In vivo cardiovascular magnetic resonance diffusion tensor imaging shows evidence of abnormal myocardial laminar orientations and mobility in hypertrophic cardiomyopathy

This study was approved by the National Research Ethics Service. Eleven healthy controls and 11 patients with HCM were recruited and gave written informed consent. Data in systole from 6 patients was previously reported for reproducibility [24]. Imaging was performed using a 3 T scanner (Skyra, Siemens, Erlangen, Germany) with an 18 element anterior matrix coil and 8-12 elements of a matrix spine coil. After localization of the LV short and long axes, a retro-gated balanced Steady State Free Precession (bSSFP) cine sequence with a 40 ms temporal resolution was used to find the timing and duration of the subject specific end systolic and diastolic pauses, taking care to avoid the atrial kick in diastole.

In order to ensure the same myocardial region was imaged with cDTI in systole and diastole, a breath-hold spoiled gradient echo (GRE) sequence with a spatial modulation of magnetization (SPAMM) tagging pre-pulse and a 45 ms temporal resolution was used to acquire four-chamber and two-chamber views. The linear tags were separated by 16 mm and were perpendicular to the long-axis with an acquired spatial resolution of 2.1 × 1.7 mm in-plane and slice thickness of 6 mm. The displacement of the linear tag closest to the central mid-ventricular slice was manually tracked from the systolic to the diastolic phase. The positioning of the cDTI central slice was manually adjusted between the systolic and the diastolic acquisition accordingly, with the inter-slice distance kept constant.

The in vivo cardiac diffusion weighted stimulated echo acquisition mode (STEAM) single shot echo planar imaging (EPI) sequence used has been previously described [23]. In short, this sequence runs over 2 heart beats and assumes that myocardium returns to the same position at the same encoding times in consecutive cycles, in this work, end systole and the late diastolic pause (diastasis). In order to minimize the length of the single shot EPI readout, parallel imaging and zonal excitation were implemented. Data were acquired in a series of breath-holds to minimize the effects of respiratory motion. Each breath-hold had a duration of 18 heartbeats. The following sequence parameters were used: reference plus 6 diffusion encoding directions at b =350 s/mm² (at a heart rate of 60 beats per minute), fat saturation, TR = 2 RR intervals = 2000 ms (at a heart rate of 60 beats per minute), TE = 23 ms, BW = 2442Hz/pixel, GRAPPA [25] parallel imaging acceleration factor of 2, EPI echo train length = 22-28 readouts, depending on field of view, EPI echo-train readout duration = 11-14 ms, in plane acquired spatial resolution = 2.7 × 2.7 mm², interpolated to 1.35 × 1.35 mm² (k-space zero-filling by a factor of two), field of view 360 × 123-157 mm² and a minimum of 8 averages (one average per breathold). Three slices with 8 mm thickness were acquired in the mid ventricular short axis with an inter-slice gap of 4 mm. No partial Fourier was used. The sequence trigger time was set to acquire cDTI data during the systolic and diastolic rest periods. Localized first and second-order shimming and frequency adjustment were performed over the extent of the LV within the imaging planes. It should be noted that the sequence runs a spoiler gradient in place of the diffusion encoding gradients for the reference (b0) acquisitions, which introduces a small but not negligible diffusion weighting to the "b0" images. This is taken into account during the tensor calculation.

Additional breath-hold CMR was performed to assess cardiac morphology and function in patients and controls. Retrospectively gated bSSFP cine sequences were acquired in 3 long-axis planes, followed by a contiguous stack of short-axis slices from the atrioventricular ring to the apex [26]. Late gadolinium enhancement (LGE) imaging for the assessment of regional fibrosis was also performed, using an inversion recovery prepared spoiled GRE.

Ventricular volumes, function, mass, and ejection fraction was measured using a semi-automated threshold-based technique (CMRtools, Cardiovascular Imaging Solutions, London, UK), with indexing to body surface area [27]. The LV epicardial and endocardial borders were defined by manual planimetry excluding papillary muscles. Each short-axis slice was divided into 12 segments. The presence of LGE was recorded, and when present its extent was quantified using CMR42 software (Circle Cardiovascular Imaging, Calgary, Canada) with a full-width half-maximum threshold and expressed as percent of LV mass.

