Alzheimer's Disease Diagnosis Based on Ensemble of Multi-model Convolutional Neural Networks

Alzheimer's disease (AD) is one of the most life-threatening diseases which is disrupting the everyday routine of elderly peoples in 21st century. Emerging researches have a rapid effect on improving the everyday life of Alzheimer's individuals. Clinicians and cognitive researchers perform several experiments in the early prediction of the disease. When the disease progresses to later stage, Alzheimer's can cause severe damage and structural changes to the patient's brain. Hence, this research studies those structural changes and predicts the stage of Alzheimer's disease at the early onset so that patient's brain can be saved from severe damage. This disease occurs in 3 stages, they are mild, moderate and severe cognitive impairment. Alzheimer disease does not affect people in one common way. Each and every person may undergo different symptoms and the symptoms differ in different stages of Alzheimer's. Usually during the mild stage, the person behaves normally as if he is not having many symptoms. Instead, they may experience few memory lapses in their daily activities. They may forget familiar words, their basic skills and their daily activities they often do and also forget the position of regularly using objects. This paper implements the deep learning Convolution Neural Network (CNN) algorithm that differentiates various stages of disease and the conversion probability from mild to advanced stage of disease. The existing systems had utilized machine learning algorithms by training using all the symptoms related features. The proposed approach extracts all the features and textures from the images of MRI scans and utilize in the hidden layers of Multi-Class Classifier. These layers use the pattern from the image features and classifies the disease stage to help in diagnosis at the early onset of symptoms. This makes the disease prediction accuracy higher and the model developed using CNN is effective with a prediction accuracy of 96%.