請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/69137
標題: | 使用卷積神經網路之三維乳房彈性成像的電腦輔助腫瘤診斷 Computer-aided Tumor Diagnosis for 3-D Breast Elastography using Convolutional Neural Network |
作者: | Yuan-Feng Zhu 朱元豐 |
指導教授: | 張瑞峰(Ruey-Feng Chang) |
關鍵字: | 乳癌,彈性超音波,超音波影像,三維卷積神經網路,電腦輔助診斷系統, breast cancer,ultrasound image,elastography,3-D convolutional neural network,computer-aided diagnosis system, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 根據統計,乳腺癌不僅是常見的癌症,而且也是女性癌症死亡的第二大原因。因此,早期的檢查和治療是降低死亡率的最有效的方法。近年來,超音波技術中的彈性成像已成為乳腺癌的診斷的一種有效途徑。本文提出了基於三維卷積神經網路(CNN)结合乳房弹性超音波影像的電腦輔助診斷(CAD)系統,對腫瘤進行良惡性分類。首先,我們使用U-Net分割方法從三維B-mode影像中提取腫瘤區域。完成分割後,我們得到了只含有腫瘤區域的B-mode影像和彈性影像,其中彈性影像包含了腫瘤組織彈性的資訊信息。我們也從U-Net中獲得了腫瘤的遮罩影像,遮罩影像包含了腫瘤輪廓的形狀特徵。最後,用我們的CAD系统結合五類影像,分別是B-mode影像,彈性影像,遮罩影像,只含有腫瘤的B-mode影像和只含有腫瘤的彈性影像來診斷腫瘤的良惡性。此外我們使用集成學習的方法降低預測的變異數進而提高診斷性能。在此實驗中,利用63個良性腫瘤和20個惡恶性腫瘤組成的83個經過病理驗證的腫瘤來測試我們所提出的方法。從實驗結果可知,我們提出的基於3-D CNN的集成模型結合了B-mode影像,遮罩影像和只含有腫瘤的彈性影像有最好的性能,腫瘤診斷系統的準確率、靈敏度、特異性分别為90.36% (75/83),90.00% (18/20)和90.48% (57/63),ROC曲線下的面積為0.9374。 Breast cancer is the common and second leading cause of cancer death in women. Therefore, early examination and treatment are the most effective way to reduce the mortality rate. Recently, the breast elastography techniques had become a helpful examination for cancer diagnosis. As a result, the computer-aided diagnosis (CAD) system based on 3-D convolutional neural networks (CNN) using breast elastography is proposed to distinguish the tumors as benign or malignant. At first, the U-Net segmentation method is used to extract the tumor region from 3-D B-mode images. After generating the tumor region, the mask images contain complete shape features. Thus, the tumors are diagnosed in the proposed CAD systems using B-mode images, elastographic images, mask images, masked elastographic images, and masked B-mode images. In addition, the ensemble method is adopted to reduce prediction variance and enhance diagnosis performance. In this experiment, totally 83 biopsy-proved lesions composed of 63 benign tumors and 20 malignant tumors are used to evaluate our proposed method. According to results, the best diagnosis performance is using the ensemble method combined with B-mode images, mask images, and masked elastographic images and the accuracy, sensitivity, specificity, and AUC are 90.36% (75/83), 90.00% (18/20), 90.48% (57/63), and 0.9374, respectively. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/69137 |
DOI: | 10.6342/NTU201801692 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-1.pdf 目前未授權公開取用 | 1.71 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。