Alzheimer's disease classification based on combination of multi-model convolutional networks

Abstract

Alzheimer's disease (AD) is an irreversible neurodegenerative disorder with progressive impairment of memory and cognitive functions. It brings increasingly troubling in today's society. Its early diagnosis and warning are particularly important. Structural magnetic resonance images (MRI) play an important role to help understand the brain anatomical changes related to AD. Traditional methods are usually based on extraction of handcrafted features and training a classifier to distinguish AD from other groups. Motivated by the success of deep learning in image classification, this paper proposes a classification method based on combination of multimodel 3D convolutional networks to learn the various features from MR brain images. First, a deep 3D convolutional neural network (3D CNN) is built to hierarchically transform the MR image into more compact high-level features. Second, the multiscale 3D convolutional autoencoders (3D CAEs) are constructed to extract features from MR brain images. The features learned by these models are combined with the upper fully connected layers for image classification in AD diagnosis. The proposed method can automatically learn the generic features from the imaging data for classification without segmentations of brain tissues and regions. Our method is evaluated using T1-weighted MR brain images on 428 subjects including 199 AD patients and 229 normal controls (NC) from Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Experimental results show that the proposed method achieves an accuracy of 88.31% for AD classification and an AUC (area under the ROC curve) of 92.73%, demonstrating the promising classification performances.