Multimodal Classification of Alzheimer's Disease Using Longitudinal Data Analysis and Hypergraph Regularized...

Area of Science:

• Neuroimaging
• Machine Learning
• Biomedical Data Analysis

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder impacting memory and cognition.
• Magnetic Resonance Imaging (MRI) is crucial for AD diagnosis, but current methods often overlook longitudinal data.
• Analyzing temporal changes in neuroimaging is vital for accurate AD monitoring and diagnosis.

Purpose of the Study:

• To develop a multi-task feature selection algorithm for Alzheimer's disease classification using longitudinal imaging and hypergraphs (THM2TFS).
• To improve the accuracy of Alzheimer's disease diagnosis by leveraging temporal dependencies in neuroimaging data.
• To identify key biomarkers associated with Alzheimer's disease progression.

Main Methods:

• A multi-task learning framework was established, treating feature selection at each time point as a separate task.
• Group sparse regularization, incorporating hypergraph-induced and fused sparse Laplacian regularization, was used to model subject relationships and temporal changes.
• Multi-kernel Support Vector Machines (SVM) integrated selected features for final classification.
• Functional MRI and structural MRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) across four time points were utilized.

Main Results:

• The THM2TFS method achieved high classification accuracies: 96.75% (AD vs. NC), 93.45% (MCI vs. NC), and 83.78% (AD vs. MCI).
• The algorithm effectively captured relevant information from longitudinal imaging data.
• The proposed method demonstrated improved classification accuracy compared to existing approaches.

Conclusions:

• The THM2TFS algorithm offers a robust approach for Alzheimer's disease classification using longitudinal neuroimaging data.
• The method enhances diagnostic accuracy and aids in identifying critical biomarkers for Alzheimer's disease.
• This work highlights the importance of incorporating temporal dynamics in neuroimaging analysis for neurodegenerative disease research.