Qualitative Approach of Empirical Mode Decomposition-Based Texture Analysis for Assessing and Classifying the Severity of Alzheimer’s Disease in Brain MRI Images

Abstract

Medical image processing systems are widely adopted in several real-time diagnostic systems due to their significant nature of information extraction and processing which helps to predict the early stages of illness. Alzheimer’s disease (AD) is one of the most chronic and challenging diseases in the medical diagnostic field. This disease is responsible for neurodegenerative brain disorder and attacks the brain cells and nerves that result in affecting the brain functionality and finally cause dementia. In this work, the prime focus is on the early prediction of Alzheimer’s disease using image processing-based machine learning techniques. Erstwhile, extensive studies are researched using pathological- and MRI-based systems which show the issues caused due to the brain’s white matter damage. Nevertheless, these studies do not provide that how white matter damage is associated with the AD and its classification at multiple stages. Conferring to the proposed approach, an improved feature extraction technique is introduced by combining empirical mode decomposition and gray-level co-occurrence matrix (GLCM). In order to abstract robust features, several image preprocessing steps are applied such as image enhancement, and later feature extraction is applied followed by the classification where multiple classifiers such as KNN, decision tree classifier, RBF, and support vector machine classification are castoff to assess the performance of feature extraction technique. Projected methodology obtains promising performance for the classification of various stages of AD and consequently can be employed for real-time application for early prediction of AD. An extensive experimental study is carried out for the anticipated approach that is implemented on OASIS brain imaging dataset. Experimental study shows that support vector machine, KNN, decision tree, and RBF classification techniques achieve that the accuracy as 39.06%, 80.20%, 98.43%, and 77.08%, respectively. Combination of the proposed feature extraction technique and decision tree classification scheme achieves a promising classification performance.