Determinants of myocardial function characterized by CMR-derived strain parameters in left ventricular non-compaction cardiomyopathy

Clinical presentation of left ventricular non-compaction cardiomyopathy (LVNC) can be heterogeneous from asymptomatic expression to congestive heart failure. Deformation indices assessed by cardiovascular magnetic resonance (CMR) can determine subclinical alterations of myocardial function and have been reported to be more sensitive to functional changes than ejection fraction. The objective of the present study was to investigate the determinants of myocardial deformation indices in patients with LVNC. Twenty patients with LVNC (44.7 ± 14.0 years) and twenty age- and gender-matched controls (49.1 ± 12.4 years) underwent functional CMR imaging using an ECG-triggered steady state-free-precession sequence (SSFP). Deformation indices derived with a feature tracking algorithm were calculated including end-systolic global longitudinal strain (GLS), circumferential strain (GCS), longitudinal and circumferential strain rate (SRll and SRcc). Twist and rotation were determined using an in-house developed post-processing pipeline. Global deformation indices (GLS, GCS, SRll and SRcc) were significantly lower in patients with LVNC compared to healthy controls (all, p < 0.01), especially for midventricular and apical regions. Apical rotation and twist were impaired for LVNC (p = 0.007 and p = 0.012), but basal rotation was preserved. Deformation indices of strain, strain rate and twist correlated well with parameters of the non-compacted myocardium, but not with the total myocardial mass or the thinning of the compacted myocardium, e.g. r = 0.595 between GLS and the non-compacted mass (p < 0.001). In conclusion, CMR deformation indices are reduced in patients with LVNC especially in affected midventricular and apical slices. The impairment of all strain and twist parameters correlates well with the extent of non-compacted myocardium.


Material and Methods
The study was conducted in accordance to the Declaration of Helsinki and its later amendments. The study design was approved by the Ethics Committee of the Canton of Zurich and written informed consent was obtained. All data used for this study were handled anonymously.

Study population.
In total, 20 patients with LVNC were prospectively included into this study from our outpatient clinic, between October 2011 and July 2016. Before the acquisition of the CMR, the enrolled patients had been diagnosed with LVNC according to current recommendations for transthoracic echocardiography (TTE) 2,11 . TTE parameters defined the non-compacted to compacted ratio > 2 in systole in a short-axis view predominantly localized in midventricular and apical segments with additional color Doppler evidence of perfused intertrabecular recesses as well as absence of coexisting cardiac abnormalities 2 . CMR diagnostic criteria defined the ratio of the thickness of LV non-compacted myocardium to compacted myocardium as greater than 2.3 in any long-axis view during diastole 11 . The LVNC group as well as 20 additional age-and gender-matched healthy controls received CMR. Cardiovascular magnetic resonance. Imaging was conducted on a 1.5 T MR imaging system (Achieva, Philips Healthcare, Best, The Netherlands) using a 5-channel phased array coil. After scout and reference scans, functional and geometric assessment was achieved by electrocardiogram-triggered, cine steady-state-free precession (SSFP) images in standard long-axis geometries (two-, three-and four-chamber view) as well as in short-axis orientation with full ventricle coverage from basis to apex (repetition time (TR)/echo time (TE) = 3.3/1.6 ms, flip angle (FA) = 60°, spatial resolution = 8 × 1.5 × 1.5 mm 3 , 2 slices per breath-hold, minimum phases: 25). Late gadolinium enhanced imaging 15 minutes after contrast agent (Gadovist, Bayer Healthcare, Berlin, Germany 0.2 mmol/kg) administration was performed in order to detect myocardial scarring or fibrosis. Therefore, a gradient-spoiled turbo fast-field-echo sequence with a non-selective 180° inversion pre-pulse was acquired at end-diastole with anatomical reference taken from SSFP images.
