3-D MRI Brain Scan Feature Classification Using an Oct-Tree Representation

This paper presents a procedure for the classification of specific 3-D features in Magnetic Resonance Imaging (MRI) brain scan volumes. The main contributions of the paper are: (i) a proposed Bounding Box segmentation technique to extract the 3-D features of interest from MRI volumes, (ii) an oct-tree technique to represent the extracted sub-volumes and (iii) a frequent sub-graph mining based feature space mechanism to support classification. The proposed process was evaluated using 210 3-D MRI brain scans of which 105 were from “healthy” people and 105 from epilepsy patients. The features of interest were the left and right ventricles. Both the process and the evaluation are fully described. The results indicate that the proposed process can be effectively used to classify 3-D MRI brain scan features.