Detection of alzheimer's disease using deep learning

Abstract

Machine Learning has been on top of its form over the last few years. It covers a vast area of predictive web browsing, email and text classification, object detection, face recognition etc. Among all of the other applications of machine learning, deep learning has gained more popularity over the last several years. It is helping researchers in the field of biomedical problems like detection of different types of diseases such as Cancer, Alzheimer, Malaria, Blood cell detection etc. Deep learning is a subset of machine learning algorithms that is used for classification, image processing etc. by extracting features. In our research, we used Convolutional Neural Network (CNN) for classi cation of Alzheimer patients and healthy patients from Magnetic Resonance Imaging (MRI) data. The dataset (OASIS-1) contains 416 subjects classi ed into non-demented and mild to moderate Alzheimer's disease. The classification of this type of medical data is very significant for creating a prediction model or system to examine the presence of the disease in different subjects or to estimate the phase of the disease. Classification of Alzheimer's disease has always been difficult and selecting the distinctive features is the most complicated part of it. By using different CNN architectures like InceptionV3, Xception, MobileNetV2, VGG16, VGG19 we have classified Alzheimer's patients from healthy subjects by calculating different model accuracy, confusion matrix and ROC curve from their MRI data. Among all the models, the basic CNN and the InceptionV3 provide the best accuracy up to 90.62%. This research shows us how different CNN architectures perform on our MRI data of Alzheimer's subjects and healthy subjects in case of classification and helps us to find the best models for the detection of Alzheimer's disease.