請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/78329
標題: | 結合倍頻顯微術和深度神經網絡以從事乳房外柏德氏症之術前邊緣鑑定 Margin Assessment of Extramammary Paget’s Disease Based on Harmonic Generation Microscopy with Deep Neural Networks |
作者: | Chia-I Chen 陳佳怡 |
指導教授: | 孫啟光(Chi-Kuang Sun) |
關鍵字: | 乳房外柏德氏症,倍頻顯微術,深度學習,三維卷積神經網絡,非侵入式,非線性光學, Extramammary Paget’s disease (EMPD),harmonic generation microscopy (HGM),deep learning,3D convolutional neural network (3D-CNN),noninvasive,nonlinear optics, |
出版年 : | 2021 |
學位: | 碩士 |
摘要: | 在現有的技術中,乳房外柏德氏症很難透過其臨床的病灶特徵來定義清除病灶的手術邊緣。在此研究中,我們提出一種新的診斷技術,此技術將倍頻顯微影像術與深度學習結合,能自動且即時判讀當下拍攝的三維皮膚組織影像為乳房外柏德氏症的病灶組織或是一般正常的皮膚組織。在此研究中我們從新鮮的乳房外柏德氏症手術檢體上的不同位置提取足夠的三維影像作為深度學習模型的訓練資料:利用倍頻顯微影像系統從皮膚表面開始垂直拍攝至皮膚深度180微米處。進而再對同一檢體進行病理組織切片及染色以後,我們將病理結果作為標準檢測結果對應至三維倍頻顯微影像為訓練深度模型作標記。我們僅使用2095張三維影像作為訓練和驗證數據集訓練深度神經網絡模型,並在191張的三維影像測試資料集上,取得93.20%的靈敏度、95.45%的特異度和94.24%的準確度於乳房外柏德氏症病灶組織與正常皮膚組織的分類。因此,研究結果表示此技術基於三維卷積神經網路,具有極高的潛力能夠透過非侵入式的方法及時準確提供該成像範圍為亞微米等級的區域是否為惡性或是正常皮膚組織的建議。 Surgical border of Extramammary Paget’s disease (EMPD) is difficult to be identified via its clinical appearance. In this thesis, we proposed a new diagnostic technique which combines nonlinear harmonic generation microscopy (HGM) with the deep learning method to instantaneously determine whether the imaged 3D stack is malignant EMPD or surrounding normal digitally. To demonstrate our proposal, in this thesis different locations of fresh EMPD surgical samples were 3D imaged starting from the surface up to a depth of 180 μm using stain-free HGM. With the followed histopathological H E examination of the same sample, we mapped the gold-standard results to 3D HGM image stacks with labels for the training of the deep learning model. With only 2095 3D image stacks as training and validation dataset, the results of EMPD and normal skin tissue classification achieve 93.20% sensitivity, 95.45% specificity and 94.24% accuracy on the test dataset of 191 3D image stacks. The thesis supports our proposed 3D convolutional-neural-network-based technique with a high potential to assist physicians to quickly map the EMPD preoperational margin by providing noninvasive instant suggestion regarding the imaged sub-millimeter site as malignant or surrounding normal with a high accuracy. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/78329 |
DOI: | 10.6342/NTU202100277 |
全文授權: | 有償授權 |
顯示於系所單位: | 光電工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-3001202122201300.pdf 目前未授權公開取用 | 2.35 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。