Diffusion tensor cardiovascular magnetic resonance in hypertrophic cardiomyopathy: a comparison of motion-compensated spin echo and stimulated echo techniques

Diffusion tensor (DT) cardiovascular magnetic resonance (CMR) is used to obtain non-invasive measures of myocardial microstructure [1‐3]. Parameters such as mean diffusivity (MD) and fractional anisotropy (FA) describe the freedom of water motion, and degree of myocardial microstructural organisation. Helix angle (HA) and second eigenvector angulation (E2A) relate to the orientation of cardiomyocytes and sheetlets, respectively. Cardiomyocytes take a left-handed helical arrangement at the epicardium and progress smoothly through a circumferential orientation at the mesocardium to a right-handed helical arrangement at the endocardium. This microarchitecture is generally preserved through the cardiac cycle [4]. Sheetlets are dynamic bundles of myocytes that align near wall-perpendicular in diastole and re-orientate to become more wall-parallel in systole, thus effecting wall thickening [4‐6]. E2A is a measure of sheetlet alignment and the change in E2A between peak systole and diastole is known as sheetlet mobility [4].

There are currently two main DT-CMR sequences employed in the in vivo assessment of the myocardium (Fig. 1); STimulated Echo Acquisition Mode (STEAM) and second-order motion-compensated spin echo (M2-SE). STEAM is the more established technique, having been used to characterise both healthy cohorts and also patients with cardiac pathology including cardiomyopathy [2, 4, 7‐9]. In hypertrophic cardiomyopathy (HCM), STEAM DT-CMR has shown elevated diastolic E2A, which results in reduced E2A mobility [7] and that reduction of FA in hypertrophied myocardium is associated with ventricular arrhythmia and possibly myocyte disarray [9].

However, STEAM suffers from an intrinsically low signal–noise-ratio (SNR), the limited potential confounding effects of strain [1, 10], as well as dependence upon a regular heart rate. Alternatively, M2-SE could offer benefits; it is strain-insensitive, has a higher SNR and has shown promising results in healthy volunteers scanned using high-performance gradient systems at 1.5 T during systole [3, 11]. In recent work, we compared STEAM and M2-SE at multiple cardiac phases in healthy volunteers at 3 T using a widely available gradient specification [12]. However, while a limited number of studies in disease have used a related diffusion-prepared technique for diffusion-weighted and, more recently, diffusion tensor imaging with limited success [13, 14], M2-SE DT-CMR has only been used in patients in one recent study [15] and has yet to be compared to STEAM in a patient cohort.

This study compares STEAM and M2-SE DT-CMR techniques in HCM patients. It aims to establish technique viability, compare diffusion parameters from both sequence types and better understand the relative value of the different approaches to characterise HCM.

This study was approved by the National Research Ethics Committee and all participants provided informed written consent. HCM patients were prospectively recruited. The diagnosis of HCM followed the 2011 ACCF/AHA Guideline for the Diagnosis and Treatment of Hypertrophic Cardiomyopathy and the 2014 ESC Guidelines on diagnosis and management of hypertrophic cardiomyopathy criteria [16, 17]. Patients who had undergone previous alcohol ablation or septal myectomy were excluded.

The baseline characteristics of the patients are shown in Table 1. All patients had asymmetrical septal hypertrophy and presence of LGE. Within the slice imaged with DT-CMR, all patients had areas of hypertrophy and LGE in the septum but not in the lateral wall.

All patients completed the scan protocol. Typical scan duration was 90 min. Median [IQR] breath-holds in diastole were 11 [4] for STEAM and 12 [7] for M2-SE (p = 0.20) and in systole 11 [2] in STEAM and 11 [3] for M2-SE and (p = 0.53). Median [IQR] scores for quality of HA maps were STEAM diastole: 2 [2], M2-SE diastole: 0 [1], STEAM systole: 2 [2], M2-SE systole: 1 [2]. Data were analysable from all STEAM acquisitions, but only 7/10 (systole) and 4/10 (diastole) M2-SE acquisitions. Typical examples of DT-CMR parameter maps are shown in Fig. 2.

Missed-triggers occurred in 0 STEAM systole, 1 M2-SE systole, 3 STEAM diastole and 7 M2-SE diastole acquisitions. For the STEAM data and M2-SE systole data, we estimate that there was a maximum missed-trigger rate of 2.5%. In contrast, for the M2-SE diastole data there were acquisitions with 0.7, 1, 5, 7, 12, 18 and 19% missed-trigger rates. Of the acquisitions with missed-triggers rates > 10%, 2 of the 3 were rejected due to poor quality data. Median [IQR] SNR for data sets scoring > 0 was 7.4 [1.9], STEAM diastole, 4.2 [1.3] M2-SE diastole, 8.5 [1.6] STEAM systole and 5.6 [1.9] M2-SE diastole.