Standard analysis of functional and geometric LV indices was obtained including left and right ventricular end-diastolic volume (LVEDV/RVEDV), left and right ventricular ejection fraction (LV-/RV-EF), indexed left ventricular compacted mass with papillary muscles (LVMi-C), indexed left ventricular compacted and non-compacted mass (LVMi-MM), indexed left ventricular non-compacted mass (LVMi-NC) as well as their ratio (LVMi-NC/MM) 12 . In addition, the end-diastolic extent of compacted (LV-C) and non-compacted (LV-NC) myocardium as well as their ratio were measured in one long axis geometry (LV-NC/C) 11 .
Myocardial feature tracking (FT) analysis was performed using TomTec software (Image-Arena VA Version 3.0 and 2D Cardiac Performance Analysis MR Version 1.1.0; TomTec Imaging Systems, Unterschleissheim, Germany) on the basis of a previously established algorithm 13 . Cine images in long-and short-axis view were used for the analyses of end-systolic global longitudinal (GLS) and global circumferential (GCS) strain of the left ventricle, a global right ventricular longitudinal strain (RV-GLS) and longitudinal (SR ll ) as well as circumferential (SR cc ) peak diastolic strain rate. For strain analysis, endocardial contours were manually drawn independent of the cardiac cycle followed by subsequent software-driven automatic tracking of image features, such as tissue patterns or signal inhomogeneities (Fig. 1). Quality adjustment was performed afterwards and contours were amended manually if necessary. The phase of end-systolic strain was defined by aortic valve closure in three-chamber view for the long-axis geometries and the peak contraction around the nadir of the global strain curve for short-axis geometries 5 . In addition, intra-and interobserver agreement (both observers > 3 years' experience in cardiac imaging) was achieved by repeating strain analysis on 28 randomly chosen subjects (14 LVNC and 14 controls).
In addition to strain parameters, peak apical and basal rotation as well as the end-systolic twist was obtained. End-systolic twist was defined by the subtraction of minimum basal rotation, obtained around end-systole, from maximum apical rotation around end-systole 14 . www.nature.com/scientificreports www.nature.com/scientificreports/ Customized motion tracking. As a result of unsatisfactory low reproducibility of the rotation parameters determined with the commercially available software, a custom-built, post-processing pipeline was built in-house to aim for more reproducible rotation parameters from short-axis cines images.
Dense motion fields via image registration were estimated using a previously established algorithm publicly available as Matlab toolbox 15,16 . The algorithm does not require myocardial segmentation and receives as an input N cine images that correspond to the different cardiac phases. The regularized registration problem is solved by estimating N spatial image transformations that minimize the nuclear norm of image patches. Such group-wise image similarity metric allows to non-parametrically enforce image alignment via patch correlation 17 . For the current short-axis images, cubic image interpolation was used, linear splines for displacement parametrization with grid stride of 4 pixels, L1-regularization of second order displacement derivatives in temporal domain with the weight of 5·10 -3 , vectorial total variation regularization in the spatial domain with weight of 5·10 -2 . Estimated displacement fields were then used to deform myocardial contour between different cardiac phases, e.g. displacement fields that map phase t to phase 0 were used to propagate the contour at phase 0: c t = c 0 ° τ t,0 , (e.g end-systole to end-diastole).
After segmenting the myocardium in a basal, midventricular and apical short-axis slice by applying epi-and endocardial contours, its length (for endo-and epicardium) s l , and mean in-plane width s w was calculated. Circumferential (S t c ) strain at the cardiac phase t was then defined as previously described 18 : where s l 0 is the contour length at initial end-diastolic phase. In addition, rotation was estimated for a basal and an apical slice. To estimate the amount of rotation between cardiac phases 0 and t, a singular value decomposition was used to fit a rigid body motion model to displacement vectors inside the myocardium and then extracted Euler angle from the orthogonal rotation matrix. Statistical analysis. Statistical analysis was performed using SPSS 24.0 (SPSS Inc., Chicago, IL, US).
