Analysis of Mild Cognitive Impairment Utilizing Covariance Matrices of Brain Regions

Mild Cognitive Impairment (MCI) represents a prevalent condition which impacts cognitive function and is often regarded as a precursor to Alzheimer’s disease. Resting-state functional magnetic resonance imaging (fMRI) has shown changes in brain activity patterns and connectivity between brain regions in individuals with MCI compared to healthy individuals. This study aims to classify MCI patients and healthy controls by analyzing the variations in inter-dependencies among various brain regions from six different brain networks. To address the challenge of dealing with multiple time series from each region, we use regional homogeneity (ReHo) maps to extract representative time series for each region. Covariance matrices derived from these time series are utilized as features for classification. Our dataset consists of resting-state fMRI scans from 50 MCI patients and 50 healthy controls obtained from publicly available Alzheimer’s disease neuroimaging initiative database (ADNI). We identify significant differences in the covariance matrices of several brain regions, particularly the regions in default mode and frontoparietal networks. Tangent vectors derived through Riemann geometry (RG-TS) are found to yield the highest accuracy across all methods of analysis (93.15%), utilizing regions chosen from the default mode network. These findings suggest that the covariance matrices of brain regions, especially those in the default mode network, hold the potential as a valuable tool in the early detection and monitoring of MCI.