Extracting Valuable Associations among Textural Features of Medical Images

As a matter of fact, the textural features extracted from medical images have physical meaning. Some of them express the contrast, the uniformity, the homogeneity, the distortion and the suavity of the image. The idea of information digging for finding associations among these features in retina images is introduced. The proposed technique works in three stages. In the first stage, the Gray Level Co-event Matrix (GLCM) textural features are extracted and averaged for the 0°, 45°, 90° and 135° directions. In the second stage, feature selection is adopted using The relief algorithm to choose the most relevant textural features and discretize them to fall between two ranges of values (high and low). Finally, the FP-growth algorithm is applied on the selected features to discover useful associations among them. The results show that the normal retina images are highly associated with certain textural features in a certain range of values. Hence, These associations can be utilized for effective diagnosis of normal retina images.