Unless otherwise stated, continuous variables are presented as mean ± standard deviation (SD). Normal distribution was tested using the Shapiro-Wilk test. Categorical variables are reported as percentage. Data between www.nature.com/scientificreports www.nature.com/scientificreports/ the two different groups were analyzed by 2-sided unpaired Student's t-tests for normally distributed data and Mann-Whitney U-test for not normally distributed data. For post-hoc analyses, the LVNC patient group was divided into two subgroups with either preserved (LVEF ≥ 50%) or reduced (LVEF < 50%) systolic function. χ 2 test was used to examine significant differences between nominal classifications. For intra-and interobserver agreement, coefficients of variation (CoV) were calculated by dividing the SD of the differences by the mean. In addition, intraclass correlation coefficients (ICC) were assessed using a model of absolute agreement. There was excellent agreement when ICC > 0.74, good when ICC = 0.60-0.74, fair when ICC = 0.40-0.59, and poor when ICC < 0.4 19 .
Pearsons correlation was performed to calculate correlations between functional and geometric CMR parameters. Two-tailed p-values below 0.05 were considered statistically significant.

Results
Patient characteristics. 20 LVNC patients (mean age 44.7 ± 14.0 years) were included into this study. The age-and sex-matched 20 healthy control subjects showed similar body surface area (BSA) characteristics. Further clinical baseline characteristics are summarized in Table 1. Only 4 patients presented with small LGE (1x anterior, 1x anterolateral and 2x septal) and 14 did not show LGE. One patient did not receive contrast agent and for 1 patient, LGE images were non-diagnostic.
LVNC patients had a significantly lower LVEF in comparison to the controls ( Table 2), but according to recent heart failure guidelines mean LVEF in the patient group was still preserved 20 . Eight LVNC patients showed a (mid-range) reduced LVEF with values < 50%. Despite no difference in LVMi-C, there were significantly higher values in non-compacted parameters as well as their ratios for patients with LVNC. No differences could be seen for RV functional and geometric indices.

CMR -strain. Feature tracking analysis revealed significantly lower GCS and GLS values for the LVNC group
compared to controls (GLS LVNC vs. controls: −17.7 ± 4.2 vs. −22.1 ± 4.1%, p = 0.002; GCS LVNC vs. controls: −24.7 ± 7.6 vs. −30.8 ± 3.8, p = 0.003). On a regional level, GLS parameters were not different for basal and midventricular segments, but for apical ones (Fig. 2a). For GCS, significantly lower values were detected for basal, midventricular and apical slices with the highest differences for apical segments (Fig. 2b). ANOVA analysis indicated a difference of GLS between controls and LVNC patients with LVEF > 50% (N = 12) and LVEF < 50% (N = 8) (p = 0.002). In post-hoc Bonferroni correction, this difference only remained for the comparison of controls and the subgroup of patients with reduced LVEF < 50% (p = 0.001), but not between the two subgroups. For GCS, ANOVA indicated a difference between the groups as well (p < 0.001), which was most pronounced between controls and patients within the subgroup of reduced LVEF (p < 0.001), less pronounced between the two subgroups (p = 0.022). There was no difference between controls and patients within the subgroup of preserved LVEF > 50%. Three patients with LGE presented with GLS and GCS below the mean of the LVNC group.
Besides a correlation with LVEF, further correlation analyses revealed that GLS best correlated to parameters including the noncompacted areas (Table 3), such as LVMi-NC, LVMi-MM, LV-NC and LV-NC/C. The same applied for GCS. In addition, there was an association between decreasing thickness, but not mass, of compacted area (LV-C) and decreasing GCS (R = −0.449, p = 0.004).
Mean differences ±1.96 SD, CoV and ICC for GLS and GCS divided by control and diseased groups are summarized in Table 4 for intra-and interobserver variability. There was good to excellent inter-and intraobserver reproducibility for GLS and GCS in the healthy and diseased groups (ICC 0.68-0.97). Intraobserver agreement showed higher reproducibility than interobserver agreement.