All quantitative tensor information was post-processed with custom software, written in-house using MATLAB (Mathworks, Massachusetts, USA). In a first step, all the raw diffusion-weighted images were analyzed visually to reject frames corrupted by breathing motion or abnormal RR intervals. A minimum of two averages per direction, after frame rejection, was set as a requirement for inclusion in the study. The accepted frames were then co-registered to correct for intra-subject variations in breath-hold position (in-plane only) with a cross-correlation based algorithm for rigid displacement [28]. The subject's RR-intervals during diffusion acquisition were then calculated from the acquisition times recorded in the DICOM files, and assumed to be half the time between frames (each frame runs over two heart-beats). This information was used to adjust the b-value for each image. A rank-2 diffusion tensor, and its eigensystem (set of three eigenvalues λ1, λ2, and λ3; and the respective eigenvectors: E1, E2, and E3) were then calculated for each voxel of the diffusion dataset. The eigenvalues and respective eigenvectors were sorted in a descending order according to the magnitude of the eigenvalues. A small percentage (3.2%) of the eigenvalues was negative due to motion, noise or mis-registration artifacts. This violates the assumption of a positive definite diffusion tensor, and therefore these values were averaged to the mean of neighbouring voxels.

Two diffusion tensor measurements were calculated from the eigensystem: Helix-angle (HA), and secondary eigenvector angle (E2A) [8],[29]. HA and E2A represent directional information from the first (E1) and second eigenvectors (E2) respectively. These were calculated by identifying the 3 orthogonal cardiac coordinate directions of each myocardial voxel: longitudinal, which is parallel to the left-ventricular long-axis and pointing towards the base; circumferential, which is tangential to the wall with a counter-clockwise direction when viewed from base to apex; and radial, which is given by the cross product of the previous two and pointing outward (Figure1). The primary eigenvector E1 was then projected radially to the local wall tangent plane. HA was defined as the angle in this plane between the E1 projection and the circumferential direction in the range -90 to 90 degrees, being positive (right-handed helix) if rotated counter-clockwise from the circumferential as viewed from the outside, and negative (left-handed helix) if rotated clockwise. The cross-myocyte plane, perpendicular to E1, was then calculated for each voxel. Then E2 was projected onto this plane and E2A calculated in it, between E2 and the illustrated cross-myocyte direction. This angle was measured in the range [-90, 90], being positive if rotated clockwise from the cross-myocyte direction when viewed in the more circumferential direction, and negative if rotated counter-clockwise (Figure1).

For 3D visualization and comparison of E1 distributions, principal eigenvector tractography was performed, color coded according to helix-angle (Figure2A,B,E,F). It must be emphasised that the obtained tracts are only used to aid on the visualisation of E1 orientation. No further analysis was performed on the tractograms, considering the limited number of slices acquired.

To display mean intra voxel sheetlet angulation in the septal and anterior wall regions, the full diffusion tensor was represented by superquadric glyphs, and superimposed with long thin white cylinders, which represent the orientation of E2. The tensors were color coded according to their absolute E2A value (Figure2C,D,G,H). 3D visualization of tensor information and tractography was performed using Python and PARAVIEW (Kitware, NM, USA).

The measured HA and E2A values were binned and plotted as histograms for all subjects in order to visualize the angle distribution. For further analysis of the effects of hypertrophy, E2A data was presented in a scatter plot. E2 mobility was defined as the change of mean absolute E2A values from diastole to systole. E2 mobility uses the absolute value of the angle so polarity is ignored to provide a measure of change of angulation relative to the local wall tangent plane from diastole to systole. Global E2A results included all LV segments in all 3 slices. For further insights in the HCM group, myocardial segments were categorized by consensus by 2 observers (12 segments per short-axis slice): segments with hypertrophy and LGE (H+LG+); segments with hypertrophy but no LGE (H+LG-); and segments with no hypertrophy or LGE (H-LG-). Segments were included in the H+ or LG+ categories even if hypertrophy or late-enhancement were only partially included. Mean absolute E2A values were compared between groups with the Mann-Whitney-Wilcoxon test. Statistical significance was considered present when p < 0.05 after Bonferroni correction for the number of tests performed.

