Detection of Alzheimer's Disease Using a Convolutional Neural Network

Abstract

This paper proposes a relatively simple neural network for detecting Alzheimer's disease and its stages by analyzing cerebral magnetic resonance images (MRI). The proposed network architecture was built starting with the simplest architecture with only an input and an output layer, adding the necessary to improve performance, arriving at a network that contains three types of hidden layers: convolutional and pooling layers for extracting features from images and the fully connected layers for the classification of these images. The use of convolution is due to the limitations of multilayer perceptron when it comes to images. Indeed, this type of network which processes the image pixel by pixel does not really model what the human brain does to process the images. Data used in this study to evaluate and to test the proposed model as well as each intermediate structure come from the kaggle platform. It is a database accessible to the public under the Open DataSet license. The results obtained are encouraging, in comparison with other works that use data augmentation and transfer learning.