Optimizing cardiac diffusion tensor imaging in vivo: More directions or repetitions?

Cardiac diffusion tensor imaging (cDTI) is sensitive to imaging parameters, including the number of unique diffusion encoding directions (ND) and number of repetitions (NR; analogous to number of signal averages). However, there is no clear guidance for optimizing these parameters in the clinical setting.

Spin echo cDTI data with second-order motion-compensated diffusion encoding gradients were acquired in 10 healthy volunteers on a 3T magnetic resonance imaging scanner with different diffusion encoding schemes in pseudo-randomized order. The data were subsampled to yield 96 acquisition schemes with 6 ≤ ND ≤ 30 and 33 ≤ total number of acquisitions (NA_all) ≤ 180. Stratified bootstrapping with robust fitting was performed to assess the accuracy and precision of each acquisition scheme. This was quantified across a mid-ventricular short-axis slice in terms of root mean squared difference (RMSD), with respect to the full reference dataset, and standard deviation (SD) across bootstrap samples, respectively.

For the same acquisition time, the ND = 30 schemes had on average 48%, 40%, 34%, and 34% lower RMSD and 6.2%, 7.4%, 10%, and 5.6% lower SD in mean diffusivity (MD), fractional anisotropy (FA), helix angle (HA), and absolute sheetlet angle (|E2A|) compared to the ND = 6 schemes. Given a fixed number of high b-value acquisitions, there was a trend toward lower RMSD and SD of MD and FA with increasing numbers of low b-value acquisitions. Higher NA_all with longer acquisition times led to improved accuracy in all metrics, whereby quadrupling NA_all from 40 to 160 volumes led to a 20%, 39%, 11%, and 5.4% reduction in RMSD of MD, FA, HA, and |E2A|, respectively, averaged across six diffusion encoding schemes. Precision was also improved with a corresponding 53%, 50%, 53%, and 36% reduction in SD.

We observed that accuracy and precision were enhanced by (i) prioritizing number of diffusion encoding directions over NR given a fixed acquisition time, (ii) acquiring sufficient low b-value data, and (iii) using longer protocols where feasible. For clinically relevant protocols, our findings support the use of ND = 30 and NA_b50:NA_b500 ≥ 1/3 for better accuracy and precision in cDTI parameters. These findings are intended to help guide protocol optimization for harmonization of cDTI.