The baseline demographics and clinical characteristics of control and patient groups showed no significant differences (Table2). The time spent in diffusion acquisition was 42 ± 7 min (mean ± SD) for controls, and 46 ± 7 min for HCM patients. The number of required breath-holds was 57 ± 4 for controls, and 63 ± 9 for HCM patients. The SD recorded on the RR interval, during diffusion acquisition, were 38 ± 9 ms for controls, and 41 ± 20 ms for HCM patients.

Average LV mass, wall thickness and ejection fraction were greater in the HCM group and end-systolic volume was smaller. The ventricular morphology, risk factors for sudden death and LGE patterns for the HCM patients are summarized in Table3. Ten out of 11 patients had asymmetrical septal hypertrophy and 1 patient had anterior wall hypertrophy. The median number of risk factors for sudden death was 0 (range 0-2). Six patients had LV outflow tract obstruction. All patients had regional fibrosis on LGE imaging usually affecting the septum and representing 9.1% of LV mass.

Figure2 shows 3-dimensional graphic representations of the orientations of E1 (mean intravoxel helix-angle) and absolute E2 (greater cross-myocyte direction of diffusion) rendered for 1 control and 1 HCM patient at both cardiac phases. The expected helix-angle distributions are observed, transitioning inwards transmurally from left-handed, through circumferential to right handed, without obvious changes of layer-wise orientation between phases, or between the control and the patient. In contrast, glyph representations of the diffusion tensor, and in particular orientation of E2, show unmistakable reorientation, reflecting phasic changes in absolute E2A. E2 tilts towards parallel to the wall-plane in diastole, and towards perpendicular in systole, particularly through mid-wall and subendocardial voxels, consistent with the predicted sheetlet reorientations during the cardiac cycle. In the HCM patient however, most obviously in the most hypertrophic wall region, the phasic changes are less pronounced, with the glyphs remaining relatively perpendicular to the local wall plane in diastole as well as in systole.

cDTI was successfully performed on all subjects. Only 3.2% of all eigenvalues were spuriously negative, for which we assigned the positive mean values of adjoining voxels. In total the number of myocardial voxels analyzed for each group was: controls-systole 28,836 voxels; controls-diastole 26,034 voxels; HCM-systole 45,791 voxels; HCM-diastole 44,661 voxels. The greater number of voxels in HCM patients occurred because of LV hypertrophy. The measured E2 angles were found to be meaningful (more details in Additional file 1: Appendix 1). An example of a mid-slice averaged magnitude image, together with the respective HA, E2A and absolute E2A maps at both cardiac phases is shown in Figure3 for a control and an HCM patient.

The distribution of HA is shown in Figure4. In general, HA maintained the same distribution between cardiac phases and also when comparing controls and HCM patients, all showing greatest frequency around the circumferential orientation (i.e. angles around 0 degrees). The main difference was in the controls, where diastolic distribution shows a narrower peak when compared to the systolic distribution. Contrasting with HA results, the E2A values did not follow an obvious spatial pattern within the myocardium, but showed substantial changes between the 2 cardiac phases, especially for controls (Figure3). A histogram of the measured myocardial E2A values for both groups is shown in Figure5. Systolic values peak at ±90 degrees for both groups, implying sheetlet orientations tilted in towards perpendicular to local wall planes, generally even more so in HCM than controls. However marked differences between the 2 groups were seen in diastole. In controls the diastolic angles peak at approximately 0 degrees, implying mean intravoxel sheetlet orientations towards parallel to local wall plane. However, in HCM, the diastolic distribution curves were flattened, showing a more homogeneous distribution across the range of angles without the tendency towards wall-parallel orientation seen in controls. The histograms were approximately symmetric, except for a slight predominance of negative systolic angles in controls.

