Brain MRI classification using an ensemble system and LH and HL wavelet sub-bands features

A new classification system for brain images obtained by magnetic resonance imaging (MRI) is presented. A three-stage approach is used for its design. It consists of second-level discrete wavelet transform decomposition of the image under study, feature extraction from the LH and HL sub-bands using first order statistics, and subsequent classification with the k-nearest neighbor (k-NN), learning vector quantization (LVQ), and probabilistic neural networks (PNN) algorithms. Then, an ensemble classifier system is developed where the previous machines form the base classifiers and support vector machines (SVM) are employed to aggregate decisions. The proposed approach was tested on a bank of normal and pathological MRIs and the obtained results show a higher performance overall than when using features extracted from the LL sub-band, as usually done, leading to the conclusion that the horizontal and vertical sub-bands of the wavelet transform can effectively and efficiently encode the discriminating features of normal and pathological images. The experimental results also show that using an ensemble classifier improves the correct classification rates.