Bendlets and Ensemble Learning Based MRI Brain Classification System

Abstract

Brain tumours are composed of cells where the growth is unrestrained. Though the incidence rate is lower, it is a serious threatening disease to human lives. For effective treatment, an accurate and quick method to classify Magnetic Resonance Imaging (MRI) is required. To identify the meaningful patterns and to interpret images, pattern recognition algorithms are developed. In this work, an extension of Shearlet transform named Bendlets is employed to interpret MRI images and decision making is done by ensemble learning using k-Nearest Neighbor (kNN), Naive Bayesian and Support Vector Machine (SVM) classifiers. The Bendlet and Ensemble Learning (BEL) based system utilizes Bendlet Co-Occurrence Features (BCFs) and Histograms of Positive and Negative Bendlet Coefficients (HPBC & HNBC) from the dominant sub-band as texture descriptors. The rate of classification by the BEL system for the 200 images from REpository of Molecular BRAin Neoplasia DaTa (REMBRANDT) is 99.5% at the initial stage (normal/abnormal classification) and 99% at the final stage (low-risk/high-risk classification). Based on the results, the implementation of BEL system could provide continuous monitoring of the progress of brain tumour very effectively and also offers a real-time response.