Identification of early mild cognitive impairment using multi-modal data and graph convolutional networks

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early mild cognitive impairment (EMCI) is an early stage of Alzheimer's disease (AD), characterized by brain changes.
• Identifying EMCI is challenging due to the complexity of structural and functional brain alterations.
• Combining multiple features like grey matter volume and shortest path length shows promise for improved EMCI detection.

Purpose of the Study:

• To develop a novel framework for accurate identification of early mild cognitive impairment (EMCI).
• To leverage multi-modal data and graph convolutional networks (GCNs) for enhanced EMCI diagnosis.
• To address the challenge of selecting and combining optimal features for EMCI identification.

Main Methods:

• Extracted grey matter volume (T1w MRI) and shortest path length (rs-fMRI) using the AAL atlas.
• Applied multi-task feature selection to identify informative features for EMCI detection.
• Constructed a non-fully labelled subject graph incorporating imaging and non-imaging data.
• Utilized a GCN model for the EMCI identification task.

Main Results:

• Evaluated on 210 subjects (105 EMCI, 105 normal controls) from the ADNI database.
• Achieved 84.1% accuracy and 0.856 AUC for EMCI/NC classification.
• Demonstrated superior performance compared to existing EMCI identification methods.

Conclusions:

• The proposed framework is effective for automatic EMCI diagnosis.
• The GCN-based approach shows significant promise for clinical application in identifying early Alzheimer's disease stages.