Figure 3 and Supplementary Table 1 show that the average myocardial DT-CMR results over the mid short-axis slice varied by sequence at both cardiac phases. For MD, M2-SE yielded higher values at both phases (p = 0.02 and p = 0.6 in systole and diastole, respectively). Conversely, FA was lower when obtained using M2-SE at both phases (p = 0.02 and p = 0.1). Biphasic HAG was steeper and E2A was higher when measured using STEAM at both phases, though not significantly.

Example DT-CMR maps alongside an LGE image and ECV map are shown in Fig. 4. Results of the 12-segment analysis comparing MD, FA and E2A in LVH + LGE + to LVH − LGE − regions are found in Supplementary Tables 2 and 3 and comparative plots are shown in Fig. 5. Statistical analysis between these regions was only performed in systole due to insufficient data points in diastole and to avoid excessive sub-analysis. Using STEAM, LVH + LGE + areas were distinguished from LVH − LGE − areas by elevated diastolic E2A and reduced E2A mobility (p = 0.002). Using M2-SE, FA was significantly lower in LVH + LGE + regions (p = 0.02).

Pearson correlations and linear regression were performed between the ECV and native T1 measured in each segment and the MD and FA values in the same segments (Fig. 6). Significant correlations exist between ECV and both MD and FA using STEAM (R² = 0.17, p < 0.001 and R² = 0.29, p < 0.001, respectively). Using M2-SE, only FA showed a negative correlation with ECV (R² = 0.1, p = 0.003). As expected, MD increases and FA reduces with increasing ECV. There were no significant correlations between native T1 and FA or MD measured in systole using either sequence.

This study is the first to evaluate and compare two DT-CMR techniques in a patient cohort. We studied patients with HCM as similar DT-CMR sequences [7, 13, 26] have previously identified microstructural abnormalities in these patients independently. We sought to establish and compare the feasibility of both STEAM and M2-SE, provide an initial comparison of DT-CMR results and initial insights into the future clinical and scientific utility of the two methods in patients with HCM.

Results show that currently only STEAM can reliably perform biphasic microstructural assessment in HCM patients. M2-SE is feasible for systolic assessment, but has a high failure rate (60%) in diastole. Our prior work comparing M2-SE and STEAM in healthy volunteers using the same sequences, protocols and scanner found that: 1/15 systolic, 3/15 sweet spot and 7/15 diastolic M2-SE were not evaluable, whereas all but one diastolic STEAM data set were interpretable [12]. While the success rates of M2-SE imaging in both systole and diastole were reduced in the cohort of HCM patients studied here, the reduction relative to our success rates in healthy volunteers is not statistically significant. Based on a binomial distribution, the 95% confidence intervals of the success rates observed in our healthy volunteer study are 68–100% in systole and 27–79% in diastole. This non-significant difference in success rates between healthy volunteers and HCM patients could be a consequence of patient size: (body surface area was substantially larger in HCM subjects, 2.03 [0.4] vs. 1.7 [0.2]); age (HCM subjects were substantially older, 64 [11] vs. 24 [11]); anxiety and experience in the MRI scanner, which is likely to be less and more, respectively; or simply statistical fluctuations. We have also demonstrated the greater influence of eddy current-related artefacts in the M2-SE than in the STEAM acquisitions using a phantom.

DT-CMR data sets were accepted or rejected based on 50% of the myocardium demonstrating the expected progression of HA from a negative or left-handed helical angle at the epicardium, to a positive or right-handed helical arrangement on the endocardium. HA maps were used to assess DT-CMR quality, as HA appears relatively preserved in most pathologies we have studied, thereby avoiding the need to define an arbitrary threshold of some parameter, like MD or SNR, that may also be abnormal as a consequence of disease. Rejected data sets, where < 50% of the myocardium showed a normal appearing HA, tended to have low SNR in the regions of abnormal HA (see Fig. 4 and Supplementary Figure S2 in [12] for examples). The lower success rate of diastolic imaging in both healthy volunteers and HCM patients could relate to the complex cardiac motion during diastole that second-order motion compensation fails to address. However, assessing the magnitude of such motion components would require a technique able to produce accurate and precise measures of tissue displacement or velocity with high temporal resolution, covering the whole cardiac cycle. Neither the DENSE data or feature tracking of the cine data available in this HCM cohort were appropriate for this analysis. In addition, as demonstrated by our analysis of the DICOM timestamps in this work, the longer diffusion gradients (required for motion compensation) extend close to the next R wave and increase the likelihood of missing the next trigger for M2-SE diastolic imaging. This mis-triggering leads to signal loss. We visually inspect all data acquired and aim to reject all acquisitions where there is visible artefact or abnormal signal intensity. However, rejecting images clearly results in a reduction in the diffusion tensor signal-to-noise ratio and a careful balance is required between minimising artefacts and maintaining sufficient data.

Another study demonstrated that diastolic M2-SE DT-CMR is less precise than systolic, with increased uncertainty in all parameters measured in diastole [27]. However, an earlier smaller comparison study of M2-SE and STEAM with sweet-spot imaging at 1.5 T did not reject any data sets and the authors found improved signal to noise ratio when using M2-SE [3].

