Author + information
- Received September 29, 2016
- Revision received December 6, 2016
- Accepted December 9, 2016
- Published online February 5, 2018.
- aMenzies Institute for Medical Research, University of Tasmania, Hobart, Tasmania, Australia
- bBaker IDI Heart and Diabetes Institute, Melbourne, Victoria, Australia
- ↵∗Address for correspondence:
Dr. Kazuaki Negishi, Menzies Institute for Medical Research, University of Tasmania, 17 Liverpool Street, Private Bag 23, Hobart, 7000, Australia.
Objectives The aim of this study was to perform a systematic review and meta-analysis to estimate the normal ranges of magnetic resonance imaging (MRI)-based feature tracking (FT) and to identify sources of variations. Similar analyses were also performed for strain encoding, displacement encoding with stimulated echoes, and myocardial tagging.
Background MRI-FT is a novel technique for quantification of myocardial deformation using MRI cine images. However, the reported 95% confidence intervals (CIs) from the 2 largest studies have no overlaps.
Methods Four databases (EMBASE, SCOPUS, PUBMED, and Web of Science) were systematically searched for MRI strains of the left (LV) and right (RV) ventricles. The key terms for MRI-FT were “tissue tracking,” “feature tracking,” “cardiac magnetic resonance,” “cardiac MRI,” “CMR,” and “strain.” A random effects model was used to pool LV global longitudinal strain (GLS), global circumferential strain (GCS), global radial strain (GRS), and RVGLS. Meta-regressions were used to identify the sources of variations.
Results 659 healthy subjects were included from 18 papers for MRI-FT. Pooled mean of LVGLS was −20.1% (95% CI: −20.9% to −19.3%), LVGCS −23% (95% CI: −24.3% to −21.7%), LVGRS 34.1% (95% CI: 28.5% to 39.7%), and RVGLS −21.8% (95% CI: −23.3% to −20.2%). Although there were no publication biases except for LVGCS, significant heterogeneities were found. Meta-regression showed that variation of LVGCS was associated with field strength (β = 3.2; p = 0.041). Variations of LVGLS, LVGRS, and RVGLS were not associated with any of age, sex, software, field strength, sequence, LV ejection fraction, or LV size. LVGCS seems the most robust in MRI-FT. Among the MRI-derived strain techniques, the normal ranges were mostly concordant in LVGLS and LVGCS but varied substantially in LVGRS and RVGLS.
Conclusions The pooled means of 4 MRI-derived myocardial strain methods in normal subjects are demonstrated. Differences in field strength were attributed to variations of LVGCS.
Cardiac wall motion analysis plays a central role for assessment of ventricular contractile function. Cardiac magnetic resonance (CMR) is a radiation-free, reference standard for assessment of cardiac anatomy and wall motion because of its excellent endocardial border definition due to high spatial and contrast resolution. However, the current assessment of wall motion is primarily subjective, and results are skill and experience dependent. The quantification of myocardial deformation provides further insights into cardiac function in a variety of subclinical cardiac diseases (1,2). Although several quantitative assessment techniques have been proposed, such as myocardial tagging (MT), phase contrast velocity imaging, displacement encoding (DENSE), and strain encoding (SENC) for strain analysis (3,4), these methods require additional sequences and time.
Recently, magnetic resonance imaging (MRI) feature tracking (MRI-FT) has been introduced using cine images and provides a fast and accurate assessment of both ventricular (5–12) and atrial strains (13,14) with speckle tracking echocardiography (STE) or MT (15). To date, however, only a few data exist on MRI-FT normal reference values, and they are based on small or modest sample-size studies. The reported 95% confidence intervals (CIs) from the 2 largest MRI-FT studies have no overlaps (16,17). Therefore, in this study, we aimed to: 1) perform a systematic review for studies that reported MRI-FT strain values from a normal healthy population; 2) estimate the pooled means of their myocardial strains by meta-analysis; and 3) elucidate possible sources of variation affecting the strain values by meta-regression analyses. We also performed the same systematic review and meta-analysis for DENSE, SENC, and MT.
