Classification of Mild Cognitive Impairment by Fusing Neuroimaging and Gene Expression Data

Abstract

As reversing the pathology of Alzheimer's disease (AD) is impossible, the diagnosis of mild cognitive impairment (MCI), which is considered as the precursor of AD, has become a more tractable goal. Because both brain structural and functional alterations have been observed in MCI patients, many multimodal fusion approaches have been proposed to classify MCI from normal controls (NC) in clinical studies. Given the complex relationships between brain structure and function, deep learning based models can be helpful in revealing potential non-linear relationships buried in multimodal neuroimaging data. Meanwhile, RNA expression microarray profile can be a complementary feature in brain diseases analysis from another aspect, that is, the knowledge from molecular biology and genetics may benefit the classification of AD/MCI patients. To incorporate both imaging and molecular biomarkers, we propose a new deep fusion model: by integrating a cross-model deep network working on multi-modal brain image data and a fully-connected neural network working on gene expression data, a parameter representing the ratio of imaging and genetic features can be learned automatically during the classification process. Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, this method achieves an overall 82.3% accuracy, by fusing brain structural and functional connectivity as well as gene expression intensity information.