Toward quantitative fast diffusion kurtosis imaging with b‐values chosen in consideration of signal‐to‐noise ratio and model fidelity

Diffusion kurtosis (DK) imaging is a variant of conventional diffusion magnetic resonance (MR) imaging that allows assessment of non-Gaussian diffusion. Fast DK imaging expedites the procedure by decreasing both scan time (acquiring the minimally required number of b-values) and computation time (obviating least-square curve fitting). This study aimed to investigate the applicability of fast DK imaging for both cerebral gray matter and white matter as a quantitative method.Seventeen healthy volunteers were recruited and each provided written informed consent before participation. On a 3-Tesla clinical MR system, diffusion imaging was performed with 12 b-values ranging from 0 to 4000 s/mm2 . Diffusion encoding was along three orthogonal directions (slice selection, phase encoding, and frequency encoding) in separate series. Candidate b-values were chosen by first determining the maximum b-value (bmax ) in the context of signal-to-noise ratio and then assessing the model fidelity for all b-value combinations within bmax . Diffusion coefficient (D) and diffusion kurtosis coefficient (K) were derived from these candidates and assessed for their dependence on b-value combination.Our data suggested bmax to be 2200 s/mm2 as a trade-off between the percentage (~80%) of voxels statistically detectable against background and the sensitivity to non-Gaussian diffusion in both gray matter and white matter. The measurement dependence on b-value was observed predominantly in areas with a considerable amount of cerebrospinal fluid. In most gray matter and white matter, b-value combinations do not cause statistical difference in the calculated D and K.For fast DK imaging to be quantitatively applicable in both gray matter and white matter, bmax should be chosen to ensure adequate signal-to-noise ratio in the majority of gray/white matter and the two nonzero b-values should be chosen in consideration of model fidelity to mitigate the dependence of derived indices on b-values.