Diffusion Tensor Imaging (DTI) of the heart in vivo has previously been performed using multiple breath-holds or respiratory navigation, providing a subset of short-axis slices of the left ventricle (LV), comprising ~25% of the LV in about 20 minutes (1,2). Here we demonstrate a clinically feasible whole heart free-breathing DTI approach, with scan time under 15 minutes, based on a simultaneous multi-slice (SMS) acquisition technique combined with a retrospective image alignment and selection method. This approach avoids navigator echoes and synchronized breathing and enables anatomic coverage of the entire LV.

DTI was performed in healthy volunteers (n=7) on a clinical 3T scanner (Skyra, Siemens, Erlangen, Germany) equipped with a 45 mT/m gradient system and a 34-element cardiac receive coil. Imaging was performed with an ECG-gated diffusion-encoded STE sequence, volume selected in the phase-encode axis using a slab selective radiofrequency (RF) pulse. Acquisition parameters were: FOV = 360 x 200 mm², resolution 2.5 x 2.5 mm², slice thickness = 8 mm, in-plane GRAPPA rate 2, TE = 34 ms, b-values = 0 and 500 s/mm², 10 diffusion-encoding directions, and 8 magnitude averages (repetitions). Twelve short-axis and six 4-chamber slices were acquired in the systolic sweet spot of the cardiac cycle to mitigate strain effects (3). SMS excitation was followed by a blipped-CAIPI (Controlled Aliasing in Parallel Imaging) readout (4). Free-breathing scans were performed with rate 2 SMS and compared with breath-hold scans acquired with rate 3 SMS.

Breath-hold DTI was performed with the default interleaved gradient ordering scheme. For free-breathing DTI a sequential gradient ordering scheme was implemented, which ensured that enough uncorrupted samples of each diffusion-encoding direction could be selected to determine the diffusion tensor accurately. Spatiotemporal registration (STR) was performed by matching radial intensity profiles over all repetitions, and reduced the misregistration resulting from respiratory motion. With free breathing, following STR, an entropy-based retrospective image selection (ERIS) was applied to accept or reject both the diffusion-free and diffusion-weighted images. An entropy measure (E_M), based on Shannon’s definition of entropy, was calculated from the distribution of signal intensity within the myocardium, across repetitions (5). Images with E_M within a prescribed range E_R were accepted, while those adversely affected by artifact were rejected.

For each voxel, we computed a dyadic tensor comprising the primary, secondary, and tertiary eigenvectors and their associated eigenvalues. Mean diffusivity (MD), fractional anisotropy (FA) and myofiber helix angle (HA) were calculated, as previously described (6). Tractography was performed by numerically integrating the primary eigenvector field into streamlines using an adaptive 5^th order Runge-Kutta approach (6). Comparisons between the breath-hold and free-breathing data were performed using paired t-tests.

Combining STR and ERIS yielded the best quality of MD, FA, and HA maps during free breathing (Fig 1). For a mid-ventricular slice, the entropy measure E_M varied between 6.39 and 7.17, rejecting images with E_M less than 6.54, corresponding to the bottom 20% of E_M values. DTI and tractography of the entire heart in both the short-axis and 4-chamber views could be successfully performed with the blipped-CAIPI SMS approach during both breath-hold and free-breathing (Fig 2). Each approach produced similar tractograms. Using STR and ERIS, no significant differences were found in MD (0.9 ± 0.09, 0.89 ± 0.08 ×10-3 mm²/s), FA (0.42 ± 0.06, 0.42 ± 0.05) or HA profiles between the free-breathing and breath-hold scans (Fig 3).Clinical translation of cardiac diffusion imaging has proven technically challenging. Previous studies have required multiple breath-holds or synchronized breathing (3,7), with partial anatomic coverage of the LV and comparatively long scan times (1,2), even with accelerated techniques such as SMS (8). In this study, we establish a free-breathing DTI approach with complete coverage of the human heart in less than 15 minutes. Respiratory motion during diffusion encoding can cause cancellation of signal in the myocardium. Lengthy breath-hold acquisitions will be burdensome for many patients. Free-breathing acquisition overcomes this hurdle using respiratory-gating with navigator echoes, or by retrospectively eliminating respiration-affected images. In this work, we used STR and ERIS to align and select the usable scans.This is the first demonstration of whole heart DTI producing 3D tractograms of the entire LV using free breathing, of either the short-axis or 4-chamber views, in under 15 minutes. This clinically feasible protocol could facilitate the translation of whole heart DTI to patients with heart failure and other disease states. The characterization of myocardial microstructure using DTI in the clinical setting could provide important insights into disease pathogenesis and treatment.