The Alzheimer's Disease continuum Supervised Classification Using the Merged Connectome of Metabolic and Structural Imaging

Abstract

Previous connectome research about glucose metabolic network, obtained with 18F-FDG PET data by sparse inverse covariance estimation (SICE), which has been revealed in neurodegenerative diseases. However, this metabolic network construction suffers from the lack of robustness with the metabolic connection estimation. Metabolic connection matrices have been observed to present similar patterns as the structural connection matrices obtained from diffusion MRI. Further, we aim to use the structural connectivity regularize the sparse estimation of metabolic connection. The merged connectome of metabolic and structure imaging based on FDG PET and diffusion MRI imaging is employed to measure the metabolic connection. The proposed method is then applied in a clinical dataset including healthy subjects and Alzheimer's disease continuum patients. The merged results had proved the improvement in the accuracy of the estimated metabolic connection network (Accuracy: 0.94 [HC-AD], 0.89 [HC-MCI], 0.83 [MCI-AD]). Compared to standard SICE, the structural weighting has shown more stable performance in the supervised classification.