Feature-tracking myocardial strain in healthy adults- a magnetic resonance study at 3.0 tesla

We analyzed feature-tracking derived circumferential and longitudinal strain in healthy volunteers who underwent cardiac magnetic resonance imaging (CMR) at 3.0 T. 88 healthy adults (44.6 ± 18.0 years old, 49% male), without prior cardiovascular disease, underwent CMR at 3.0 T including cine, and late gadolinium enhancement in subjects >45 years. LV functional analysis and feature-tracking strain analyses were carried out. Global strain had better reproducibility than segmental strain. There was a sex specific difference global longitudinal strain (mean ± SD, −18.48 ± 3.65% (male), −21.91 ± 3.01% (female), p < 0.001), but not global circumferential strain (mean ± SD, −25.41 ± 4.50% (male), −27.94 ± 3.48% (female), p = 0.643). There was no association of strain with ageing after accounting for sex for both global longitudinal and circumferential strain. Feature-tracking strain analysis is feasible at 3.0 T. Healthy female volunteers demonstrated higher magnitudes of global longitudinal strain when compared to male counterparts. Whilst global cine-strain has good reproducibility, segmental strain does not.

One of the most important components of a clinical imaging study is the assessment of left ventricular (LV) pump function. The LV ejection fraction (LVEF) is the difference between LV end-diastolic and systolic volumes, divided by the LV end-diastolic volume, hence myocardial contractibility is not directly assessed. Whilst echocardiography is the standard of care, cardiovascular magnetic resonance (CMR) is regarded as the gold-standard for LV functional assessment 1 . The LVEF can be within normal reference ranges in a number of pathological states, which might otherwise have abnormal peak systolic strain values, i.e. identify sub-clinical LV dysfunction. The LVEF cannot be used to provide a detailed assessment of cardiac mechanics due to the complex architectural arrangement of myofibers in circumferential and longitudinal directions. Strain is described as local shortening, thickening and lengthening of the myocardium as a measure of regional and global LV function 2 . Circumferential and longitudinal strain are denoted as negative magnitudes of strain to reflect shortening, whilst radial strain is positive as it reflects myocardial thickening 3 .
Feature-tracking is a technique which has gained traction since being described by Hor et al., in 2011 4 , and has resulted in the assessment of myocardial strain from routinely acquired cine imaging sequences in a myriad of pathologies [5][6][7][8][9][10] . The estimation of strain is reasonably quick 11 . Feature-tracking algorithms are designed to focus on border displacement, with a stronger weighing of endocardial deformation explaining some of the differences in results found in direct comparisons of feature-tracking and other strain modalities 12,13 . Feature-tracking uses optical flow 14 to track myocardial borders, by tracking a number of points using both 1D and 2D techniques through the cardiac cycle 4 . Feature-tracking has been clinically validated against tagging [13][14][15][16] .
There has been an increase in clinical cardiac MR imaging performed at 3.0T 17 . There are a number of advantages to utilizing 3.0 T MR scanners, notably an increase in signal-to-noise ratio, and image resolution 18,19 . Schuster et al. 20 reported that intra-observer variability in cine-strain assessment with feature-tracking at 3.0 T is similar to that observer at 1.5 T. However, balanced steady state free precession imaging is more likely to experience artifact related to the high magnetic field 21 and feature-tracking reference ranges for healthy volunteers at 3.0 T are currently unavailable with clinically approved (for example, Food and Drug Administration approved) feature-tracking software. A recent meta-analysis 22 on cine-strain has reported there has been one publication looking at strain using an investigational tissue-tracking software in healthy Chinese volunteers 23 .
There is an increasing body of evidence of the incremental utility of strain in patients with dilated cardiomyopathy 24 , post myocardial infarction 10,25 , and congenital heart disease 5 . Health volunteer reference ranges are required to identify minor reductions in strain parameters when conventional parameters of function, such as the LV ejection fraction are unchanged 15,26,27 .
We aimed to assess circumferential and longitudinal myocardial strain utilizing feature-tracking at 3.0 T in healthy volunteers to provide reference ranges and to investigate the influence of age and sex on strain. We did not investigate radial strain due to reported inferior reproducibility 15,28,29 .

