期刊名称：International Journal of Computer Science Issues
印刷版ISSN：1694-0784
电子版ISSN：1694-0814
出版年度：2011
卷号：8
期号：6
出版社：IJCSI Press
摘要：Knee scans is very useful and effective technique to detect the knee joint defects. Unsupervised Classification is useful in the absence of domain expert. Real Knee Magnetic Resonance Images have been collected from the MRI centres. Segmentation is implemented using Active Contour without Edges. DICOM, Haralick and some Statistical features have been extracted out. A database file of 704 images with 46 features per images has been prepared. Unsupervised Classification is implemented with clustering using EM model and then classification using different classifiers. Learning rate of 5 classifiers (ID3, J48, FID3new, Naive Bayes, and Kstar) has been calculated. At the obtained learning rate minimal feature set has been obtained for unsupervised classification of Knee MR Images.