Additional analysis of the E2A data is shown in the E2 mobility plots of Figure6. Figure6A shows the global mean absolute values for both cohorts at the 2 cardiac phases. There was a significant difference in global E2A in diastole between HCM and controls (p < 0.001). In systole, there was also a statistically significant difference between the 2 groups of lesser magnitude (p =0.026). Figure6B and Table4 show further analysis of E2 mobility in the HCM cohort, with the myocardial absolute E2A values divided into 3 categories: segments with hypertrophy and LGE (H+LG+); segments with hypertrophy but no LGE (H+LG-); and regions with no hypertrophy or LGE (H-LG-). There was a significant difference in E2A at diastole between non-hypertrophic segments (H-LG-) and the other two groups H+LG+ and H+LG- (p =0.0028 and p =0.0022 respectively, both after Bonferroni correction for 2 tests). There was also a significant difference in E2A in systole between hypertrophic regions and controls for H+LG+ segments and H+LG- segments (p =0.0060 and p =0.0030 respectively both after Bonferroni correction for 3 tests).

Impairment of E2 mobility in hypertrophic segments is shown in Figure7, where slices from 2 HCM hearts with asymmetric hypertrophy are compared with the corresponding slice from a control. The non-hypertrophic lateral wall in both HCM hearts approaches the absolute E2A angles measured in the control heart, while differences between the hypertrophic regions and the control values are clearly apparent in diastole. There was no obvious effect from the presence or absence of LGE on the E2A values in the hypertrophied anteroseptal wall, in accord with the results shown in Figure6B.

The microstructure of myocardium is not isotropic. However, with a view to its potential relevance to cDTI, Reese et al. considered the effects of cyclically applied strain on aqueous diffusion through a structurally isotropic material [31] They used an elastic gel phantom, periodically indented by a hemi-spherical driver. This caused radially directed compression of the gel with counter-stretching in the orthogonal circumferential directions, analogous with the radial thinning and bidirectional elongation of LV walls in diastole. On release, elastic restoration of the gel was associated with radially directed thickening and bidirectional shortening, analogous with systolic deformations of the LV wall, as imaged macroscopically. As predicted by the underlying theory of diffusion through an isotropic material during deformation, DTI in the `systolic' phase showed anisotropy, with increase of the radial and reduction of the circumferential components of diffusion. We have schematically illustrated this effect in Figure9A. The converse was found by DTI in the diastolic phase. Given normal, healthy myocardial strain, comparable strain-induced directions and amounts of anisotropy might be thought to account for the changes of E2A that we found in healthy volunteers. Strain might be taken as the underlying explanation, except for the following crucial considerations:

Firstly, the myocardium is not structurally isotropic, but orthotropic in most LV regions [9], meaning that it has different constituent properties in three orthogonal directions: the myocyte, sheetlet and sheetlet normal directions. Diffusion taking place through it is likely to be constrained by microstructural boundaries in ways that modify the postulated effects of strain on diffusion, particularly for the inherent long mixing-times (one heart-beat) of the STEAM sequence. Furthermore, the constituent microstructures undergo their own distinct small-scale deformations, including the shearing and cyclic reorientation of extracellular fluid in the interstitial shear layers. This is perhaps the most pertinent of several types of micro-deformation in contracting myocardium that may contribute to the measured anisotropies of diffusion. Others include the even smaller scale interdigitation of actin and myosin nanofilaments within sarcomeres, the lengthwise shortening and slight circumferential counter-thickening of whole myocytes and the reorientations of the sheetlets in which the myocytes are arrayed. These micro-deformations underlie but can, especially in the cross-myocyte directions, differ in magnitude and sign from deformations imaged macroscopically. The marked systolic thickening of myocardium is known to be caused more by laminar reorientations than by the thickening of individual myocytes [7]. This amount of radial thickening, combined with systolic shortening in cross-myocyte as well as myocyte directions in wall tangent planes, is only achievable through the presence and mobility of laminar microstructures.

