Multi-modal imaging genetics data fusion by deep auto-encoder and self-representation network for Alzheimer's disease diagnosis and biomarkers extraction

Abstract

Alzheimer's disease (AD) is an incurable neurodegenerative disease, so it is important to intervene in the early stage of the disease. Brain imaging genetics is an effective technique to identify AD-related biomarkers, which can early diagnosis of AD patients once they are clinically verified. With the development of medical imaging and gene sequencing techniques, the association analysis between multi-modal imaging data and genetic data has garnered increasing attention. However, current imaging genetics studies have problem with non-intuitive data fusion. Meanwhile, the characteristics of multi-modal imaging genetics data are high-dimensional, non-linearity, and fewer subjects, so it is necessary to select effective features. In this paper, a multi-modal data fusion framework by deep auto-encoder and self-representation (MFASN) was proposed for early diagnosis of AD. First, a multi-modality brain network was constructed by combining information from the resting-state functional magnetic resonance imaging (fMRI) data and structural magnetic resonance imaging (sMRI) data. Then, we utilized the deep auto-encoder to achieve non-linear transformations and select the informative features. A sparse self-representation module was employed to capture the multi-subspaces structure of the latent representation. At last, a multi-task structured sparse association model was developed to fully mine correlations between the genetic data and multi-modal brain network features. Experiments on AD neuroimaging initiative datasets proved the superiority of the proposed method, while discovering discriminative biomarkers were strongly associated with AD.