Alzheimer’s Disease Detection using Correlation based Ensemble Feature Selection and Multi Support Vector Machine

Abstract

In recent decades, machine learning techniques have been playing a crucial role in the field of computer aided diagnosis. This paper address the issue of automated Alzheimer’s disease detection on the basis of magnetic resonance imagining, and proposed a new supervised machine learning technique for Alzheimer’s disease diagnosis. Initially, an adaptive histogram equalization and region growing are employed on the collected brain scans for contrast improvement and skull removal. Next, Fuzzy C Means (FCM) clustering algorithm is applied in the enhanced brain scans to segment tissues like White Matter (WM), Cerebro Spinal Fluid (CSF), and Grey Matter (GM). In addition, feature extraction is accomplished in the segmented brain tissues using Gabor and local directional pattern variance features. In order to decrease the dimension of the extracted feature vectors, the correlation based on ensemble feature selection algorithm was proposed. Finally, the obtained optimal feature vectors are fed to Multi Support Vector Machine (MSVM) to classify Mild Cognitive Impairment (MCI), Alzheimer’s disease, and healthy controls classes. From the simulation outcome, the proposed ensemble feature selection with multi support vector machine model shows 9.58% and 5.09% improvement in classification accuracy on Open Access Series of Imaging Studies (OASIS) and Alzheimer’s Disease Neuroimaging Initiative (ADNI) datasets compared to the existing models.