Using 3D-CAPSNET and RNN for Alzheimer’s Disease Detection Based on 4D fMRI

Abstract

An early prediction of Alzheimer’s disease (AD) progression can help slow down cognitive decline more effectively. Several studies have been devoted to applying different methods based on convolutional neural networks (CNNs) for automated AD diagnosis using resting-state functional magnetic resonance imaging (rs-fMRI). The methods introduced in these studies encounter two major challenges. First, fMRI datasets suffer from being of small size resulting in overfitting. Second, the 4D information of fMRI sessions needs to be efficiently modeled. Some of the studies applied their deep learning methods to functional connectivity matrices generated from fMRI data to model the 4D information, or to fMRI data as separate 2D slices or 3D volumes. However, this results in information loss in both types of methods. In this study, a new model based on Capsule network (CapsNet) and recurrent neural network (RNN) is proposed to model the spatiotemporal (4D) information of fMRI data for AD diagnosis. Experiments were conducted to evaluate the efficiency of the proposed model. According to the results, it has been observed that the proposed model could achieve 94.5% and 61.8% accuracy for the AD versus normal control (NC) and late mild cognitive impairment (lMCI) versus early mild cognitive impairment (eMCI) classification tasks, respectively.