請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70971
標題: | 以卷積神經網路實現棘上肌鈣化性肌腱炎之診斷 Diagnosis of Supraspinatus Calcific Tendinopathy using Convolutional Neural Networks |
作者: | Pei-Hsin Chiu 邱霈欣 |
指導教授: | 張書瑋(Shu-Wei Chang) |
關鍵字: | 棘上肌鈣化性肌腱炎,超音波,深度學習,機器學習,卷積神經網路,DenseNet121,醫療影像辨識, Supraspinatus calcific tendinopathy,Ultrasound,Deep learning,Machine learning,Convolutional Neural Network,DenseNet121,Medical images identification, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 以超音波檢查進行棘上肌鈣化性肌腱炎診斷時,常受限於其對操作者的經驗依賴性、放射科醫師人手不足以及城鄉之間民眾對於放射科服務可及性的差異,為了減緩以上之限制,本研究應用卷積神經網路建立人工智慧之機器學習模型,透過此模型建立電腦輔助系統以協助放射科醫師在超音波檢查中判斷長軸以及短軸的棘上肌鈣化性肌腱炎的存在。 本研究提出三個基於不同訓練集的機器學習模型,分別是僅基於長軸的長軸模型、僅基於短軸的短軸模型以及同時基於長、短軸的長短軸模型,以卷積神經網路之DenseNet121預訓練模型進行模型之建立與訓練,比較三者之間對於棘上肌鈣化性肌腱炎的判斷準確度以及其餘相關評估指標可知:同時基於長、短軸訓練之長短軸模型,其準確度為91.32%、靈敏度為87.89%、特異度為94.74%,是三種模型中最適合向放射科醫師提供超音波檢查中判斷長軸以及短軸的棘上肌鈣化性肌腱炎的存在的協助之模型。 Diagnosing supraspinatus calcific tendinopathy with ultrasound examination is usually limited by the dependence of ultrasound examination on experience of operators, the shortage of physicians, and the difference of the radiology service accessibility between urban and rural community. To eliminate these limitations, convolutional neural network was applied to build the machine learning models in this study. It was expected to build the computer aided system in order to assist physicians with diagnosing the existence of longitudinal view and transverse view of supraspinatus calcific tendinopathy in ultrasound examinations. Three models trained on different training dataset were proposed in this study, which were the longitudinal model trained on only longitudinal view of supraspinatus calcific tendinopathy, the transverse model trained on only transverse view of supraspinatus calcific tendinopathy, and the longi-trans model trained on both longitudinal view and transverse view of supraspinatus calcific tendinopathy. These models were built and trained by DenseNet121, which is a pre-trained model in convolutional neural network. Compared the accuracy and other evaluation index of these three models against the longitudinal view and transverse view of supraspinatus calcific tendinopathy, it was found that the accuracy of the longi-trans model was 94.74%, which was the highest among the three models, and its sensitivity and specificity were 87.89% and 94.74% respectively. In conclusion, the longi-trans model is the most suitable model to provide physicians assistance of diagnosing the existence of longitudinal view and transverse view of supraspinatus calcific tendinopathy in ultrasound examinations. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70971 |
DOI: | 10.6342/NTU202004111 |
全文授權: | 有償授權 |
顯示於系所單位: | 土木工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-2008202000084400.pdf 目前未授權公開取用 | 3.25 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。