跳到主要內容

臺灣博碩士論文加值系統

(44.192.48.196) 您好!臺灣時間:2024/06/14 15:59
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

: 
twitterline
研究生:陳昱翔
研究生(外文):Yu-Cheung Chen
論文名稱:基於深度學習技術之智慧牙科診斷輔助工具開發和應用
論文名稱(外文):Development and application of an intelligent dental diagnosis assistant tool based on deep learning technology
指導教授:蔡政穆張傳旺
指導教授(外文):Cheng-Mu TsaiChuan-Wang Chang
口試委員:曾信賓黃俊銘
口試委員(外文):Shin-Pin TsengChun-Ming Huang
口試日期:2023-07-17
學位類別:碩士
校院名稱:國立中興大學
系所名稱:精密工程研究所
學門:工程學門
學類:機械工程學類
論文種類:學術論文
論文出版年:2023
畢業學年度:111
語文別:中文
論文頁數:39
中文關鍵詞:CNN開咬深咬遠端看診診斷建議基因演算法
外文關鍵詞:CNNOpen biteDeep biteRemote consultationsDiagnostic recommendationsGenetic Algorithms
相關次數:
  • 被引用被引用:0
  • 點閱點閱:35
  • 評分評分:
  • 下載下載:4
  • 收藏至我的研究室書目清單書目收藏:0
本論文研究以CNN開發一個預測牙齒咬合情況的智慧醫療系統,可以預測牙齒矯正中的開咬、深咬和正常三種情況。傳統上醫生都是用目視和經驗的檢查方式去判斷一位患者有無咬合問題,此方式不僅會增加醫療療程,也會增加判斷上的錯誤率,因此開發一個系統提供醫師遠端看診和診斷建議,可以極大地提高診斷準確性和治療效果。模型選擇上,CNN除了處理圖片數據外,還可以處理非圖片的矩陣數據,但此數據的每個網格數據必須與相鄰網格數據相近,這與我牙齒座標數據相符,除此之外,還嘗試使用其他種類的傳統演算法,只因最後結果不如CNN來的準確,因此還是採用CNN作為本研究開發系統的演算法。資料部分,由於初始資料中三個類別的比例不平衡且資料量有限,所以需要從最初的資料產生虛擬資料,而訓練用資料與測試用資料則按8:2的比例分開產生,從原本在牙體技術所得到258筆資料上升到1128筆資料,此為訓練與測試資料總數,產生虛擬資料的方式參考了基因演算法交配產生子代的方式,將初始資料當作母代,交配產生的子代當作用於訓練的虛擬資料,此方法將原先訓練資料的正確率提升了近15%,從原先80%提升近95%,在測試資料的部分,正確率提升約25%,從原先60%提升至85%。
This paper presents the development of an intelligent dental occlusion status software using a CNN to predict three types of tooth occlusion: open bite, deep bite, and normal bite. Traditional diagnosis of bite problems by doctors relies on visual inspection, which can be time-consuming and prone to errors. To address this issue, a system has been developed to provide remote consultations and diagnostic recommendations to doctors, which can greatly improve the accuracy of diagnosis and treatment outcomes. The CNN was chosen as the model because it can handle not only image data but also matrix data that is similar to the dental coordinate data, provided that each grid data is similar to the adjacent grid data. Due to the limited and imbalanced initial data, virtual data was generated from the initial data to increase the training data. The training and validation data were separated in an 8:2 ratio, resulting in a total of 1128 data points for training and validation, up from the original 258 data points obtained from dental technology. The virtual data was generated using a Genetic Algorithms, mimicking the process of mating between parents to produce offspring. This method improved the accuracy of the training data from 80% to nearly 95%, and the accuracy of the validation data from 60% to nearly 85%.
摘要 i
Abstract ii
目錄 iii
圖目錄 v
表目錄 vii
第1章 緒論 1
1.1 文獻回顧 1
1.2 研究目的與動機 2
1.2.1牙齒矯正的問題和需求 2
1.2.2智慧醫療系統的發展和優勢 3
1.3 論文架構 3
第2章 資料來源和資料擴充 4
2.1 光學取模系統和齒列矯正規劃系統 4
2.2 資料型態和定義 5
2.3 資料量擴充 7
2.4 資料歸一化 7
第3章 模型理論和選擇 9
3.1 卷積神經網路的原理和應用 9
3.1.1卷積對座標資料操作流程 9
3.1.2池化層操作理論 12
3.1.3神經元(資料轉換) 12
3.1.4啟動函數(activation function) 14
3.1.5損失函數(loss function): 15
3.1.6優化器(Optimizer) 15
3.1.7卷積神經網路總流程 18
3.2傳統演算法種類 19
3.2.1鄰近演算法 19
3.2.2決策樹 21
3.2.3隨機森林 21
第4章 實驗過程與方法 23
4.1研究流程 23
4.2資料準備 24
4.3設計神經網路 25
4.4設計KNN模型 29
4.5設計決策樹模型 30
4.6設計隨機森林模型 30
4.7最終模型評估與選擇 31
第5章 結果分析與未來展望 34
5.1資料前處理與數據擴充的效果 35
5.2模型性能的評價和提高 35
5.3未來展望 36
參考文獻 37
[1]G. Muhammad, F. Alshehri, F. Karray, A.E. Saddik, M. Alsulaiman and T.H. Falk, “A comprehensive survey on multimodal medical signals fusion for smart healthcare systems,” Information Fusion, Volume 76, pp.355-375, 2021.