Previous PCI, n(%) 0 (0) 0 (0) 1.000 Family history LVNC, n (%) 7 (35) 0 (0) 0.004   CMR -strain rate. Like GLS and GCS, peak diastolic SR ll and SR cc were significantly lower for LVNC compared to controls (Fig. 3a,b). On a regional level, the difference was most pronounced in midventricular and apical slices for SR ll and in all slices for SR cc (Fig. 3a,b). ANOVA analysis indicated a difference of SR ll between controls and both LVNC subgroups (p = 0.004). In post-hoc Bonferroni correction, this difference only remained for the comparison of controls and patients within the subgroup of reduced LVEF (p = 0.004), but not between the two subgroups. For SR cc , ANOVA indicated a difference between the groups as well (p < 0.001), which was most pronounced between controls and patients within the subgroup of reduced LVEF (p < 0.001), less pronounced between the two subgroups (p = 0.004). There was no difference between controls and patients within the subgroup of preserved LVEF. Like for strain, three patients with LGE presented with SR ll and SR cc below the mean of the LVNC group.
Correlation analyses revealed the best correlation of SR ll and SR cc to parameters of the non-compacted myocardium, such as LV-NC or LVMi-NC (Table 5). For both parameters, there was no correlation to the compacted myocardium.
Mean differences ±1.96 SD, CoV and ICC for SR ll and SR cc divided by healthy and diseased groups are summarized in Table 4 for intra-and interobserver variability. There was good to excellent reproducibility for these parameters in the healthy and diseased groups (ICC 0.9-0.99). Like for global strain, intraobserver agreement showed higher reproducibility than interobserver agreement. The in-house written MATLAB-based script showed excellent reproducibility for rotational parameters in the healthy and diseased groups (ICC 0.95-0.99) ( Table 4). Reproducibility was higher in apical segments for the healthy and in basal segments for the diseased group.

CMR -rotation.
Analyses showed that there was no difference in basal rotation, but a significant difference in maximum apical rotation with lower values for the LVNC group (p = 0.007) (Fig. 4). This difference remained for end-systolic twist with lower values for the LVNC group (p = 0.012). ANOVA analysis indicated a difference of apical rotation and twist between controls and both LVNC subgroups (p = 0.005 and p = 0.007). In post-hoc Bonferroni correction, this difference only remained for the comparison of controls and patients within the subgroup of reduced LVEF < 50% for apical rotation (p = 0.004) and twist (p = 0.006). The 4 patients with LGE showed heterogeneous values in rotational analyses with values below or above the mean.  www.nature.com/scientificreports www.nature.com/scientificreports/ Similar to strain values, twist correlated well with parameters of the non-compacted myocardium, e.g. LV-NC/C (r = −0.454, p = 0.003) ( Table 6).

Discussion
The present study provides CMR myocardial deformation indices in patients with LVNC compared to controls on a global and regional level. GLS and GCS, as well as SR ll and SR cc were found to be significantly impaired in LVNC, especially in midventricular and apical segments with a good correlation to parameters of the non-compacted myocardium. By implementing a software for assessing ventricular rotation, impaired apical rotation and twist with good reproducibility could be shown as well.
During its heterogeneous course, LVNC may lead to heart failure or arrhythmic events 21,22 . Besides TTE, CMR is frequently used to confirm the diagnosis of LVNC or for follow-up strategies. As additional parameter to LVEF, CMR-derived strain can easily be assessed by post-processing standard cine images without the need for further sequences like tagging, or displacement-encoding (DENSE). In addition, CMR yields the advantage of a reproducible, in-plane image acquisition. Therefore, these parameters can be of additional value for the description of functional changes in LVNC 7,23 .
