1353057 - Diffusion Direction Orientation Considerations for Cardiac Diffusion Tensor Imaging

Background: Cardiac diffusion-tensor imaging (cDTI) provides metrics of myocardial microstructure including Mean Diffusivity (MD) and Fractional Anisotropy (FA). [1,2] cDTI operates in a low signal-to-noise (SNR) regime owing to both the prolonged echo times needed to accommodate motion compensated diffusion encoding and the signal attenuating effects of diffusion. How to balance the number of encoding directions plus signal averages and how to select the orientation of the directions remain open questions. We hypothesize that acquiring more diffusion directions while balancing total acquisition time with signal averaging yields more accurate DTI metrics. We also hypothesize that encoding along the scanner coordinate (XYZ) versus patient-specific coordinates (PRS) leads to more accurate cDTI metrics because encoding directions may be less aligned with the axis of largest cardiac motion.

Methods:

Methods:

Healthy volunteers (n=5, 25.2±3.7 years) were imaged (IRB approved, consented) at 3T (Skyra, Siemens) to obtain three (basal, mid-ventricular, and apical) slices. We used an ECG-gated, second-order motion compensated spin-echo EPI sequence [3] (2×2×8mm³
resolution, TE=95, TR=3×R-R intervals, PF=6/8, GRAPPA=2). The following diffusion directions and average combinations were acquired: (1) 3 directions with 10 averages (3x10); (2) 6x5; and (3)10x3. A ground-truth dataset with 30 directions with 10 averages was also collected (30x10). The protocol was repeated with directions prescribed along XYZ and PRS coordinates (Fig. 1). Post-processing included Gibb’s ringing removal, affine registration, acquisition rejection, respiratory sorting, averaging, denoising and filtered zero-filled interpolation. MD and FA were computed in the LV. Median and 95% Confidence Interval (CI) are reported and paired t-tests evaluated for both hypotheses with Bonferroni correction (p < 0.05).

Results: For our first hypothesis, the MD and FA appeared smoother with increasing number of directions and reduced averages (Fig. 2). There was a significant decrease in median MD from 6x5 to 30x10 (p < 0.001) and 10x3 to 30x10 (p=0.002).  There was also a significant decrease in FA from 6x5 to 30x10 (p < 0.001) and 10x5 to 30x10 (p < 0.001) (Fig. 3). For our second hypothesis, visually there were no observable differences between PRS and XYZ for DTI maps (Fig. 2). There was a significant decrease in FA between XYZ and PRS (p=0.01) (Fig. 3).

Conclusion: We conclude that acquiring more diffusion directions tends to produce DTI metrics closer to ground truth, suggesting that our first hypothesis is correct. This effect is less clear in MD but observed in FA. We also conclude that if acquiring fewer directions, then directions distributed off-axis in the PRS coordinates yield DTI metrics closer to ground-truth, so the second hypothesis is not true. Overall, this study demonstrates the need to consider directions, averages, and orientation when designing DTI protocols.

Funding: NSF DGE-1656518, NIH R01 HL152256