In this work, we implement a velocity and acceleration compensated spin-echo (SE) cardiac diffusion tensor imaging (cDTI) sequence^1,2,3 on a clinical 3T scanner with standard gradients and show initial comparisons with a stimulated echo (STEAM)^4,5 sequence at multiple points in the cardiac cycle. We also correlate parameter differences between sequences with strain, T1 and T2.A SE EPI cDTI sequence was implemented with 0^th, 1^st and 2^nd order motion-compensated diffusion gradients (M012)^1,2. Mid-ventricular short-axis cDTI was performed in 20 healthy volunteers on a 3T Siemens Skyra (gradients 45mT/m@200T/m/s per axis) with M012-SE and STEAM⁵. Acquisitions were performed at end-systole, end-diastole and the sweet-spot. The sweet-spot is the trigger-time at which strain effects in STEAM should be eliminated⁶ and DENSE analysis from 13 subjects found an average of 150ms from the R-wave. Time from R-wave to diffusion-encoding was matched between sequences. Optimal parameters were used for both sequences^2,7. M012 acquisitions used b_main=450smm^-2, TE=73ms and water-selective excitation. STEAM acquisitions used b_main=800smm^-2, TE=23ms and fat saturation. Both acquisitions used 6 diffusion directions, b_ref=150smm^-2, 6 averages, TR=2RR-intervals, reduced phase field-of-view, 360x135x8mm³ at 2.8x2.8mm² resolution, SENSE x2 and an identical EPI echo train. Each breath-hold was 20RR for both sequences. Since STEAM requires 2RR for diffusion-encoding, M012-SE was triggered to alternate R-waves. Strain information was obtained from a breath-hold 2D spiral cine DENSE acquisition⁸ in the same plane as cDTI. T1 and T2 maps were acquired in the same plane using a 5(3)3 modified Look-Locker sequence⁹ and a T2-prep based 3-point method¹⁰ respectively.cDTI data were processed using in-house software to produce pixel-wise maps of mean diffusivity (MD), fractional anisotropy (FA), helix angle (HA), tensor mode (mode), signal to noise ratio (SNR)⁵ and absolute second eigenvector angle (E2A)¹¹. Circumferential and radial strain data was obtained from the DENSE data using the University of Virginia DENSE analysis software. Mean T1 and T2 values were calculated from a septal region of interest.Global left ventricular cDTI parameters were compared between sequences using a Wilcoxon signed-rank test (p<0.01). To test the influence of strain, Pearson’s correlation coefficient was calculated between systolic – diastolic cDTI parameters and peak radial and circumferential strain. The influence of T1 and T2 weighting was investigated by correlating differences in cDTI parameters between sequences with T1 and T2 values.cDTI was performed successfully in all subjects and time points using STEAM. Processed M012-SE data was considered insufficient quality for further analysis in 2/20 systolic, 5/20 sweet-spot and 6/20 diastolic acquisitions due to bulk-motion related signal loss. Figure 1 shows examples of good cDTI parameter maps from one subject at all time points acquired using both sequences. Figure 2 shows examples of sweet-spot parameter maps in one subject where the M012-SE maps are poor. Global cDTI parameters are plotted in figure 3. There were significant differences between M012-SE and STEAM in MD, FA, E2A and Helical Angle gradient (HAg, °/mm) at all time points and SNR in diastole. Few of the correlations were significant (see table for all p<0.05) and only systolic FA (STEAM – M012-SE) vs. T1 reached p<0.01.

Previous studies using M012-SE have used high performance gradient systems^1,2,3, but we have demonstrated the feasibility of M012-SE cDTI using standard clinical gradients with a similar TE. While M012-SE can be performed at points throughout the cardiac cycle in most subjects, STEAM was successful in all subjects. This is due to the short diffusion gradients used with STEAM and despite the motion compensated design of M012-SE, failure was likely due to uncompensated cardiac motion during the long diffusion-encoding gradients. This also results in a long TE when using M012-SE and hence an increased T2 sensitivity. Differences between sequences in systolic SNR and FA, and diastolic MD were found to significantly correlate with T2.

STEAM cDTI is usually considered to have poor SNR efficiency and to be affected by strain⁴. However, few differences in cDTI parameters between systole and diastole demonstrated a significant correlation with peak strains and we did not observe increased SNR when using M012-SE, but did find higher STEAM SNR in diastole.

These preliminary results demonstrate that there are systematic differences between cDTI parameters derived from M012-SE and STEAM. While strain, T1 and T2 partially account for some differences, the factor 50 difference in mixing-time between sequences (~20ms M012-SE vs. ~1s STEAM) is likely to be responsible for the majority (figure 4). This is consistent with the higher FA and lower MD found using STEAM.

Future work will focus on a comprehensive analysis of these techniques, their relative advantages and suitable applications.