Feature Ranking of Spatial Domain Features for Efficient Characterization of Stroke Lesions

摘要

Development of automatic framework for efficient characterization of brain lesions is a significant research concern due to the complex properties exhibited by the brain tissues. This study focuses on observing the properties of such composite structures in order to identify optimal features for characterizing the properties of normal and abnormal brain tissues. This work initially applies Fuzzy C Mean algorithm to identify the region of interest. After segmentation, four different types of features are extracted from the region of interest. These features include first-order parameters, Gray-level Co-occurrence Matrix (GLCM) parameters, Laws texture features, and Gray-Level Run-Length Matrix (GLRLM) parameters. These identification features were ranked in order of pertinence with the help of Mutual Information and Statistical Dependence-based feature ranking algorithms. Based on the inference obtained from the Mutual Information and Statistical Dependence-based feature ranking algorithms, twelve best features are selected for characterizing the properties of the normal and abnormal brain tissues.