[2] R. Beckers, Z. Kwade and F. Zanca, “The EU medical device regulation: Implications for artificial intelligence-based medical device software in medical physics,” Physica Medica, Volume 83, pp.1-8, 2021.

[3]K. A. Yaeger, M. Martini, G. Yaniv, E. K. Oermann and A. B. Costa, “United States regulatory approval of medical devices and software applications enhanced by artificial intelligence,” Health Policy and Technology, Volume 8, pp.192-197, 2019.

[4] M. Bastan and O. Yilmaz., “Multi-View Product Image Search with Deep ConvNets Representations,” World Scientific, 2018.

[5] Z. Zhongheng, “Introduction to machine learning: k-nearest neighbors,” Annals of Translational Medicine, 2016.

[6] B. Blakley, “Decision Analysis Using Decision Trees for a Simple Clinical Decision,” National Library of Medicine, 2012.

[7] Y. Mishina, R. Murata, Y. Yamauchi, T. Yamashita and H. Fujiyoshi, “Morphology and fracture behavior of lithium disilicate dental crowns designed by human and knowledge-based AI,” IEICE Transactions on Information and Systems, Volume E98.D, pp.1630-1636, 2015.

[8]A. Zhou and Z. Hua, “T1 Open-environment machine learning,” JF National Science, Volume 9, 2022.

[9]Y. Chen, J. K. Y. Lee, G. Kwong, E. H. N. Pow and J. K. H. Tsoi, “Morphology and fracture behavior of lithium disilicate dental crowns designed by human and knowledge-based AI,” Journal of the Mechanical Behavior of Biomedical Materials, Volume 131, 2022.

[10]K. Karako, P. Song, Y. Chen and W. Tang, “Realizing 5G- and AI-based doctor-to-doctor remote diagnosis: opportunities, challenges, and prospects,” BioScience Trends, Volume 14, pp.314-317, 2020.

[11]J. Liu, Y. Chen, S. Li, Z. Zhao and Z. Wu, “Machine learning in orthodontics: Challenges and perspectives,” National Library of Medicine, 2021.

[12]A. G. Brega, L. Jiang, R. L. Johnson, A. R. Wilson, S. J. Schmiege and J. Albino, “Health Literacy and Parental Oral Health Knowledge, Beliefs, Behavior, and Status Among Parents of American Indian Newborns,” National Library of Medicine, 2020.

[13]A. Bogosavljevic, V. Misina, J. Jordacevic, M. Abazovic, S. Dukic, L. Ristic and D. Dakovic, “Treatment of teeth in the esthetic zone in a patient with amelogenesis imperfecta using composite veneers and the clear matrix technique: A case report,” National Library of Medicine, 2016.

[14]T. F. Lou and W. H. Hung, “Revival of Classical Algorithms: A Bibliometric Study on the Trends of Neural Networks and Genetic Algorithms,” Dept Management Informat Syst, 2023.

[15]The FDI tooth position representation , Retrieved from https://commons.wikimedia.org/wiki/File:Zahneschema_UK_1.jpg

[16]D. Shin, B. Zaid and F. Biocca, A. Rasul, “In Platforms We Trust?Unlocking the Black-Box of News Algorithms through Interpretable AI,” Behaviour and Information Technology, pp.235-256, 2022.

[17]Z. Li, S. H. Wang, R. R. Fan, G. Cao, Y. D. Zhang and T. Guo, “Teeth category classification via seven-layer deep convolutional neural network with max pooling and global average pooling,” Wiley Online Library, volume 29, pp 577-583, 2019.

[18]D. Boob, S. S. Dey and G. Lan,” Complexity of training ReLU neural network,” Discrete Optimization, Volume 44, 2022.

[19]H. Chen and X. Yao, "Regularized Negative Correlation Learning for Neural Network Ensembles," IEEE Transactions on Neural Networks, Volume 20, pp. 1962-1979, 2009.

[20]Z. Yan, J. Chen, R. Hu, T. Huang, Y. Chen and S. Wen, “Training memristor-based multilayer neuromorphic networks with SGD, momentum and adaptive learning rates,” Neural Netw, Volume 128, pp.142-149, 2020.
連結至畢業學校之論文網頁點我開啟連結
註: 此連結為研究生畢業學校所提供,不一定有電子全文可供下載,若連結有誤,請點選上方之〝勘誤回報〞功能,我們會盡快修正,謝謝!
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top