Parametric and non-parametric statistical analysis of DT-MRI data.

Area of Science:

• Medical Imaging
• Statistical Modeling
• Neuroscience

Background:

• Diffusion Tensor Magnetic Resonance Imaging (DT-MRI) generates complex data requiring robust statistical analysis.
• Existing methods may not fully capture the nuances of diffusion tensor data, especially in the presence of noise.
• Accurate statistical modeling is crucial for reliable interpretation of DT-MRI findings.

Purpose of the Study:

• To propose and evaluate parametric and non-parametric statistical methods for DT-MRI data analysis.
• To introduce a Multivariate Normal Distribution as a parametric model for diffusion tensor data.
• To develop and validate a novel non-parametric bootstrap methodology (DT-MRI bootstrap) for DT-MRI data.

Main Methods:

• Parametric approach: Modeled diffusion tensor data using a Multivariate Normal Distribution, assuming Johnson noise.
• Non-parametric approach: Implemented a DT-MRI bootstrap methodology for empirical probability distribution estimation and hypothesis testing.
• Validation: Utilized Monte Carlo (MC) simulations and applied methods to in vivo human brain and phantom DT-MRI data.

Main Results:

• The Multivariate Normal Distribution model was tested via MC simulations for its efficacy in parametric analysis.
• The DT-MRI bootstrap successfully estimated empirical distributions and performed hypothesis tests on DT-MRI data.
• The bootstrap method provided robust statistics of DT-MRI parameters within voxels and ROIs, and assessed intrinsic variability independent of noise.

Conclusions:

• Both proposed parametric and non-parametric statistical methods offer effective tools for analyzing DT-MRI data.
• The DT-MRI bootstrap is a valuable non-parametric approach for understanding DT-MRI data variability and performing hypothesis testing.
• The validated methods show promise for application in both research and clinical settings involving DT-MRI analysis.