Model Based on Deep Feature Extraction for Diagnosis of Alzheimer’s Disease

Abstract

Alzheimer's disease (AD) is a neurodegenerative disease that results in loss of cognitive ability of the patient. Computational intelligence, more specifically Deep Learning, has been a powerful method for AD diagnosis. In this work we propose a model for AD diagnosis based on deep feature extraction for the classification using magnetic resonance imaging. This model aims to classify AD vs. HC (Healthy Controls). The database used in this project is the Minimal Interval Resonance Imaging in Alzheimer's Disease (MIRIAD), for validation of the proposed method. We select thirty slices from the upper region of the brain, above the eyes, for the apprenticeship in this work. The Convolutional Neural Network (CNN) architecture is designed in three convolutional layers to extract the best features of the selected region. After that, we put the selected attributes in a vector for learning and detection of patterns by another technique of computational intelligence. Finally, the data are partitioned with the 10-folds cross-validation method and trained with the Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbor (K-NN) algorithms with different parameters for evaluation. The results of accuracy are 0.8832, 0.9607 and 0.8745, for the algorithms mentioned above, respectively. According to a comparative analysis performed with other works of the literature, we can prove the efficiency and reliability of the model for the diagnosis of Alzheimer's disease.