A Novel Multimodal MRI Analysis for Alzheimer's Disease Based on Convolutional Neural Network

Recent years, Alzheimer's disease (AD) has become a significant threat to human health while the accurate screening and diagnosis of AD remain a tough problem. Multimodal Magnetic resonance imaging (MRI) can help to identify the variation of brain function and structure in a non-invasive way. Deep learning, especially the convolutional neural networks (CNN), can be utilized to automatically detect appropriate features for classification, which is well adapted for computer-aided AD screening and identification. This paper proposed a multimodal MRI analytical method based on CNN, which is also suitable for single type MRI data analysis. First, the human brain network connectivity matrix were extracted from multimodal MRI data, used as the input data for CNN. Then a novel CNN framework was proposed to process the network matrix and classify AD, amnestic mild cognitive impairment (aMCI) patients and normal controls (NC). The advantage of this method lies in that we combined multimodal MRI information through CNN convolution kernel, and achieved a higher classification accuracy. In our experiments, the comprehensive classification accuracy of AD, aMCI patients and NC was as high as 92.06% when using multimodal MRI data as input, which is effective enough to provide a reference for multimodal MRI data analysis.