Minimal Feature Set for Unsupervised Classification of Knee MR Images
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.
Keywords: Unsupervised Classification, Segmentation, Feature Extraction, Knee MR Images.
Download Full-Text
ABOUT THE AUTHORS
Rajneet Kaur
Assistant Professor in CSE dept. in Sri Guru Granth Sahib World University
Rajneet Kaur
Assistant Professor in CSE Dept. in Sri Guru Granth Sahib World University
Naveen Aggarwal
Assistant Professor in CSE Dept. in Panjab University
Rajneet Kaur
Assistant Professor in CSE dept. in Sri Guru Granth Sahib World University
Rajneet Kaur
Assistant Professor in CSE Dept. in Sri Guru Granth Sahib World University
Naveen Aggarwal
Assistant Professor in CSE Dept. in Panjab University