A deep learning framework for identifying Alzheimer's disease using fMRI-based brain network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) diagnosis is challenging due to unclear pathological mechanisms.
• Previous research focused on brain network abnormalities or deep learning for AD classification.
• Convolutional Neural Networks (CNNs) are prominent deep learning algorithms for clinical diagnosis.

Purpose of the Study:

• To develop a novel framework combining brain topological features and CNN-extracted deep features for AD detection.
• To utilize resting-state functional magnetic resonance imaging (rs-fMRI) with Phase Synchronization Index (PSI) and 2D-CNN.
• To detect abnormal brain functional connectivity in AD patients.

Main Methods:

• Constructing brain networks using PSI from region of interest (ROI) signals and extracting eight topological features.
• Applying 2D-CNN to the PSI matrix to extract eight deep features, analyzing local and global network patterns.
• Performing classification using Support Vector Machine (SVM) with 5-fold cross-validation on combined features.

Main Results:

• The combined PSI and 2D-CNN framework achieved a classification accuracy of 98.869% for AD detection.
• The method effectively identified abnormal brain functional connectivity patterns.
• Extracted features characterized brain functional abnormalities in AD.

Conclusions:

• The proposed framework adaptively combines brain network features for enhanced AD detection.
• This approach offers new insights into the underlying pathogenesis of Alzheimer's disease.
• The framework effectively explores network synchronization and functional connections.