Nonparametric Aggregation of Geodesic Trends for Longitudinal Data Analysis.

Area of Science:

• Medical Imaging
• Biostatistics
• Neuroimaging

Background:

• Longitudinal imaging studies often have unbalanced and sparse data, with few measurements per individual over short periods.
• Existing methods may struggle to model individual trajectories and aggregate them into a representative group trend, especially for non-linear changes.

Purpose of the Study:

• To develop a novel technique for analyzing longitudinal imaging data, specifically addressing the challenges of sparsity and imbalance.
• To model individual changes using diffeomorphic geodesic regression and aggregate these into a nonparametric group average trend.

Main Methods:

• Utilized diffeomorphic geodesic regression to estimate individual longitudinal trends, suitable for short time windows and low sample sizes.
• Developed a novel nonparametric regression approach to aggregate individual geodesic trends into an average group trend, accommodating non-geodesic patterns.

Main Results:

• Successfully demonstrated the method's ability to capture non-geodesic group trends.
• Applied the technique to analyze hippocampal volume and perform diffeomorphic registration on 3D MRI data from the longitudinal OASIS dataset.

Conclusions:

• The proposed technique offers a robust framework for analyzing complex longitudinal imaging data, particularly in neuroimaging research.
• This approach effectively models individual variability and establishes a representative group trend, even with sparse and unbalanced datasets.