On the averaging of cardiac diffusion tensor MRI data: the effect of distance function selection

Area of Science:

• Medical Imaging
• Biophysics
• Computational Biology

Background:

• Diffusion tensor magnetic resonance imaging (DT-MRI) offers insights into tissue microstructure.
• Selecting appropriate distance functions for diffusion tensors is critical for clinical applications.
• Existing research debates the optimal DT-MRI distance metric.

Purpose of the Study:

• To compare Euclidean, affine-invariant Riemannian, and log-Euclidean metrics for DT-MRI data.
• To evaluate these metrics using high-resolution DT-MRI rat heart data.
• To introduce a novel framework for quantitative comparison of tensor distance functions.

Main Methods:

• Temporal averaging of consecutive DT-MRI datasets to reduce noise.
• Application of three distinct tensor distance metrics: Euclidean, Riemannian, and log-Euclidean.
• Quantitative, unbiased, and longitudinal (same-voxel) comparison of metric performance.

Main Results:

• Statistical analysis indicated that the three DT-MRI distance functions produced largely equivalent results.
• The tensor manifold for cardiac DT-MRI studies exhibits near-zero curvature.
• The 'swelling effect' from Euclidean averaging was found to be negligible.

Conclusions:

• The choice of distance function has minimal impact on cardiac DT-MRI analysis.
• Cardiac tensor spaces can be approximated as flat (zero curvature).
• Noise reduction techniques enhance the reliability of tensor comparisons.