Comparison of the PCA and FLD Approaches in Glial Tumors Classification Systems

Computer-aided diagnosis (CAD) systems based on machine learning methods are an important component of medical practice. Principal Component Analysis (PCA) and Fisher Linear Discriminant (FLD) are among the main linear methods for feature extraction and reduction in recognition tasks. In this regard, a comparative analysis of the efficiency of the systems constructed by PCA and FLD algorithms, respectively, is carried out. As a continuation of our previous research, a classification system model for Magnetic Resonance Images (MRIs) of the brain is built using FLD. By analogy, the built CAD system detects the presence of a glial tumor, with a subsequent two-level and three-level gradation according to the degree of malignancy of the tumor. The input images are again not pre-processed and additional wavelet features are used: the normalized energy of the subimages and their non-normalized Shannon entropy. The comparative analysis of the pair of CAD systems is carried out through the quality measures: F1-score and the Matthews Correlation Coefficient. The validation of the obtained results is based on the diagnosis by three independent radiologists.