Uncertainty Quantification for Cardiac Diffusion Tensor Imaging Without Additional Datasets.

Area of Science:

• Medical Imaging
• Biophysics
• Cardiovascular Research

Background:

• Cardiac diffusion tensor imaging (cDTI) is susceptible to physiological and thermal noise, leading to errors in diffusion measurements.
• Existing methods for assessing precision involve dataset decimation, but fitting entire datasets requires specific uncertainty quantification (UQ) methods.
• Robust UQ is crucial for integrating accurate error estimation into cDTI post-processing pipelines.

Purpose of the Study:

• To develop and demonstrate uncertainty quantification (UQ) methods for cardiac diffusion tensor imaging (cDTI).
• To account for non-idealized errors inherent in cDTI data acquisition and processing.
• To enable the output of reliable uncertainty measures from cDTI post-processing.

Main Methods:

• Employed repetition bootstrap methods with whole-image resampling to approximate the sampling distribution of diffusion measures.
• Applied UQ to voxel-wise diffusion measures and myocardial summary statistics.
• Demonstrated methods on datasets from healthy volunteers and hypertrophic cardiomyopathy patients.

Main Results:

• Uncertainty weighting of myocardial averages significantly increased group differences for measures like Mean Diffusivity (MD), Fractional Anisotropy (FA), and |E2A|.
• The difference in group medians for |E2A| increased from 24.0° (unweighted) to 36.7° (uncertainty-weighted).
• Uncertainty maps effectively highlighted regions with less reliable diffusion measures and aided in identifying outlier cases.

Conclusions:

• Uncertainty quantification is feasible in cardiac diffusion tensor imaging using repetition bootstrapping.
• The proposed UQ methods provide a reliable way to assess the precision of diffusion measures.
• Successful application requires a cDTI dataset design suitable for repetition bootstrapping techniques.