Although the quality of in-vivo human cardiac diffusion weighted imaging (DWI) has increased considerably over the last years, most knowledge about whole heart myocardial architecture results from ex-vivo experiments (1). In-vivo experiments are typically limited to 1 to 3 slices located in the mid-ventricle with only few studies showing whole heart coverage (2,3). The aim of this study was to provide a comprehensive description of whole heart in-vivo myocardial fiber architecture using the fiber architecture matrix (FAM) (4), and to validate results using ex-vivo and simulated data.

Nine healthy volunteers (5 Female, mean age 24 [22-34])were imaged with a 3T scanner (Philips, Achieva, Release 5.1.7) using a 32-channel cardiac coil. DWI was performed using a cardiac triggered SE-EPI sequence in free breathing with asymmetric bipolar gradients (5) and additional flow compensation (2). The b-values were 0, 10, 20, 30 ,50, 100, 200 and 400 s/mm² with 6, 3, 3, 3, 3, 3, 3, and 24 gradient-directions, respectively. Other imaging parameters were; TR: 14 heart-beats, FOV: 280x150 mm², slices: 14, voxel size: 7x2.5x2.5 mm³, acquisition matrix: 112x48, SENSE: 2.5, partial Fourier: 0.85, trigger delay: 220 ms, and acquisition time: 12 min. Ex-vivo DWI data was acquired using a multi shot STEAM-EPI sequence using Stejskal-Tanner gradients. In total twelve datasets of formalin fixed porcine hearts were acquired (6): nine with a “low” resolution of 6x2x2 mm³ and three with a “high” resolution of 2x1x1 mm³ with b-values of 500, 1000, 2000 and 3000 s/mm² and 30 gradient-directions per b-value.

Data processing was done using DTITools (Mathematica 10) and comprised the following steps: registration to correct for subject motion and eddy current deformations (in vivo: 2D b-spline, ex-vivo: 3D affine), Rician noise suppression, IVIM correction (7), tensor calculation using WLLS, manual segmentation of the ventricles (Figure 1A), calculation of the local myocardial coordinate system (LMCS) (Figure 1), voxel wise FAM calculation (4) and whole heart fiber tractography of all three eigenvectors (ε₁ ε₂ and ε₃) using the vIST/e toolbox (Figure 2). The FAM is defined as the angle between the LMCS and the projection of the eigenvalues along the axes of the LMCS (see Figure 3). Transmural profiles (180 radial profiles per slice) of the FAM were obtained for 15 points along the myocardial wall using first order interpolation along the radial axes of the LMCS (green vectors in Figure 1C). To validate the obtained transmural profiles, they were fitted using three quadratic transmural functions describing the transmural rotation around the three axes of the LMCS.

Figure 2 shows whole heart fiber tractography of the eigenvectors for one high resolution ex-vivo (top) and one in-vivo dataset (bottom). One can appreciate similarity, with the exception of the direction of the second and third eigenvector, which are switched ex-vivo compared to in-vivo. The change in eigenvectors between the in-vivo and ex-vivo data is further confirmed by evaluation of the FAM maps which are shown in Figure 4. The high and low spatial resolution ex vivo cases show identical results (Figure 4A and B respectively). FAM maps of the in vivo data are also identical with the exception of those belonging to ε₂ and ε₃. Transmural FAM profiles of all datasets are shown in Figure 5. To allow comparison between the angles of the second and third eigenvectors of the in-vivo and ex vivo data, the profiles of the second and third eigenvectors of the in-vivo data are switched (blue lines in the bottom two rows). Figure 5 shows is great similarity between in-vivo and ex-vivo transmural FAM profiles. Simulated transmural FAM profiles are also shown in Figure 5 and demonstrate that the profiles can be approximated using a simple quadratic transmural rotation along the LMCS (see Figure 3 top left). To clarify the FAM profiles, transmural fiber tractography of a high resolution ex-vivo data is compared to those obtained from the simulation (Figure 3).In this study we showed that whole heart fiber architecture described by the FAM can be obtained from in-vivo DWI. However, compared to ex-vivo data the direction of ε₂ and ε₃ are exchanged. This means that, in contrast to prior ex-vivo studies, in-vivo diffusion along the radial axis of the heart is greater than the diffusion direction associated with the sheet structure of the heart. This could be due to residual motion encoding of the contracting heart along this direction or changes in water diffusion due to the fixation process commonly used for ex-vivo data.