Characteristics of The Study
Participants. The characteristics of the participants (n = 88) and their LV mass and function are presented in Table 1.
Inter-observer Analysis. All cine imaging was of diagnostic quality. Global longitudinal strain had excellent reproducibility both with inter-and intra-observer analyses and strong positive correlations. Global circumferential strain had excellent reproducibility, as identified by the intra-class correlation co-efficient and strong positive correlations (Fig. 1). Segmental longitudinal strain analyses had a higher bias than global strain analyses, with lower intra-class correlation co-efficient for both intra-and inter-observer analyses. Correlations between reproducibility analyses for segmental longitudinal strain were moderately strong ( Table 2). Segmental circumferential strain reproducibility analyses revealed higher biases than global circumferential strain parameters, with a lower intra-class correlation co-efficient, and a lower correlation strength (Fig. 2, Table 2).
Myocardial Strain. Feature-tracking circumferential and longitudinal strain was analyzable for all participants.
Strain and healthy ageing. Both global circumferential (R = −0.12), and longitudinal strain (R = −0.17) had a poor negative correlation with healthy ageing We carried out linear regression analyses accounting for the effect of sex as well as age. There was no association between global circumferential or longitudinal strain and ageing, after accounting for gender (Table 5).

Discussion
We have investigated the feature-tracking derived circumferential and longitudinal strain in a large sample of healthy volunteers at 3.0 T utilizing commercially available feature-tracking software. The main findings of this study are that: (1) Global circumferential and longitudinal strain have good reproducibility, unlike segmental strain.
(2) There is a gender specific difference in magnitudes of longitudinal but not circumferential strain.
(3) There are no changes in magnitudes of strain with healthy ageing.
A number of studies have assessed reproducibility of feature-tracking derived strain, in healthy volunteers 20,30 and in patients 31 . There is concern that unlike global strain 31 , segmental strain derived with feature-tracking is not yet ready for clinical use because of poor reproducibility 2,13,27 . In our study, we identified that inter-observer www.nature.com/scientificreports www.nature.com/scientificreports/ analyses of segmental strain yielded moderate (ICC = 0.66, segmental circumferential strain) and good (ICC = 0.77 segmental longitudinal strain) reproducibility as assessed by intra-class correlation co-efficient, associated with a significant mean bias and standard deviations, in keeping with what has been previously reported 2,13,27 . This has important implications when it comes to choosing the strain modality to be used in studies investigating regional myocardial biomechanics.
In our study we identified that females generate larger magnitudes of longitudinal strain, whilst the observed difference in circumferential strain between different genders was not statistically significant. This result is in keeping with other studies using feature-tracking to look at healthy volunteers. Augustine et al. 15 (n = 145, 37%  www.nature.com/scientificreports www.nature.com/scientificreports/ male), and Taylor et al. 26 (n = 100, 50% male) reported that circumferential strain was not associated with sex, whilst André et al. 32 (n = 150, 50% male) and Liu et al. 23 (n = 130, 46% male), identified a statistically significant difference in circumferential strain between the sexes, with females having larger magnitudes of strain There was a significant difference in longitudinal strain magnitudes between sexes in our study in keeping with Liu et al. 23 , Augustine et al. 15 , Taylor et al. 26 and André et al. 32 . This difference. The difference in strain magnitudes could be partly attributed to the gender specific difference in myocardial volumes 34 which would imply that greater myocardial shortening would be required to generate similar cardiac output between sexes.
In our study, we did not observe a relationship between feature-tracking strain and ageing. Looking to the literature, André 32 using feature-tracking, Oxenham 35 using tagging, and Neizel 36 using strain-encoded CMR did not identify any association between age and strain. Taylor 26 reported an age related increase in circumferential but not longitudinal strain, but no information was provided on the potential associations with sex. Kuznetsova et al. 37 using echocardiography in 236 healthy volunteers reported an inverse association between longitudinal strain and age.
In conclusion, we have described circumferential and longitudinal strain at 3.0 Tesla in a reasonably large sample of healthy adults across a broad age range with feature-tracking software. We have observed that longitudinal and circumferential strains varied in a regional distribution with higher strain values in the anterior and lateral LV territories. Longitudinal strain values were higher in females than in males. There was no age related difference in strain after accounting for age.

