Classification of Brain Tumor using PCA-RF in MR Neurological Images

Brain tumour identification is a very crucial step that is required to diagnose and identify the degree of severity. The identification of tumor and classification helps the radiologist to diagnose properly and treat in according with that. Therefore, this paper presents a multi-class brain tumor classification in MR neurological images using random forest (RF) classifier with three different approaches. In the proposed approach, gray level co-occurrence matrix (GLCM), shape and local binary pattern (LBP) features are computed and further principal component analysis (PCA) is used for dimensionality reduction of the computed feature vector. All the features are extracted into the small patches having size of 3 x 3 in a local window. Several experiments are performed on a brain tumor dataset having 3064 T1-weighted contrast-enhanced images and comparative analysis of RF, RF-PCA and RF-PCA with random selection is presented. The experiment results show that RF-PCA with random selection performs better than the other approaches with a testing and validation accuracy of 88.72% and 85.56%.