請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/63349
標題: | 應用於結腸鏡檢查質量保證之電腦輔助盲腸到達率監視之方法 Computer-aided Cecal Intubation Rate Surveillance for Colonoscopy Quality Assurance |
作者: | Fa-Yao Tseng 曾法堯 |
指導教授: | 陳中平(Chung-Ping Chen) |
關鍵字: | 盲腸,盲腸到達率,大腸鏡影像,影像辨識,分類器,類別不平衡,深度學習,卷積神經網路,集成學習, Cecum,Cecal Incubation Rate,Colonoscopy,Image recognition,Classifier,Class imbalance,Deep learning,Convolutional neural network,Ensemble learning, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 結腸鏡檢查為當今預防大腸癌發生的最佳方法,不過因為其為人為操作的技術,品質管理與監測亦為相當重要的一環,當中以四項指標來評估結腸鏡檢查之品質,如盲腸到達率(Cecal Intubation Rate)、腺瘤偵測率(Adenoma Detection Rate)、腸道準備(Bowel Preparation)以及退出時間(Withdrawal Time),研究發現,盲腸到達率低會導致結腸鏡檢查後大腸癌發生率提高,為了提高醫療質量,美國胃腸病學院(ACG)和美國胃腸內視鏡學會(ASGE)均建議計劃進行結腸鏡檢查的患者向內視鏡醫師詢問其個人盲腸到達率,但是大多數內視鏡醫師很可能不容易了解這些數據,僅能提供估算的結果。因此,在本論文中,我們針對盲腸到達率提出一套新的偵測方法,透過深度學習和卷積神經網路採用新穎的Xception及EfficientNet架構,並以F1分數處理類別不平衡的問題,最後使用集成學習進一步改善預測結果,自動化判斷每一次結腸鏡檢查是否到達盲腸,以增加結腸鏡手術的品質。實驗結果顯示,我們提出的方法與之前相比,提升了9%以上的準確率並達到90.66%的靈敏度以及86.60%的特異度。 Colonoscopy is the best way to prevent the colorectal cancer (CRC) nowadays. However, it is a highly operator-dependent examination. Therefore, quality assurance and surveillance are also quite important. There are four indicators used to evaluate the quality of colonoscopies, namely Cecal Intubation Rate, Adenoma Detection Rate, Bowel Preparation, and Withdrawal Time. In the previous study, it has been proved that poor cecal intubation rate is correlated to increase risk of post-colonoscopy CRC in order to improve the quality of medical care, both the American College of Gastroenterology (ACG) and the American Society for Gastrointestinal Endoscopy (ASGE) recommend that patients planning to undergo a colonoscopy ask the endoscopist for their personal cecal intubation rate. However, most of endoscopists do not understand these data clearly and can only provide a guesstimate of this metric. Therefore, in this thesis, we propose a novel methodology for detecting cecal intubation, which adopts deep learning and convolutional neural network (CNN) techniques, two strengthening architectures are used, Xception and EfficientNet. Handle class imbalance problem through the F1 score indicator. Furthermore, the ensemble method enhances results to make sure successful cecal intubation for each examination, thereby increasing the quality of colonoscopy. Comparing with our previous work, the proposed methodology improved the detecting accuracy by over 9% and achieved 90.66% in sensitivity and 86.60% in specificity. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/63349 |
DOI: | 10.6342/NTU202000779 |
全文授權: | 有償授權 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-109-1.pdf 目前未授權公開取用 | 2.49 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。