Secondly, the effects of strain cannot account for the exaggerated systolic anisotropy of diffusion that we found in the hypertrophied regions of HCM hearts. As is well known in HCM, strain is reduced, especially in hypertrophic regions [20]. However, in systole, we measured above normal E2As implying marked predominance of radial over tangential components of diffusion. In diastole, on the other hand, in spite of the fact that even the hypertrophic myocardium had undergone radial thickening with slight longitudinal and circumferential shortening during systole, albeit of limited amount, there remained a predominance of radial over tangential components of cross-myocyte diffusion, with E2As greater than 45°. From the postulated effects of strain alone, the opposite would have been the case. To appreciate an alternative, more plausible interpretation of the HCM DTI results, the function of laminar microstructures should be considered further.

The contraction of myocytes of one myocardial layer, for example the sub-epicardial layer, imposes cross-myocyte shortening on the simultaneously contracting myocytes of a different layer, in this case the subendocardial. As mentioned in the introduction, it is this potential conflict that requires the presence and function of laminae: the sheetlets and shear layers not only offer compliance that allows cross-myocyte shortening in spite of their individual thickening, but the changing angulations of laminae, like the folds of a compressed concertina [7], contribute to wall thickening. The reorientations of laminae of a particular region are driven not by their own force, but rather by the contraction of deeper and/or more superficial myocytes of the same region. Contraction of myocytes through all depths of any given segment causes laminae to adopt steeper, through-wall obliquity, with increase of E2A values and of wall thickness. This underlies our interpretation of the elevated E2As found in the hypertrophic regions of HCM patients, in diastole as well as systole.

We had hypothesized that sheetlet mobility in HCM would be impaired, which we found. Our results also show that the diastolic orientations appear markedly abnormal, with the E2A of HCM hearts retaining relatively systolic conformation in diastole. This abnormality was associated with increased wall thickness, independent of the presence of myocardial LGE (Figure6B). This suggests that the wall's state of contraction or hypertrophy (not presence of LGE) is the dominant determinant of E2A in HCM, a finding that is consistent with findings by Aletras et al., that regional contractile heterogeneity in HCM could not be explained by the distribution of fibrosis [20]. It must however be noted that although capable of detecting regional fibrosis, LGE imaging is not suitable to image diffuse fibrosis [32]. T1-mapping is an emerging technique more suitable to detect and quantify diffuse fibrosis [33].

Although less marked than the diastolic abnormality, we also found exaggerated E2A values in systole, consistent with a hypercontracted state.

Our in vivo data, which suggests relatively systolic sheetlet orientations and impaired relaxation is consistent with current hypotheses of the pathophysiological consequences of HCM-causing mutations from biochemical studies and transgenic models. A common property of HCM-causing mutations is increased myofilament or myosin ATPase calcium sensitivity [34]-[37]. These mutations increase cardiomyocyte tension for a given cytoplasmic calcium level, particularly in the diastolic range, slowing relaxation kinetics, limiting diastolic relaxation, and under some circumstances increasing the tension generated during systole. Exaggeration and persistence of systolic tone, and delayed relaxation, have implications for myocardial perfusion given that coronary flow predominantly occurs in diastole enabled by myocardial relaxation [38], and can be compromised in HCM [39].

The improved understanding of pathophysiological mechanisms has potential implications for novel approaches to therapeutic intervention [40],[41]. It may be feasible to use cDTI to identify abnormal myocardium, before the onset of fibrosis, and monitor its response to the effects a drug, known or novel, initially in animal experiments and, if apparently successful, ultimately in patients.

No macroscopic evidence of disarray in the HA maps of HCM patients was found. These findings contrast with the work of Tseng et al., which found more longitudinally oriented HAs in the hypertrophied epicardial septum when compared to the normal heart [42]. We consider there may be 2 factors, which have contributed to this difference. In some of the HCM subjects, longitudinal oriented HAs that were considered to be hypertrophy from the septo-marginal trabeculations of the right-ventricle were excluded from analysis of the left ventricle. Additionally our cohort of HCM subjects was relatively old with relatively few risk factors and might therefore have had limited myocardial disarray.

