甲狀腺癌是一種常見的內分泌惡性腫瘤, 佔所有惡性腫瘤的 1-4%。其中, 乳突癌(Papillary Thyroid Carcinoma, PTC) 是最常見的類型, 佔甲狀腺癌病例的 90 - 95% 以上。在過去的 25 年中, 台灣甲狀腺癌的新發個案數量以接近 7 倍的速度增加。儘管甲狀腺癌的致死率相對較低, 但仍有 10% 至 15% 的病人會發展成高度惡性且致命的去分化癌病變和遠端轉移。由於甲狀腺癌的發生率不斷上升, 此癌症是一種越早發現、越早治療, 其治療成功率越高的癌症。因此, 開發細胞學診斷技術有助於早期發現甲狀腺癌避免後續癌變。近年來醫學影像科技與深度學習技術提升很多，有效運用可以改進甲狀腺癌的早期診斷準確度。本論文嘗試開發了一種運用人工智慧技術的輔助診斷系統。我們根據臨床醫生細膩診斷中所觀察的病灶特徵, 例如細胞核大小變化、核質比及核膜不規則等病灶特徵，設計網路模型及特徵提取方法。本論文共使用了 14 例的細胞檢體抹片 (1534 張病例圖像) 來建立 AI 模型訓練及驗證的資料, 並使用 3 例細胞檢體抹片 (127 張病例圖像) 作為模型的測試資料。它集成了用於目標檢測的 YOLOv7 模型、用於增強目標分類 Xception 模型, 及用於細胞團塊、核分割的DeepLabv3+ 模型, 並提出半監督式樣本標記方法, 快速擴增模型訓練資料, 提升模型預測精準度。透過以上深度學習模型及特徵提取方法可進一步分析甲狀腺細胞核的特徵變化。最後比較了四種常見的機器學習模型對於病灶特徵的分類表現。結果表明隨機森林分類器 (Random Forest Classifier) 表現最佳, 在測試數據中準確度高達 94.78%。此外, 系統中也引入了品質控制解決方案, 剔除掉品質不佳的病例圖像, 以保證臨床診斷的可靠性。此系統藉由目標檢測模型，可以有效的協助細胞醫檢師更快速的確認病灶區域。並且提供病灶區域的量化特徵及模型預測結果。提供醫檢師完成初步篩檢的效率與正確性，最終輔助醫師精準快篩潛在病患。

Thyroid cancer is a common endocrine malignancy, accounting for 1-4% of all malignant tumors. Papillary Thyroid Carcinoma (PTC) is the most common type, representing over 90 - 95% of thyroid cancer cases.In the past 25 years, the number of new thyroid cancer cases in Taiwan has increased close to 7 times. Although the mortality rate of thyroid cancer is relatively low, 10% to 15% of patients still develop highly malignant and fatal poorly differentiated carcinoma and distant metastasis. With the continuous increase in the incidence of thyroid cancer, early detection and treatment are crucial for higher treatment success rates. Therefore, developing cytological diagnostic techniques helps detect thyroid cancer and prevent subsequent carcinogenesis. In recent years, advancements in medical imaging technology and deep learning techniques have greatly improved the accuracy of early diagnosis of thyroid cancer. In this thesis, an AI-assisted diagnostic system was developed. The system was designed based on the fine observations made by clinical doctors during diagnosis, such as changes in nuclear size, nuclear-cytoplasmic ratio, and irregular nuclear membranes, Designing the network
model and feature extraction method. This paper used 14 cell samples (1,534 images) to establish the training and validation data for the AI model and 3 cell samples (127 images) as test data for the model. The system integrates the YOLOv7 model for object detection, the Xception model for enhanced target classification, and the DeepLabv3+ model for cell clump and nucleus segmentation.Additionally, we proposed a semi-supervised sample labeling method to augment the model’s training data and improve the accuracy of model predictions.
Further, we analyzed thyroid cell nucleus characteristics through deep-learning models and feature extraction methods. Finally, we compared four standard machine learning models regarding their classification performance on lesion features. The results show that the random forest classifier performs the best, with an accuracy of up to 94.78% on the test data. Furthermore, a quality control solution is implemented in the system to exclude poor-quality case images and ensure the reliability of clinical diagnosis. This system utilizes a target detection model to effectively assist cytotechnologists in quickly identifying lesion areas. It also provides quantified features of the lesion area and model prediction results, offering improved efficiency and accuracy in the initial screening process. Ultimately, it assists physicians in accurately and swiftly identifying potential patients through precise and rapid screening.

[1] W. Duan, L. Gao, J. Liu, C. Li, P. Jiang, L. Wang, H. Chen, X. Sun, D. Cao, B. Pang, R. Li, and S. Liu, “Computer-assisted fine-needle aspiration cytology of thyroid using
two-stage refined convolutional neural network,” Electronics, vol. 11, no. 24, 2022.
[2] M. Hirokawa, H. Niioka, A. Suzuki, M. Abe, Y. Arai, H. Nagahara, A. Miyauchi, and
T. Akamizu, “Application of deep learning as an ancillary diagnostic tool for thyroid
fna cytology,” Cancer Cytopathology, vol. n/a, no. n/a, 2022.
[3] N. T. Duc, Y.-M. Lee, J. H. Park, and B. Lee, “An ensemble deep learning for automatic prediction of papillary thyroid carcinoma using fine needle aspiration cytology,” Expert Systems with Applications, vol. 188, p. 115927, 2022.
