A Model of a Two-Stage Classification System for Glial Tumors in Magnetic Resonance Imaging

Using medical images from different modalities of diagnostic imaging is a prerequisite for the development of multiple algorithms for the search and classification of certain information. Almost all proposed classifiers include some pre-processing of the image such as background noise removal, segmentation, etc. In this paper, a model of a system for classifying magnetic resonance images (MRIs) of the brain is proposed, with the input data coming directly from the relevant medical database. The goal is to find sufficiently informative and discriminating features of the images without pre-processing them. A model of a two-level classification system for glial tumors is implemented. By means of a descriptor obtained by the method of principal components analysis (PCA), the images are divided into normal and abnormal depending on the absence or presence of a glial tumor, respectively. During the second stage, additional wavelet energy and entropy signatures are used to further classify the abnormal images of low- and high-grade gliomas. The quality measures: F1-score and the Matthew's correlation coefficient are used to evaluate the system's performance.