Longitudinal functional principal component analysis.

Area of Science:

• Statistics
• Functional Data Analysis
• Neuroimaging Analysis

Background:

• Analyzing longitudinal functional data requires sophisticated statistical methods.
• Understanding dynamic changes in subject-specific variability is crucial in many scientific fields.
• Existing models may not fully capture the complexity of functional data observed over multiple time points.

Purpose of the Study:

• To introduce novel statistical models for analyzing functional data with multiple time points.
• To decompose the dynamic behavior of functional data into various components of variability.
• To apply these models to neuroimaging data for studying brain connectivity in healthy and diseased populations.

Main Methods:

• Development of functional mixed-effects models, replacing classical random effects with random processes.
• Utilizing principal component bases for functional processes to ensure computational feasibility for large datasets.
• Application to Diffusion Tensor Imaging (DTI) data to analyze brain connectivity.

Main Results:

• The proposed models effectively decompose functional data into population average and various subject-specific variability components.
• Computational feasibility is demonstrated for moderate to large datasets.
• The methodology is successfully applied to DTI data, enabling analysis of brain connectivity differences.

Conclusions:

• The introduced functional mixed-effects models offer a powerful and computationally feasible approach for analyzing longitudinal functional data.
• These models provide insights into dynamic changes in subject-specific variability.
• The application to DTI data highlights the potential for advancing neuroimaging research, particularly in understanding neurological conditions like multiple sclerosis (MS).