Feature Selection and Classification for Mammographic Microcalcification Clusters 邱正宏、傅家啟
E-mail: [email protected]
ABSTRACT
The microcalcifications in X-ray Mammography are the major index of breast cancer. The successful classification of
mammographic microcalcifications by type (benign and malignant) is a key factor of effective treatment. Most published literature related to computer aided diagnosis (CAD) development focus on detection of mammographic microcalcifications. In this thesis, a diagnostic method with a two-staged scheme and the sequential forward selection (SFS) based feature extraction is developed to detect microcalcification clusters in X-ray mammograms. This thesis proposed a computer-aided diagnosis (CAD) system for the automatic detection two kinds of digitized X-ray mammogram database of microcalcification clusters. The proposed system consists of two main steps. Microcalcifications are detected in the first stage, the algorithm is applied for clustering and feature extraction for 35 cluster features. The discriminatory power of these features is analyzed via sequential forward selection method. Experiment results show that the Dutch’s Nijmegen database’s minimum MSE error for SVM test set was reached when the top 34 features was selected. The Taiwan’s Chang-Hua Christian Hospital’s database’s minimum MSE error for SVM test set was reached when the top 28 features was selected. Cluster mean and maximum gray in -viiCluster are the significant features for
microcalcification clusters classification for both types of mammographic databases. Data analysis showed that features selected by the SFS out perform the features without being selected. Therefore, SFS has the potential to simultaneously reduce system complexity and increase classification performance. Keywords: Microcalcification Cluster, Feature Selection, Support Keywords : Microcalcification Cluster, Feature Selection, Support Vector Machines
Table of Contents
封面內頁 簽名頁 授權書...iii 中文摘要...v
Abstract...vii 誌謝...ix 目錄...x 圖目
錄...xii 表目錄...xiv 第一章 緒論...1 1.1研究目的 與動機...1 1.2研究範圍...3 1.3研究方法...3 第二章 文獻 探討...4 2.1徵像檢測...4 2.2良惡性判讀...4 2.2.1特徵萃 取...5 2.3叢集的演算法...12 2.4資料設置...14 2.5特徵選 擇...15 2.6分類器...16 2.7小結...22 第三章 研究架 構...24 3.1研究流程...24 3.1.1徵像檢測...25 3.1.2良惡性 判讀...26 3.2分類器參數選擇及特徵選擇...29 3.3績效衡量...32 第 四章 實驗結果...34 4.1 實驗設置...34 4.1.1徵像檢測流
程...35 4.1.2良惡性判讀...36 4.2實驗結果及分析...38 4.2.1徵像 檢測結果...38 4.2.2良惡性鈣化判讀結果...43 4.2.3小結...60 第 五章 結論與未來研究發展...63 5.1 結論...63 5.2未來研究發
展...65 參考文獻...66 附錄A...69 附 錄B...70 附錄C 最佳特徵組合流程...72
REFERENCES
1. 吳智誠, “資料探勘於影像資訊之應用-以乳房微鈣化特徵處理為案例”, 大同指導教授:李三剛), 2001. 2.傅家啟,李三剛,溫嘉憲,蔡明倫, 林宏銘,”數位乳房X光微鈣化之強化、 特徵萃取及辨識”,中華放射醫誌, 2003. 3. 蔡明倫, “二維點狀影像資訊之強化、特徵擷取及辨 識-以X光乳房微鈣化 檢測為案例”, 大葉大學工業工程學系碩士班論文(指導教授:傅家啟，共 同指導教授:李三剛), 2002. 4. Bazzani, A., Bevilacqua, A., Bollini, D., Branca, Lanconelli,N., Riccardi , A.,Romani,D., “An SVM classifier to separate false signals form microcalcifications in digital mammograms” Phys. Med. Biol. 46, pp.1651-1663,Mar 2001. 5.Fu, J. C., Lee, S. K., Wong, S. T. C., Yeh, J. Y., Wu, H. K., Image Segmentation、Feature Selection and Pattern Classification for Mammographic Microcalcifications, Artificial Intelligene in Medicine 2003 6.
Fuhr, H., Treiber ,O., Wanninger, F., “Cluster-oriented Detection of Microcalcifications in Simulated Low-Dose Mammographicists ,Mar 2002.
8. He ang-Kobrunner, S. H., Schreer, I., Dershaw, D. D., “Diagnostic 9. Br Mendez, A ,J., Souto, M., Vidal,Transform Techniques to the
Detection of Cluster Microcalcifications in Digital Mammograms”microcalcification features for,July 2000. 10.method of region grouping for microcalcification 12.Salfity, M, F., Kaufmann, G, H., Granitto, p., Ceccatto, H,A., “A Computer-Aided Diagnosis Method for Automated Detection and Classification of Clustered Microcalcifications in Mammograms”. 13. Soranti ., Schmidt, F., Mayer, H., Becker, M., Szepesvari, C., Graif, E., Winkler, P., Computer Aided Diagnosis of Clustered Microcalcifications Using Artificial Neural Nets ,2000. 14. S,Z, Hamid.,
P,N,Siamak., R, R,Farshid., “Shape-based and texture extraction for classification of microcalcifications in mammograms”, Med. Imag., pp.17-22, Feb. 2001. 15.Suykens, J A K, Gestel T V, Brabanter J D, Moor B E,Vandewalle J. ociation of physicyweast Imaging”, Thieme Medical Publishers, 1997. 9. Lado, M, J., Tahoces, P, G.10. Patrocinio, A, C., Schiabel, H., Benatti, R, H., Goes, C, E., Nunes, L,S., “Investigation of cluster automated classifier as part of a mammography CAD Scheme” 11. Qian ,W.,Mao,F.,Sun,X,. Zhang, Y., Song ,D., Clarke, R,A,. “An improveddetection in digital mammograms” ,June 2002. n, E Least Squares Suppor achines, Word S cientific 16. Digitized Mammogram Films
”, 17 ., “Shape-based andt Vector Mpublishing:2002. . Yu, S., Guan, L., “A CAD System for the Automatic Detection of Clustered Microcalcifications inIEEE Trans. Med. Imag., vol. 19, no. 2, pp. 115-126,Feb. 2000. . Zadeh, H, S., Nezhad, S ,P., Rad , F, Rtexture–based feature extraction for classification of microcalcifications in mammograms ” Image Processing Conference ,Feb 2001. 18. Zheng, B., Qian, W., Clarke, L. P., “Digital mammography mixed feature neural network with spectral entropy decision for detection of microcalcifications”, IEEE Trans. Med. Imag., vol. 15, no. 5, pp. 589-597, Oct. 1996. 19. http://www.doh.gov.tw/statistic/data/公佈欄資料檔/90死因/癌症8990.xls
