MCI classification based on fusion of brain network properties and local cortex signal features

Machine learning and pattern recognition have been widely used in resting-state functional magnetic resonance imaging (rs-fMRI) data to investigate Alzheimer's disease (AD). However, many previous methods have not focused on the pre-Alzheimer's disease, mild cognitive impairment (MCI), have mostly classified features from functional separation or functional integration alone, which may have overlooked the correlation between the two. We propose a novel method for MCI diagnosis using the fusion local features of brain area signals as features for functional separation and brain network properties as features for functional integration. The rs-fMRI data of 43 MCI patients and 46 normal cognitive (NC) controls were analyzed using graph theory and nonlinear time series analysis to extract the brain network properties and local cortex signal features. The classifier-SVM achieved an accuracy of 92.8% based on the above features, which is generally higher than those of conventional methods based on functional integration using brain network properties alone or functional separation using local features alone. Our method demonstrates the measure of multi-feature integration of functional integration and functional separation as a powerful tool to classify MCI patients.