查詢結果分析
來源資料
頁籤選單縮合
題名 | 乳房X光影像中微鈣化之強化、特徵萃取及辨識=Image Enhancement, Feature Extraction and Classification of Microcalcifications in Mammograms |
---|---|
作者 | 傅家啟; 李三剛; 溫嘉憲; 蔡明倫; 林宏銘; | 書刊名 | 中華放射線醫學雜誌 |
卷期 | 28:4 2003.08[民92.08] |
頁次 | 頁217-230 |
分類號 | 416.12 |
關鍵詞 | 分類; 資料探勘; 特徵萃取; 影像強化; 乳房X光影像; 微鈣化; Classification; Data mining; Feature extraction; Image enhancement; Mammogram; Microcalcification; |
語文 | 中文(Chinese) |
中文摘要 | 在乳房X光影像的診斷中,微鈣化徵象的出現視為乳癌早期訊號,因此正確的檢測出微鈣化為有效診斷之一重要指標。本研究提出一兩階段微鈣化檢測模式,第一階段計算可疑微鈣化點之位置及邊界,第二階段萃取可疑微鈣化點之紋理、空間域與頻率域等特徵後,以資料探勘方式篩選出對於微鈣化點分類有效之特徵,最後以分類器檢測微鈣化點之發生區。實驗結果顯示,經第一階段之保守濾波器處理後,其可疑微鈣化點可有效涵蓋經病理切片所檢測出之微鈣化群區域;第一階段之特徵選擇方面,經資料探勘所選擇特徵作為分類器輸入向量之檢測績效優於無資料探勘之檢測績效。第一階段保守濾波器之優點為使用者無須輸入參數或以迭代方式訓練系統,可大幅減少複雜度及系統開發時程,第二階段特徵經資料探勘選擇後,可減少將近一半的輸入向量維度之情況下同時提昇檢測績效。以荷蘭Nijmegen大學附屬醫院之乳房影像資料庫測試系統績效,實驗結果顯示結合第一階段之保守濾波器與第二階段之特徵向量資料探勘,檢測績效以ROC圖之Az值可達0.9847。 |
英文摘要 | Since microcalcifications in X-ray mammograms are the primary indicator of breast cancer, detection of microcalcifications is central to the development of an effective diagnostic system. This paper proposes a two-staged procedure for detecting microcalcifications in X-ray mammograms. The first stage, serving as a conservative filter, calculates the location and shape of all suspicious spots. The second stage classifies the spots based on features extracted from texture, spatial domain and frequency domain data. A sequential forward search data mining algorithm selects the classification input vector, which consists of features sensitive only to microcalcifications. Experimental results show that the suspicious spots located in the first stage effectively cover the microcalcification clusters, and that sequential forward search feature selection improves classification performance while reducing the dimension of the input vector by nearly 50%. Since the proposed model is data driven, it is requires neither an a priori parameter choice nor an iterative training algorithm. When tested on the Nijmegen University Hospital (Netherlands) database, the Az value of the ROC distribution of the proposed model can achieve 0.9847 |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。