We followed the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analysis) guideline when performing our systematic review and meta-analysis (18). The first search was performed on 4 August 2015, and the last search was performed on 21 October 2015. Four databases (EMBASE, PUBMED, Scopus, and Web of Science) were systematically searched for the strain values of the left ventricle (LV) or right ventricle (RV) derived from the MRI-FT technique by 2 coauthors (H.Q.V. and K.N.) under the guidance of a librarian trained in systematic review. The key terms were “tissue tracking,” “feature tracking,” “cardiac magnetic resonance,” “cardiac MRI,” “CMR,” and “strain.” The reference lists of these articles were also scrutinized to identify some additional appropriate studies. Search hedges created are listed in Online Appendix A. The study was prospectively registered with the PROSPERO database of systematic reviews (Normal Ranges of Bi-Ventricular Strain by MRI Feature Tracking: A Systematic Review and Meta-Analysis; CRD42015025616). Methods for DENSE, SENC, and MT are reported in Online Appendices B and C.
From these lists, studies were included if the articles reported strain values using cardiac MRI in healthy subjects. The 2 coauthors reviewed and chose studies if the studies met each of following criteria: 1) studies recruited extensively normal healthy subjects; and 2) studies included a control group, which was defined as normal and healthy. The definition of the normal healthy group varies with the studies. In this meta-analysis, this group of subjects was identified if subjects: 1) were not associated with any disease (diabetes, heart failure, and so on) or with overt symptoms or adverse outcomes; 2) did not have any history of heart disease; 3) were not currently implanted with any cardiac devices; 4) were indicated as being healthy; and 5) were >18 years of age. The definitions of healthy subjects are shown in Online Table S1. All discrepancies were reviewed and resolved by consensus of all authors.
The search was uniquely concentrated on human studies, published in English. Animal studies and conference presentations were excluded.
Strain data were extracted from individual studies and entered into an electronic database. Left ventricular global longitudinal strain (LVGLS), left ventricular global circumferential strain (LVGCS), left ventricular global radial strain (LVGRS), and right ventricular global longitudinal strain (RVGLS) were extracted from text, tables, and graphs. In cases where we believed that multiple articles came from a single dataset, the largest study was selected.
All demographic, common clinical characteristics, and strain information were extracted from texts and tables. In cases where the same subjects were measured several times in a day by using the same equipment (7), the first dataset was used. A study applied 2 different software programs to the same population (12). Only the data from 1 software program (TomTec Imaging Systems, Unterschleissheim, Germany) were selected and used because the vast majority of the articles used this software.
Outcomes of interest
In this meta-analysis, our outcomes of interest were normal ranges of LV and RV strains (LVGLS, LVGCS, LVGRS, and RVGLS) measured by MRI-FT.
The means and 95% CIs of LVGLS, LVGCS, LVGRS, and RVGLS were computed using random effects models weighted by inverse variance. Funnel plots with and without the Duval and Tweedie trim and fill were constructed, and the Egger test was used to assess potential publication bias. The heterogeneity between subgroups or between studies was assessed by the Cochran Q test and the inconsistency factor (I2). Meta-regressions were performed for each risk factor to examine possible study factors associated with heterogeneity. Beta coefficient and its CIs were derived using the least-mean squares fitting method. Statistical analysis was performed using R version 3.2.2 (The R Foundation for Statistical Computing, Vienna, Austria) with the “metafor” package. Two-tailed p values were applied, and the threshold of statistical significance was 0.05 except for the Egger test, where 0.1 was used.
For MRI-FT, 416 titles were matched with the key terms from the 4 databases (EMBASE [n = 218], PubMed [n = 64], Scopus [n = 59], and Web of Science [n = 75]) (Figure 1). Eighteen valid studies (659 normal participants) met the selection criteria and were included in this meta-analysis, where 17 were eligible for LVGCS, 12 for LVGLS and LVGRS, and 9 for RVGLS.
Most subjects were middle-aged (Table 1). Many studies (15 of 18 studies) had a small sample size (n ≤ 50), and only 3 studies had a sample size ≥100, with a maximum of 150. Thus, sensitivity analyses were based on the sample size.
