MLC-GCN: Multi-Level Connectomes Based GCN for AD Detection

Motivation: Alzheimer's Disease (AD) is characterized by progressive cognitive impairments that are related to alterations in brain functional connectivity (FC). Goal(s): to design a graph convolutional network (GCN) based classifier to differentiate AD from old cognitive normal controls. Approach: We assessed the FC using Pearson correlation coefficient (CC) and cross entropy (CE) measure as association analysis and proposed a multi-level generated connectome (MLC) based GCN (MLC-GCN) containing a multi-graph generation block and a GCN prediction block to classify the fMRI data. Results: Our method showed better performance than state-of-the-art GCN and rsfMRI based AD classifiers on two independent public medical datasets: ADNI and OASIS-3. Impact: The MLC-GCN classifier significantly enhances Alzheimer's disease detection by exploiting multi-level connectomes. The clinically meaningful classifier features suggest a potential of localizing disease-related nodes or regions, facilitating clinical diagnosis and future targeted interventions.