Brain tumor classification using non-negative and local non-negative matrix factorization

In this paper, a synergy of signal processing techniques and intelligent strategies is applied in order to classify different types of brain tumors, so that to assist doctors in their diagnostic task. Magnetic resonance spectroscopy (MRS) provides information on the biochemical profile of tissue and is increasingly being used as a non-invasive method of classifying brain tumor. MRS is analysed using LCModel software to yield metabolite profiles. Yet previous works have not used the nonnegative information of MRS for classification. A novel scheme is proposed in this paper. Firstly, non-negative and local nonnegative matrix factorization (NMF, LNMF) are used to extract features from metabolite profiles. Then support vector machines (SVM) and linear discriminant analysis (LDA) are applied to train classifiers based on features extracted by NMF and LNMF. The new scheme can extract meaningful features and therefore obtains a classifier with good generalization. Experimental results show that the new method has better performance than other previous ones.