治療方案的依據；故，肺結節分類具臨床重要意義。由於肺癌的治療選擇取決於肺結節的類型，但因肺結節的型態多變，各類別在電腦斷層影像上的同質性相當高，要精確分類並不容易。為了協助醫師在大量健檢資料中有效率的精準判斷肺部結節的惡性程度，本論文針對肺部亞實質結節分類提出高效的深度學習模型；改良模型，以所提改良的YOLOv4模型，包括(1)結合不同尺度的影像以增加分類所需資訊。(2)在原YOLOv4架構C2層加上橫向連接，以保留底層細部的資訊。此外，亦加入注意力模組，使模組更聚焦關注於重要特徵，以提升整體模型的準確度。模型訓練則使用成大醫院所提供具惡信度資訊的健檢資料及具病理資訊的資料，使分類模型更能具臨床使用價值，有效的協助醫師判斷後續治療方向。實驗數據顯示，本研究提出的改良YOLOv4模型在Lung-RADS (Lung imaging Reporting and Data System)結節惡性度分類上可達到準確率68.1%、F1-score 69.7%，在病理分類上，針對肺腺癌的分類達到準確率77.8%、F1-score 78.6%。即便是混合非肺腺癌的類別也能在分類上達到準確率68.8%，F1-score 69.2%。經實驗證明，我們的成果優於現有做病理分類之結果。

The classification of lung nodules is a key step in the diagnosis of early lung cancer and the basis for subsequent treatment plans, therefore, the classification of lung nodules has clinical significance. Since the choice of treatment for lung cancer depends on the type of lung nodule, but because the homogeneity of each category on the computer tomography is quite high, and it is not easy to accurately classify it. In order to assist doctors to efficiently and accurately decide the malignant degree of lung nodules from a large amount of health examination data, this paper proposes a deep learning model for the classification of subsolid nodules using a modified YOLOv4 model. Our modifications include: (1) Using images of different view areas to increase the information for classification; (2) A new horizontal connection to the C2 layer of the original YOLOv4 architecture to retain detailed information from the bottom layer. In addition, an attention module is added for focusing on important features. The model training uses the health examination data with malignancy score and the data with pathology report by National Cheng Kung University Hospital provided so that the classification model can be more clinically useful and effectively assist doctors in determining the direction of follow-up treatment. The experimental results showed an overall accuracy of 68.1% with an F1-score of 69.7% in the Lung-RADS classification of malignancy. The classification of lung adenocarcinoma achieved an accuracy of 77.8% with an F1-score of 78.6%. Even mixed non-lung adenocarcinoma achieved an accuracy of 68.8 % with an F1-score of 69.2%. Our performance is better than the state-of-the-art method on pathological classification.

[1] 衛福部, "108死因統計".
[2] 台灣肺癌協會, "低劑量胸部電腦斷層肺癌篩檢手冊(醫療版)," 2020. [Online]. Available: https://www.tlcs.org.tw/secretariatn_notice_article.php?the_no=czozOiIxODMiOw==.
[3] Sørensen L, Shaker SB, de Bruijne M, "Quantitative Analysis of Pulmonary Emphysema Using Local Binary Patterns," IEEE Trans Med Imaging, 29:559-69, 2010.
[4] Gangeh MJ, Sorensen L, Shaker SB, "A Texton-Based Approach for the Classification of Lung Parenchyma in CT Images," Med Image Comput Comput Assist Interv ;13:595-602, 2010.
[5] Anthimopoulos M, Christodoulidis S, Christe A, "Classification of Interstitial Lung Disease Patterns Using Local DCT Features and Random Forest," Conf Proc IEEE Eng Med Biol Soc:6040-3, 2014.
[6] Q. Song, L. Zhao, X. Luo and X. Dou, "Using Deep Learning for Classification of Lung Nodules on Computed Tomography Images," Hindawi, Journal of Healthcare Engineering, (7), 2017.
[7] Gao Huang, Zhuang Liu, Laurens van der Maaten, Kilian Q. Weinberger, "Densely Connected Convolutional Networks," arXiv:1608.06993 [cs.CV], 2017.
[8] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, "Deep Residual Learning for Image Recognition," arXiv:1512.03385 [cs.CV], 2015.
[9] Mian Muhammad Naeem Abid, Tehseen Zia, Mubeen Ghafoor, David Windridge, "Multi-view Convolutional Recurrent Neural Networks for Lung Cancer Nodule Identification," Neurocomputing , vol.453:299-311, 2021.
[10] Xiaoguang Tu, Mei Xie, Jingjing Gao, Zheng Ma, Daiqiang Chen, Qingfeng Wang, Samuel G. Finlayson, Yangming Ou, Jie-Zhi Cheng, “Automatic Categorization and Scoring of Solid, Part-Solid and Non-Solid Pulmonary Nodules in CT Images with Convolutional Neural Network,” Sci Rep 7:8533, 2017.
[11] Duo Wang, Tao Zhang, Ming Li, Raphael Bueno, Jagadeesan Jayender, "3D deep learning based classification of pulmonary ground glass opacity nodules with automatic segmentation," Computerized Medical Imaging and Graphics, 2021.
[12] Xiang Wang, Qingchu Li, Jiali Cai, Wei Wan, Peng Xu, Yiqian Zhang, Qu Fang, Chicheng Fu, Li Fan, Yi Xiao, Shiyuan Liu, “Predicting the invasiveness of lung adenocarcinomas appearing as ground-glass nodule on CT scan using multi-task learning and deep radiomics,” Transl Lung Cancer Res. 9(4):1397–1406, 2020.
