Integrating Time and Frequency Domain Features of fMRI Time Series for Alzheimer's Disease Classification Using Graph...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Accurate Alzheimer's Disease (AD) diagnosis is critical for effective treatment.
• Functional Magnetic Resonance Imaging (fMRI) is a key tool, but current methods using time-domain features are noise-susceptible and neglect asynchronous brain interactions.
• Existing methods focus on synchronous brain-region interactions, limiting diagnostic accuracy.

Purpose of the Study:

• To develop a robust AD diagnosis method by integrating frequency- and time-domain fMRI data.
• To overcome limitations of current methods by considering both asynchronous and synchronous brain-region interactions.
• To improve the accuracy and earliness of Alzheimer's Disease detection using advanced neural network techniques.

Main Methods:

• Proposed the Frequency-Time Fusion Graph Neural Network (FTF-GNN) model.
• Utilized Discrete Fourier Transform (DFT) for frequency-domain analysis and Fourier-based Graph Neural Network (FourierGNN) for asynchronous connectivity.
• Employed a Graph Convolutional Network (GCN) to capture synchronous connectivity patterns, fusing both domains for classification.

Main Results:

• Achieved 91.26% accuracy and 96.79% AUC for AD vs. Normal Control (NC) classification.
• Demonstrated state-of-the-art performance in distinguishing NC from Late Mild Cognitive Impairment (LMCI) (87.16% accuracy, 93.22% AUC).
• Showcased optimal performance in differentiating LMCI from AD (85.30% accuracy, 94.56% AUC) and competitive results for Early MCI (EMCI) vs. LMCI classification (77.40% accuracy, 91.17% AUC).

Conclusions:

• The FTF-GNN model effectively leverages complementary frequency- and time-domain information for robust AD diagnosis.
• Considering both asynchronous and synchronous brain-region interactions addresses limitations of existing neuroimaging approaches.
• The proposed method offers a promising solution for accurate and early detection of Alzheimer's Disease and its subtypes.