以機器學習斷層掃描影像紋理的電腦輔助診斷於低和高Fuhrman核等級腎細胞癌

Taipei Medical University Institutional Repository > 醫學科技學院 > 醫學資訊研究所 > 博碩士論文 > Item 987654321/58346

請使用永久網址來引用或連結此文件: http://libir.tmu.edu.tw/handle/987654321/58346

題名:	以機器學習斷層掃描影像紋理的電腦輔助診斷於低和高Fuhrman核等級腎細胞癌 Computer-aided Diagnosis of Low and High Fuhrman Nuclear Grades of Renal Cell Carcinoma with Machine Learning CT Texture Analysis
作者:	王震宇 Wang, Jenn-Yeu
貢獻者:	陳俊璋
關鍵詞:	電腦斷層掃描紋理分;腎細胞癌;特徵選擇;形態特徵;灰階分布圖(histogram);灰階共生矩陣的紋理特徵;Gabor紋理特徵 computed tomography texture analysis;renal cell carcinoma;feature selection;morphology feature;gray-level histogram;gray-level co-occurrence texture feature;Gabor texture feature
日期:	2019-05-23
上傳時間:	2020-01-06 13:06:03 (UTC+8)
摘要:	Fuhrman核分級可以做為腫瘤患者的標準治療資訊。探討合併電腦斷層掃描紋理分析與機器學習電腦斷層掃描圖像區分低和高Fuhrman核等級腎細胞癌的準確性。回顧性病例對照設計中，腎細胞癌患者選自癌症影像圖譜資料庫，使用機器學習分析電腦斷層掃描灰階矩陣特徵預測低和高Fuhrman核等級腎細胞癌。由專家手動圈選電腦斷層掃描橫截面圖像。對手動圈選病變進行紋理分析，並使用交叉驗證檢查重現性。評估了關於灰階分布圖(histogram)、形態特徵、灰階共生(co-occurrence) 矩陣的紋理特徵和紋理特徵 (Gabor)，選擇最具有區別性的特徵來成功分類低和高Fuhrman核等級腎細胞癌。通過驗證(validation)來評估電腦斷層掃描紋理特徵的診斷準確性。向量支持分類器放入70個特徵比較來區分多偵測器電腦斷層掃描圖像上的兩種類型的腎細胞癌顯示接受者操作特徵曲線下面積的統計有最佳的面積(0.82)。產生的電腦輔助診斷系統可以提供泌尿科醫生腫瘤分級的建議，作為醫療決策輔助。 Fuhrman nuclear grading can add a piece of information to the standard care of patient with clear cell renal cell carcinoma. To explore the accuracy of texture analysis to distinguish between high-grade and low-grade Fuhrman nuclear grades of clear cell renal cell carcinoma on computed tomography images. In a retrospective case-control design, patients with RCC are selected from the “The Cancer Image Atlas” (TCIA) database. Cross-sectional computerized tomography (CT) images were contoured manually by experts. Texture analysis was done for each lesion, and reproducibility was examined by validation. Image features regarding morphology feature, the gray-level histogram, gray-level co-occurrence matrix and Gabor texture feature, were assessed. The most relevant features were chosen to generate classifiers. Diagnostic accuracy of texture features was evaluated by validation. Support Vector Machine (SVM) classifiers using 70 features demonstrated optimal area under receiver operating characteristic (AUROC) curve (0.82) statistics. When the morphology features, intensity features and texture features are combined in classifiers, a computer-aided diagnosis (CAD) system is generated. The developed CAD system may give suggestions of tumor grading to the urologists as an aid in decision making.
描述:	碩士指導教授：陳俊璋委員：張詠淳委員：羅崇銘
資料類型:	thesis
顯示於類別:	[醫學資訊研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	144	檢視/開啟

檢視Licence

在TMUIR中所有的資料項目都受到原著作權保護.

TAIR相關文章

資料載入中.....