The maximum gradient strength available on the system used in this work was less than that available in many other M2-SE studies (43 mT/m vs. 80 mT/m) [3]. While eddy current-related artefacts would likely be more severe, higher strength gradients would assist in reducing M2-SE diffusion encoding gradient durations and, therefore the time period over which the sequence is sensitive to higher-order motion-related signal loss.

There was a systematic difference in MD and FA findings between M2-SE and STEAM. This has been reported before both in vivo and ex vivo [3, 12, 28]. The higher MD and lower FA values obtained using M2-SE reflect the shorter diffusion time and, therefore, fewer interactions between water molecules and microstructural barriers as demonstrated in recent simulation work [29]. Whilst these differences between sequences were only statistically significant for systole in this study, the trend was present in diastole and lack of statistical significance may reflect the small numbers in this group. The time-dependent nature of the measured MD and FA means that validation against reference standards is currently not possible for complex biological-like anisotropic media, but validation of HA and E2A as measures of the orientation of cardiomyocytes and sheetlets has been performed in previous studies [4, 30]. The reduction in helix angle gradient using M2-SE relative to STEAM was also previously found in healthy volunteers [12] and could reflect increased noise in the helix angle maps or a consequence of residual blood signal in epi and endocardial pixels in M2-SE data. E2A values were similar between sequences and appear higher in systole than diastole (although statistical comparisons were not performed due to small numbers of diastolic M2-SE data). Interestingly diastolic E2A was similar by both techniques at around 50°, but this value is higher than previously reported values in normals, which are in the region of 15° for STEAM and 30° for M2-SE [3, 7]. The implication is that M2-SE may not be able to clearly detect the diastolic abnormality previously reported in HCM, namely elevated diastolic E2A and failure of diastolic sheetlet relaxation [6, 7].

This study goes further than previous comparisons by examining the ability of both sequences to discriminate areas of LV hypertrophy and fibrosis. However, diastolic findings with M2-SE were not analysed due to the high failure rate and limited number of data points. Both techniques found a lower FA in LVH + LGE + areas, but this was significant only with M2-SE. This FA reduction in phenotypically diseased regions is in accordance with previous studies [9, 26]. STEAM-based studies in HCM have shown that MD was higher and FA values were lower in the septum, possibly due to septal hypertrophy and the associated increase in extracellular space or fibrosis [31]. A prior M2-SE diffusion-weighted study in HCM reported significantly elevated MD in areas of LV fibrosis, as detected by ECV [13], but diffusion-weighted techniques do not assess FA. Together these findings suggest that the shorter diffusion time may be advantageous in identifying alterations in the dimensions and integrity of the microstructural barriers to free diffusion. However, future studies will be needed to determine the underlying mechanism behind these changes.

Both techniques demonstrated a trend toward higher systolic E2A in hypertrophied and fibrosed myocardium. Previously STEAM has shown abnormalities of E2A in HCM compared to healthy controls with elevated diastolic E2A and decreased E2A mobility globally, particularly in areas of hypertrophy [4, 7]. This relates to sheetlets retaining a more systolic, wall-perpendicular orientation in diastole.

There was an inverse correlation between FA and ECV using both M2-SE and STEAM. FA reflects the organisation of the underlying tissue, and there is decreasing organisation with lower FA values. This reduction in FA may represent a combination of myocyte disarray, myocyte hypertrophy, increased extracellular space and also diffuse or focal fibrosis. However, current DT-CMR techniques cannot distinguish the relative contribution of these and perhaps other factors.

While DT-CMR is yet to find a formal clinical role, patient studies have shown clear changes relating to the underlying disruption of the microarchitecture and its function [4, 8, 9, 12, 26, 31‐34]. The scientific utility of the method is, therefore, clear and there is a growing level of interest in investigating the clinical utility of the technique within the CMR research community. When selecting the most appropriate DT-CMR technique both the sensitivity and specificity of the technique to the underlying microstructural changes of interest and the usability of method in a clinical setting must be considered. M2-SE was possible in the majority of healthy volunteers and patients during systolic contraction. Systolic FA measured using M2-SE was the only of the tensor invariants to show significant differences between LVH + LGE + and LVH − LGE − regions. In contrast, STEAM was successfully performed in both cardiac phases for all HCM patients and is the only method to have been successfully used to assess differences in dynamic sheetlet orientation during contraction via E2A mobility.

In conclusion, currently only STEAM is capable of reliable biphasic assessment of the myocardium in HCM. M2-SE has a high failure rate in diastole. STEAM is therefore the current method of choice for assessing changes in microstructure during cardiac contraction, including sheetlet reorientation and the reduction in sheetlet mobility observed in cardiomyopathies.

The globally higher MD and lower FA obtained by M2-SE reflect the shorter diffusion time of this sequence. This shorter diffusion time may also explain the ability of M2-SE to detect differences between areas of LVH and fibrosis and those without via FA. However, further studies are required to examine the relationship between DT-CMR results and underlying microstructural changes such as fibrosis.