Evaluating High Spatial Resolution Diffusion Kurtosis Imaging at 3T: Reproducibility and Quality of Fit

ABSTRACT Background Diffusion kurtosis imaging (DKI) quantifies the microstructure’s non-Gaussian diffusion properties. However, it has increased fitting parameters and requires higher b-values. Evaluation of DKI reproducibility is important for clinical purposes. Purpose To assess reproducibility in whole-brain high resolution DKI at varying b-values. Study Type Prospective. Subjects and Phantoms Forty-four individuals from the test-retest Human Connectome Project (HCP) database and twelve 3D-printed tissue mimicking phantoms. Field Strength/Sequence Multiband echo-planar imaging for in vivo and phantom diffusion-weighted imaging at 3T and 9.4T respectively. MPRAGE at 3T for in vivo structural data. Assessment From HCP data with b-value =1000,2000,3000 s/mm 2 (dataset A), two additional datasets with b-values=1000, 3000 s/mm 2 (dataset B) and b-values=1000, 2000 s/mm 2 (dataset C) were extracted. Estimated DKI metrics from each dataset were used for evaluating reproducibility and fitting quality in whole-brain white matter (WM), region of interest (ROI) and gray matter (GM). Statistical Tests DKI reproducibility was assessed using the within-subject coefficient of variation (CoV), fitting residuals to evaluate DKI fitting accuracy and Pearson’s correlation to investigate presence of systematic biases. Results Compared to dataset C, the CoV from DKI parameters from datasets A and B were comparable, with WM and GM CoVs <20%, while differences between datasets were smaller for the DKI-derived DTI parameters. Slightly higher fitting residuals were observed in dataset C compared to A and B, but lower residuals in dataset B were detected for the WM ROIs. A similar trend was observed for the phantom data with comparable CoVs at varying fiber orientations for datasets A and B. In addition, dataset C was characterized by higher residuals across the different fiber crossings. Data Conclusion The comparable reproducibility of DKI maps between datasets A and B observed in the in vivo and phantom data indicates that high reproducibility can still be achieved within a reasonable scan time, supporting DKI for clinical purposes. HIGHLIGHTS: Reproducibility and fitting accuracy of high resolution DKI were evaluated as function of available b-values. A DKI dataset with b-values of 1000 and 3000 s/mm 2 performs equally well as the original HCP three-shell dataset, while a dataset with b-values of 1000 and 2000 s/mm 2 has lower reproducibility and fitting quality. In vivo results were verified using phantoms capable of mimicking different white matter configurations. These results suggest that DKI data can be obtained within less time, without sacrificing data quality.