An Ensemble Approach for Automated Classification of Brain MRIs

Abstract

Classification of brain Magnetic Resonance Imaging (MRI) is finding widespread use in clinical practice as it greatly helps the radiologists and neurologists to diagnose the brain abnormalities with more accuracy and precision. Several machine learning models are been proposed in recent times to effectively classify the brain MRIs. Ensemble of classification models produces better classification performance compared to individual models. In this work, we have proposed and validated a Particle Swarm Optimization (PSO) based linear ensemble model for the classification of brain MRIs. The ensemble model combines the outcomes of three heterogeneous base classifiers, namely Artificial Neural Network (ANN), Least-squares Support Vector Machines (LS-SVM), k-Nearest Neighbors (k-NN). PSO is applied to optimally decide the individual contribution or weightage of the three base models towards the final results of the ensemble model as a whole. Discrete wavelet transform (DWT) is applied in the first stage for feature extraction and Principal Component Analysis (PCA) is applied for reducing the dimensionality of the extracted feature vectors before classification.