[4] K. Jassal, A. Koohestani, A. Kiu, A. Strong, N. Ravintharan, M. Yeung, S. Grodski,
J. W. Serpell, and J. C. Lee, “Artificial intelligence for pre-operative diagnosis of malignant thyroid nodules based on sonographic features and cytology category,” World
Journal of Surgery, vol. 47, 2023.
[5] Q. Guan, Y. Wang, B. Ping, D. Li, J. Du, Y. Qin, H. Lu, X. Wan, and J. Xiang, “Deep
convolutional neural network vgg-16 model for differential diagnosing of papillary thyroid carcinomas in cytological images: a pilot study,” 2019.
[6] C.-Y. Wang, A. Bochkovskiy, and H.-Y. M. Liao, “Yolov7: Trainable bag-of-freebies
sets new state-of-the-art for real-time object detectors,” 2022.
[7] L.-C. Chen, Y. Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder-decoder with
atrous separable convolution for semantic image segmentation,” 2018.
[8] F. Chollet, “Xception: Deep learning with depthwise separable convolutions,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR),
July 2017.
[9] A. Prete, P. B. de Souza, S. Censi, M. Muzza, N. Nucci, and M. Sponziello, “Update
on fundamental mechanisms of thyroid cancer,” Frontiers in Endocrinology, vol. 11,
2020.
[10] J.-F. Hang, C.-Y. Hsu, and C.-R. Lai, “Thyroid fine-needle aspiration in taiwan: the history and current practice,” 2017.
[11] H.-P. Chan, L. M. Hadjiiski, and R. K. Samala, “Computer-aided diagnosis in the era of deep learning,” Medical Physics, vol. 47, no. 5, pp. e218–e227, 2020.
[12] M. Saha, R. Mukherjee, and C. Chakraborty, “Computer-aided diagnosis of breast cancer using cytological images: A systematic review,” Tissue and Cell, vol. 48, no. 5,
pp. 461–474, 2016.
[13] G. Pellegriti, F. Frasca, C. Regalbuto, S. Squatrito, and R. Vigneri, “Worldwide increasing incidence of thyroid cancer: update on epidemiology and risk factors,” 2013.
[14] E. S. Cibas and S. Z. Ali, “The 2017 bethesda system for reporting thyroid cytopathology,” 2017.
[15] B. R. Haugen, E. K. Alexander, K. C. Bible, G. M. Doherty, S. J. Mandel, Y. E. Nikiforov, F. Pacini, G. W. Randolph, A. M. Sawka, M. Schlumberger, et al., “2015 american thyroid association management guidelines for adult patients with thyroid nodules and differentiated thyroid cancer: the american thyroid association guidelines task force on thyroid nodules and differentiated thyroid cancer,” 2016.
[16] D. Fresilli, E. David, P. Pacini, G. Del Gaudio, V. Dolcetti, G. T. Lucarelli, N. Di Leo, M. I. Bellini, V. D039;Andrea, S. Sorrenti, D. Mascagni, M. Biffoni, C. Durante, G. Grani, G. De Vincentis, and V. Cantisani, “Thyroid nodule characterization: How to assess the malignancy risk. update of the literature,” Diagnostics, vol. 11, no. 8, 2021.
[17] P. Sanyal, T. Mukherjee, S. Barui, A. Das, and P. Gangopadhyay, “Artificial intelligence in cytopathology: A neural network to identify papillary carcinoma on thyroid fineneedle aspiration cytology smears,” Journal of Pathology Informatics, vol. 9, no. 1, p. 43, 2018.
[18] T. Chang, S. Kou, F. Chen, C. Chang, K. Liaw, Y. Chang-Chien, T. Huang, C. Lin, and S. How, “Cytopathology of thyroid cancer by fine needle aspiration with liu’s stain.,” 1982.
[19] A. Goode, B. Gilbert, J. Harkes, D. Jukic, and M. Satyanarayanan, “Openslide: A
vendor-neutral software foundation for digital pathology,” 2013.
[20] T. Lin, “Labelimg,” Online: https://github.com/tzutalin/labelImg, 2015.
[21] K. Wada, “Labelme,” Online: https://github.com/wkentaro/labelme, 2020.
[22] G. Aresta, T. Araújo, S. Kwok, S. S. Chennamsetty, M. Safwan, V. Alex, B. Marami,
M. Prastawa, M. Chan, M. Donovan, G. Fernandez, J. Zeineh, M. Kohl, C. Walz, F. Ludwig, S. Braunewell, M. Baust, Q. D. Vu, M. N. N. To, E. Kim, J. T. Kwak, S. Galal,
V. Sanchez-Freire, N. Brancati, M. Frucci, D. Riccio, Y. Wang, L. Sun, K. Ma, J. Fang,
I. Kone, L. Boulmane, A. Campilho, C. Eloy, A. Polónia, and P. Aguiar, “Bach: Grand
challenge on breast cancer histology images,” Medical Image Analysis, vol. 56, pp. 122–
139, 2019.
[23] S.-R. Shih, I.-S. Jan, T.-C. Chang, and A. Chen, “Computerized cytologic features
of thyroid cancer diagnosis,” Endocrine Practice, suppl.SUPPLEMENT 1, vol. 24,
pp. 225–226, 04 2018. 著作權 - Copyright Allen Press Publishing Services Apr 2018;
最後更新 - 2019-07-31.