Brain tumor segmentation and classification using cascaded random decision forests

Abstract

Automated segmentation and classification of brain tumor is important to avoid misdiagnosis and to improve chances of patients' survival. In this paper, we present a fully automated technique for segmentation and classification of brain tumor into three different regions namely Complete Tumor, Tumor Core and Enhancing Tumor. We use a cascaded Random Decision Forest (RDF) model for classification. In our experiments, we use BRATS 2013 3D MR images dataset which contains T1, T1c, T2 and Flair MRI sequences. These sequences are standard in clinical acquisition. Using 10-fold cross validation for evaluation, we achieve promising Dice scores of 0.90, 0.79 and 0.84 for Complete Tumor, Tumor Core and Enhancing Tumor, respectively.