Hybrid Deep Learning Approach for Classifying Alzheimer Disease Based on Multimodal Data

Abstract

Alzheimer’s disease (AD) is a category of dementia that is difficult to identify under clinical supervision. Currently, there is no remedy for AD, but its initial indication is essential for effective treatment. AD causes memory, thinking, and hence behavior problems. AD symptoms usually develop gradually and become worse from time to time, which can interfere with daily activities. Traditional machine learning algorithms do AD classification usually based on only single input that is the brain’s magnetic resonance imaging (MRI) inspection. The proposed hybrid deep neural network classifies according to multimodal data in the form of MRI images and EEG signals. The hybrid method is to model the behavior of the time-watch and use the model to select the most interesting features from multimodal data. The key objective of this method is to enhance learning procedure in which the weight factor of DNN is incorporated with CNN for dealing with multimodal heterogeneous information. This paper describes the study related to how the hybrid classifier’s accuracy depends on (the number of features). As the number of features increase, the classification error decreases resulting in improving the accuracy of the classifier. Furthermore, other more traditional methods based on correlation measures and mutual information are also compared with the proposed approach. Experimental results show that the proposed approach categorization accuracy is better than other classification methods.