Main Article Content
Abstract
We have studied the effectiveness of using texture features derived from gray-level co-occurrence matrix
(GLCM) matrices for classification of cystic mass and non-cystic mass in ultra sonograms. Twenty-three (23)
region of interest (ROIs) containing cystic masses and fifty-five (55) non-cystic masses were extracted from ultra
sonogram for this study. For each ROI of 50x50 pixels, seven features (energy, inertia, entropy, homogeneity,
maximum probability, inverse difference moment, and correlation) were calculated. The importance of each
feature in distinguishing cystic masses from non-cystic masses was determined by linear discriminant analysis
with SPSS version 11.5 program. As a result of a study, it was found that all seven features can distinguishing
cystic masses from non-cystic masses with an accuracy about 91 %-92.3%. Those levels of accuracy also found
when two features (energy and inverse difference moment) was excluded from analysis. The result demonstrate
the feasibility of using texture features based on GLCM for distinguishing cystic masses from non-cystic masses
of ultra sonogram .
Keywords: Gray-level Co-occurrence Matrix Ultrasonografi, massa kistik, fitur tekstur, analisis tekstur, analisis
diskriminan