[13] Travis, W.D., Brambilla, E., Noguchi, M., Nicholson, A.G., Geisinger, K.R., Yatabe, Y., et al., "International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society International Multidisciplinary Classification of Lung Adenocarcinoma," Journal of Thoracic Oncology 6(8):244-285, vol. 6, no. 2, 2011.
[14] Jun Wang, Xiaorong Chen, Hongbing Lu, Lichi Zhang, Jianfeng Pa, Yong Bao, Jiner Su, Dahong Qian, "Feature-shared adaptive-boost deep learning for invasiveness classification of pulmonary subsolid nodules in CT images," Medical Physics, 47(4):1738-1749, 2020.
[15] Alexey Bochkovskiy, Chien-Yao Wang, Hong-Yuan Mark Liao, "YOLOv4: Optimal Speed and Accuracy of Object Detection," Computer Vision and Pattern Recognition (CVPR), 2020.
[16] Yaojun Dai, Shiju Yan, Bin Zheng, Chengli Song, "Incorporating automatically learned pulmonary nodule attributes into a convolutional neural network to improve accuracy of benign-malignant nodule classification," Phys Med Biol. 63(24):245004, 2018.
[17] Chien-Yao Wang, Hong-Yuan Mark Liao, I-Hau Yeh, Yueh-Hua Wu, Ping-Yang Chen, Jun-Wei Hsieh, "CSPNet: A New Backbone that can Enhance Learning Capability of CNN," arXiv:1911.11929 [cs.CV], 2019.
[18] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, "Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition," arXiv:1406.4729 [cs.CV], 2014.
[19] Shu Liu, Lu Qi, Haifang Qin, Jianping Shi, Jiaya Jia, "Path Aggregation Network for Instance Segmentation," arXiv:1803.01534 [cs.CV], 2018.
[20] Joseph Redmon, Ali Farhadi, "YOLOv3: An Incremental Improvement," arXiv:1804.02767 [cs.CV], 2018.
[21] Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon, "CBAM: Convolutional Block Attention Module," Computer Vision and Pattern Recognition(CVPR), 2018.
[22] S. Raschka, "Model Evaluation, Model Selection, and Algorithm Selection in Machine Learning," arXiv preprint arXiv:1811.12808, 2018.
[23] Wei Zhao, Jiancheng Yang, Yingli Sun, Cheng Li, Weilan Wu, Liang Jin, Zhiming Yang, Bingbing Ni, Pan Gao, Peijun Wang, Yanqing Hua, Ming Li, "3D Deep Learning from CT Scans Predicts Tumor Invasiveness of Subcentimeter Pulmonary Adenocarcinomas," Cancer Res. 78(24):6881-6889, 2018.
[24] Mingxing Tan, Ruoming Pang, Quoc V. Le, "EfficientDet: Scalable and Efficient Object Detection," Computer Vision and Pattern Recognition (CVPR), 2020.
[25] Shengping Wang, Rui Wang, Shengjian Zhang, Ruimin Li, Yi Fu, Xiangjie Sun, Yuan Li, Xing Sun, Xinyang Jiang, Xiaowei Guo, Xuan Zhou, Jia Chang, Weijun Peng, "3D convolutional neural network for differentiating pre-invasive lesions from invasive adenocarcinomas appearing as ground-glass nodules with diameters ≤3 cm using HRCT," QIMS, 2018.
[26] Michael C.Lee, LillaBoroczky, Kivilcim Sungur-Stasi, Aaron D.Cann, Alain C.Borczuk, Steven M.Kawut, Charles A.Powell, "Computer-aided diagnosis of pulmonary nodules using a two-step approach for feature selection and classifier ensemble construction," Artificial Intelligence in Medicine, 50 (1), 43-53, 2010.
[27] Aydın Kaya, Ahmet Burak Can, "A weighted rule based method for predicting malignancy of pulmonary nodules by nodule characteristics," Journal of Biomedical Informatics Volume 56, 69-79, 2015.
[28] Buty M., Xu Z., Gao M., Bagci U., Wu A., Mollura D.J., "Characterization of Lung Nodule Malignancy Using Hybrid Shape and Appearance Features," Medical Image Computing and Computer-Assisted Intervention ( MICCAI) Lecture Notes in Computer Science, vol 9900, 2016.
[29] Sarfaraz Hussein, Robert Gillies, Kunlin Cao, Qi Song, Ulas Bagci, "TumorNet: Lung Nodule Characterization Using Multi-View Convolutional Neural Network with Gaussian Process," arXiv:1703.00645 [cs.CV], 2017.
[30] Xinzhuo Zhao1 · Liyao Liu, Shouliang Qi, Yueyang Teng, Jianhua Li, Wei Qian, "Agile convolutional neural network for pulmonary nodule classification using CT images," Computer Assisted Radiology and Surgery (CARS), 2018.
[31] Hong Liu, Haichao Cao, Enmin Song, Guangzhi Ma, Xiangyang Xu, Renchao Jin, Chuhua Liu, Chih-Cheng Hung, "Multi-model Ensemble Learning Architecture Based on 3D CNN for Lung Nodule Malignancy Suspiciousness Classification," Society for Imaging Informatics in Medicine, 2020.