Image Data Mining for Extracting Relations between Alzheimer’s-related Gene Types and Cerebral Atrophy Patterns using Principal Component Regression

Abstract

Genome-wide association studies have discovered many genetic variants that are highly associated with specific diseases. The purpose of this research is to realize a personalized medical examination that combines an individual’s genetic type and imaging tests. From a public database Alzheimer’s Disease Neuroimaging Initiative (ADNI), we collected 129 magnetic resonance (MR) images and the type of Alzheimer’s related gene APOE. After brain morphology standardization using Statistical Parametric Mapping (SPM), principal component analysis was performed using only 60 normal cases of different ages, and brain atrophy due to normal aging was modeled on the eigen-space. We then projected 69 Alzheimer’s Disease (AD) cases onto the obtained eigenspace and used the principal component regression analysis to examine whether APOEε3 and APOEε4 cases were distributed in different locations. ROC analysis for distinguishing between APOEε3 and APOEε4 showed that the AUC was 0.85. Therefore, it was concluded that even if brain atrophy due to normal aging is taken into account, the pattern of brain atrophy in AD differs depending on the individual’s genetic type. By interpreting images that take into account individual genetic differences, it allows for early detection of the disease.