Comparative Analysis of Two Approaches to the Small Sample Size Problem in Classifying Glial Tumors
Comparative Analysis of Two Approaches to the Small Sample Size Problem in Classifying Glial Tumors

Summary/Abstract: Linear Discriminant Analysis (LDA) is one of the machine learning (ML) methods for reducing the dimensionality of the original feature space. A major drawback of the method is the so-called Small Sample Size (SSS) problem. In image classification, the dimensionality of the corresponding feature vector is much larger than the number of samples used. This leads to a singularity of the within-class scatter matrix, making the conventional LDA approach inapplicable. A number of LDA-SSS techniques have been proposed to overcome this problem, primarily in face recognition. In this work, the performance of two such approaches in the task of classifying magnetic resonance images (MRIs) of patients’ brains were investigated through the subspace LDA method (Fisherfaces) and the Moore-Penrose pseudoinverse matrix method. The evaluation of the performance of these methods is based on accuracy, the ratio of the correctly classified data to all grouping data. The MRIs belonging to the respective classes, determined by the Minimum Distance Classifier (MDC), is consistent with the independent diagnosis of three radiologists. The conducted experimental studies show that the computer-aided diagnosis (CAD) system for brain tumor classification based on the pseudoinverse algorithm shows higher performance accuracy, but also significant computational complexity.

• Details
• Contents