Robust estimation with convexity constraints for improved diffusion kurtosis imaging of the human heart in vivo

Motivation: Fitting signal representations to multi-shell diffusion data may require constraints to avoid violating principles such as diffusion propagator non-negativity. However, constrained fitting is not intrinsically robust to signal corruptions, which are common in cardiac diffusion weighted imaging. Goal(s): We sought to determine how convexity constraints and robust estimation affect cardiac Diffusion Kurtosis Imaging (cDKI) measures. Approach: We developed fitting methods combining robust estimation with convexity constraints. We applied these methods to healthy volunteer data from a Connectom scanner. Results: Constrained fitting improved kurtosis measures and gave more physiologically plausible results compared to unconstrained fitting. Robust constrained fitting yielded the best results overall. Impact: Robust estimation has now been combined with fitting constraints, allowing for reliable estimation of kurtosis measures from in vivo multi-shell cardiac diffusion MRI data. This approach will improve fidelity of kurtosis quantification in the human heart in vivo.