Our results are similar to those obtained by speckle tracking echocardiography that showed impaired GLS, GCS and strain rate parameters in patients with LVNC 6,14 . In addition, a global strain reduction has already been shown in a CMR study of children and adolescents with LVNC 6,10 . However, the disease is mainly manifested in apical and midventricular regions which is mirrored in regional CMR-derived strain parameters for the first time. This can be underlined by the new correlation of strain parameters to indices of noncompacted masses 2 . Therefore, GLS is preserved in basal regions (Fig. 2) and the difference to controls increases from basal to apical regions 6 . Surprisingly, GCS, SR ll and SR cc showed already reduced values in basal regions, although the difference to controls from basal to apical regions increases as well. One influencing factor of this finding can be the heterogeneous group of LVNC patients with different LVEF. Indeed, dividing the group of LVNC patients according to LVEF, patients within the subgroup of reduced LVEF < 50% showed the most prominent reduction in all strain parameters. From hypertrophic cardiomyopathy it is known that CMR strain indices can already be reduced even when LVEF is normal or supernormal, which leads to the hypothesis that strain indices are more sensitive to   www.nature.com/scientificreports www.nature.com/scientificreports/ capture the 3-dimensional nature of myocardial function in certain clinical settings 7 . In the present cohort, strain parameters of LVNC patients within the subgroup of preserved LVEF were not statistically different from strain parameters of controls after Bonferroni correction. However, their values were below values for normal controls and especially GLS and SR ll were not different to patients within the subgroup of reduced LVEF thereby indicating a functional state between healthy and diseased LVNC patients with impaired LVEF.
For a reliable application in clinical settings, reproducibility is important. Overall strain and strain rate assessment showed reproducible assessment with good to excellent consistency of measurements (ICC between 0.68 and 0.99), which is comparable to literature 24 . In addition, the newly-developed tool for rotational analyses was able to provide GCS results with good reproducibility. Although mean differences were slightly higher for GCS, the tendency of over-or underestimating the parameters pointed in the same direction for healthy and diseased patients with GLS being lower for repeated measurements and GCS being higher therefore preserving differences of parameters between the groups. Surprisingly and in contrast to existing literature, parameters of rotation were not reproducible using standard feature tracking software 19 . This might be due to the fact that an average of three measurements was used to calculate a global torsion value in previous studies about reproducibility of torsion. In addition, the current study chose to report inter-observer agreement as well 19 . Nevertheless, using the customized pipeline for the assessment of rotational parameters, similar CoV and ICC could be obtained as presented in literature with the advantage of only one segmentation.
As already shown using only speckle-tracking echocardiography so far, apical rotation and twist was impaired in LVNC 6,14,25 . Due to the still preserved basal rotation, a previously described 'solid body rotation' with near absent LV twist could not be confirmed in the present study 14 . This may be influenced by the different image planning of TTE and CMR. Reproducibility was poorest for diseased patients in apical rotation. Given that the disease mainly affects midventricular and apical regions with lower values, the higher CoV is not surprising 2 .
Limitations. As the incidence of LVNC in the normal population is low, a limitation of the current work is the relatively small sample size of patients with LVNC, especially for the division according to an LVEF threshold 11 . However, this limited population already showed promising data for the description of myocardial strain indices in patients with LVNC and was comparable to patient numbers previously published in the literature 6,14 .   www.nature.com/scientificreports www.nature.com/scientificreports/ As an inherent limitation, FT algorithms can only distinguish the endo-or epicardial interfaces of the myocardium and are, therefore, unable to assess transmural variations of the myocardial strain 18 . In contrast, echocardiographic speckle tracking based assessment of layered strain has already shown different subendo-and subepicardial dynamics in patients after aortic valve replacement 26 . Since LVNC is considered to have differential impact on the myocardial layers, an investigation of the transmural variation of strain would be of interest for future studies.
In addition, global radial and segmental strain were not obtained due to recent evidence of their lower reproducibility 24 . conclusion CMR deformation indices including GLS, GCS and strain rate parameters were reduced in patients with LVNC especially in affected midventricular and apical slices and correlated well with parameters of the non-compacted myocardium. By using a novel approach for post-processing CMR data, reduced twist and apical rotation with a good reproducibility could be shown. CMR-derived deformation indices may show added value to assess functional impairment in LVNC, even when LVEF is preserved.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.