The spatial-resolution achievable in vivo cDTI is limited by acquisition parameters chosen for suitability in patients with limited breath-holding ability. In controls, diastolic wall thinning reduces effective resolution of differences between layers, although this was less problematic in HCM. A zero-filling interpolation algorithm was used. This has the potential to increase spatial resolution by using the corners of k-space, although the spatial resolution improvement becomes angle dependent [43]. The impact of this interpolation in measuring mean E2 angles was tested in 18 datasets acquired for another study; results showed minimal differences of less than 3 degrees when compared to no interpolation. A maximum difference lower than 8 degrees was found for the helical-angle. We therefore expect this interpolation to have minimal impact in the diffusion tensor calculation.

There are a number of limitations to this study. As discussed above, the sensitivity to myocardial strain of the cDTI technique remains unclear. It is likely to differ from that of a homogenous, elastic gel and further work is required to investigate whether or not it contributes to the results reported.

The relation between E2A and the orientations of sheetlets within a voxel is unlikely to be straightforward. Thousands of cardiomyocytes arrayed in hundreds of sheetlets, with potentially two or more opposing populations may occupy a single 2.7 × 2.7 × 8 mm³ imaging voxel [40],[44]. The sheetlets are expected to change orientation through the cycle relative to one another as well as to the local wall plane. Angles between adjacent sheetlet populations of the mid layer might, for example, if viewed like the folds of an accordion, be obtuse when extended longitudinally in diastole, but acute when brought together in systole. Were a voxel to contain 2 equal opposing populations, this might result in very small, almost wall parallel, E2As in diastole even though individual sheetlet populations remained oblique. Interestingly, however, our analysis of E2A supports the predominance of one orientation of sheetlet population in most voxels. We have not yet identified any consistent pattern of distribution of the `positive' and `negative' slopes of these between individuals. The different and unknown orientations of 2 or more sheetlet populations in a voxel cannot be elucidated using a single tensor. cDTI techniques with higher angular diffusion resolution (multi-tensor, diffusion spectrum imaging, and q-ball) would be needed for this [45], although the acquisition times required would currently be prohibitive in a clinical setting.

Myocardial coverage was limited to 3 slices. The lack of whole heart coverage made it challenging to define the longitudinal wall tangent direction. In this work, it was taken as perpendicular to the short-axis plane, which is a better approximation for the basal and mid than the more apical slice. Intra-subject variations in breath-hold are corrected with a rigid in-plane registration algorithm; through-plane variation remains uncorrected. The SPAMM imaging was used to track manually the imaged slices from systole to diastole, with the objective of reducing differences in the imaged volume between the two phases. This method is very approximate and it does not take into account the complex cardiac deformation present entirely, particularly in the HCM cohort. Nevertheless we believe this method to be sufficient for the DTI metrics shown, as we are only showing differences in global mean values.

The diffusion-encoding scheme used had only 6 directions. Recent DTI work in the brain showed comparable Mean Diffusivity (MD) and Fractional Anisotropy (FA) between 6 directions and 30 or 60 direction schemes (acquisition time kept constant) [46]. DTI numerical simulations also show comparable tensor accuracy when trading directions for averages [47]. It remains unknown if 6 direction schemes are comparable to higher order schemes for cDTI. A recent study from our group shows that both HA and E2A maps become smoother with higher b-values (>750 s/mm²). However E2A global mean values do not change substantially [48].

The averaging of non-positive definite tensors was performed for the eigenvalues directly. We appreciate that this is not a robust method, it maintains the neighbouring FA and MD, but it lacks sophisticated tensor interpolation. Nevertheless it affects only a small percentage (3.2%) of the eigenvalues measured, so we expect the impact to be small for the global means of E2A.