Methods
Study Population. The UK Research Ethics Service (ethics reference 11/AL/0190) approved the study, all of the participants provided written informed consent and all studies were performed in accordance with relevant guidelines 38 . Healthy volunteers aged at least 18 years with no prior medical history (including cardiovascular health problems, medication or systemic illness) were invited to participate by placing advertisements in public buildings (e.g. hospital, university). The other exclusion criteria included standard contraindications to MR (e.g.       Table 5. Associations of strain, sex and age. CI-confidence intervals.
www.nature.com/scientificreports www.nature.com/scientificreports/ MR protocol. LV dimensions were assessed using b-SSFP cinematographic breath-hold sequences. Typical imaging parameters are as shown in Table 6. The heart was imaged in multiple parallel short-axis planes 7-mm thick separated by 3 mm gaps, as well as in the 2-chamber, 3-chamber, and 4-chamber long-axis views.
Participants over 45 years of age had their renal function checked and if the estimated glomerular filtration rate (eGFR) was > 30 mls/min/1.73 m 2 gadolinium contrast was administered (0.15 mmol/kg per bolus of gadolinium diethyltriaminepenta -acetic acid (Gd-DTPA, Magnevist, Bayer Healthcare). Late gadolinium enhancement images covering the entire LV were acquired 10-15 minutes after intravenous contrast agent administration using segmented phase-sensitive inversion recovery (PSIR) turbo fast low-angle shot sequence. Image Analysis. Data sets were anonymised to ensure operators were blinded to all other data. The absence of late gadolinium enhancement (myocardial fibrosis or scar) was determined qualitatively by visual assessment by D.C. (>3 years CMR experience) and C.B. (>10 years CMR experience). The absence of myocardial late gadolinium enhancement was another requirement for inclusion of the data in this analysis.
LV mass and function were analyzed in randomly ordered, de-identified scans by CMR-trained cardiologists using computer-assisted planimetry (Syngo MR ® , Siemens Healthcare, Erlangen, Germany) as previously described 27 .
Feature-tracking analysis. 3 long-axis (horizontal long axis, vertical long axis, and left ventricular outflow tract views) and 3 short-axis (basal, mid-LV, apical) slices were chosen per each volunteer. The mid-left ventricular short axis slice was chosen as the equidistant slice between the mitral valve plane and the LV apex. The LV was segmented using the anterior right ventricular-LV insertion point as the reference point. Diogenes CMR feature-tracking software (TomTec Imaging Systems, Germany) was used to quantify strain from short axis cine images at mid-left ventricular level. The operators derived strain following a standard protocol taught by the software manufacturer 15,39 .
To minimise observer bias, 10 datasets were identified at random and coded using a different code sequence to the main dataset, which was disclosed after the analysis was performed, a week apart by 2 analysts. Statistical Analysis. Statistical analysis was performed using SPSS software (SPSS Inc, Chicago, IL, USA, version 22), R V.2.15 or higher (R Foundation for Statistical Computing, Vienna, Austria). Normality was tested using the Shapiro-Wilk test. Continuous variables were expressed as mean ± standard deviation (SD). Student's t-test was used to compare means. Linear regression was used to investigate the association of age and sex with strain. Inter-and intra-observer reproducibility was assessed using Bland-Altman statistics, intra-class correlation co-efficient and Pearson correlation.
A p-value of <0.05 was considered statistically significant.
Disclosures. The University of Glasgow holds a research agreement with Siemens Healthcare.  Table 6. Typical imaging parameters, at 3.0 T MR field strength. TR: repetition time (ms); TE: echo time (ms); FoV: field of view (mm).