Identification of mild cognitive impairment using multimodal 3D imaging data and graph convolutional networks

Area of Science:

• Neuroimaging and Artificial Intelligence
• Computational Neuroscience
• Medical Diagnostics

Background:

• Mild cognitive impairment (MCI) is a prodromal stage of dementia and Alzheimer's disease (AD), necessitating early identification for timely intervention.
• Accurate diagnosis of MCI is challenging due to subtle cognitive declines and the need to integrate diverse data sources.

Purpose of the Study:

• To develop and evaluate a novel multimodal 3D imaging data integration graph convolutional network (GCN) model for accurate MCI identification.
• To assess the model's performance in classifying early MCI (EMCI) and late MCI (LMCI) against normal controls (NC).

Main Methods:

• Utilized 3D-VGGNet for feature extraction from multimodal imaging data (sMRI, FDG-PET).
• Constructed a population sparse graph integrating imaging and non-imaging features, optimizing connectivity with pairwise attribute estimation (PAE).
• Employed a population-based GCN to fuse multimodal features for MCI classification.

Main Results:

• Achieved high classification accuracies: 98.57% (NC-EMCI), 96.03% (NC-LMCI), and 96.83% (EMCI-LMCI) on the AD Neuroimaging Initiative dataset.
• Demonstrated superior performance over state-of-the-art models in AUC, specificity, sensitivity, and F1-score.
• Showcased excellent generalization capabilities with 91.43% accuracy for autism diagnosis on the ABIDE dataset.

Conclusions:

• The proposed GCN model effectively integrates multimodal imaging data for robust MCI recognition.
• This approach holds significant potential for early warning of AD and intelligent diagnosis of other neurodegenerative diseases.
• The model's strong performance and generalization highlight its clinical utility in brain disorder diagnostics.