Software used for MRI-FT was also collected to find the association of software and the variation of strain. A majority of included studies used software from 1 vendor (Diogenes or 2D CPA, TomTec Imaging Systems), the other software included Velocity Vector Imaging (Siemens Medical Solutions, Malvern, Pennsylvania), Circle (Circle Cardiovascular Imaging, Calgary, Alberta, Canada) (12), and MTT (Toshiba Medical Systems, Tochigi, Japan) (8). Further detailed information can be found in Online Table S2.
Normal ranges of MRI-FT
The pooled mean of LVGLS was −20.1% (95% CI: −20.9% to −19.3%) (Figure 2). Among 12 studies, 9 reported LVGLS from the apical 4-chamber view only and 3 from the 3 apical views, where their values were quite similar. Although no significant publication bias was identified by the funnel plot (Online Figure S1) and the Egger test, there was a significant heterogeneity in LVGLS. A univariable meta-regression was performed to find factors that have significant contributions to the heterogeneity (Online Table S3). LVGLS was not associated with sex, field strength, sequence, scanner vendor, or LV ejection fraction. All studies reported LV GLS used the same software.
LVGCS and LVGRS
The pooled means of LVGCS and LVGRS were −23.0% (95% CI: −24.3% to −21.7%) and 34.1% (95% CI: 28.5% to 39.7%). Among the 17 articles for LVGCS, 10 reported LVGCS from only 1 short-axis level (the papillary muscle level) and the remaining 7 from 3 short-axis levels. Their reported LVGCS values also overlapped (Figure 3). Similarly, 7 of 12 reported LVGRS from 1 level only, and the remaining 5 studies, from 3 levels, also showing significant overlap (Figure 4).
The Egger tests indicated a publication bias for LVGCS (p = 0.07) (Online Figure S1), but not in LVGRS. Normal range in LVGCS from the Duval and Tweedie trim and fill process was quite similar. Sample size was not associated with the LVGCS heterogeneity in both meta-regression, (β = −0.016; p = 0.3) and cumulative forest plot (Online Figure S2). Meta-regression analyses of LVGCS showed significant contributions of field strength (Online Table S3). Among our hypothesized confounding factors, there were no confounders that can significantly explain the heterogeneity of LVGRS.
All 9 studies reported RVGLS from the 4-chamber view only and the pooled mean was −21.8% (95% CI: −23.3% to −20.2%) (Figure 5). Although no publication bias was seen, there was significant heterogeneity. In meta-regression analyses, none of age, sex, field strength, sequence, LV ejection fraction, or software vendor was associated with the variability of RVGLS.
Table 2 summarizes coefficient of variance of the included articles. LVGCS tends to show smallest interobserver and intraobserver variability in MRI-FT strain.
Subsequent sensitivity analyses based on sample size (all ≥20 or ≥100) revealed no obvious effect of sample size (Online Figure S3). Additional sensitivity analyses limiting articles using TomTec software showed similar results (Online Table S4).
Similar investigations were performed for SENC, DENSE, and MT. Results are summarized in Table 3 and Online Figures S4 to S7, and Online Tables S5 to S8. MRI-FT and SENC share similar LVGLS, but MT showed smaller. MRI-FT demonstrated slightly larger LVGCS than the other MRI strain techniques. There were substantial heterogeneities in normal ranges of LVGRS and RVGLS.
This is the first systematic review and meta-analysis, to our knowledge, of pooled mean of MRI-FT among normal subjects. Although MRI-FT has substantial potential, the normal ranges from the 2 largest studies have no overlaps. This really hampers wider use of this technique. Our results warrant the need for a larger-scale study determining normal ranges for FT strains, preferably with some standardizations for the number of views used. In the meantime, estimated means (with 95% CI) would serve as a reasonable guide for end users. We also performed similar analyses for other MRI-derived strains such as MT, SENC, and DENSE as comparators.
Software from 1 vendor dominated currently, and more than one-half of the studies measured LV strains from a single view, but this had a minimal impact on these measurements. Most articles used steady-state free precession (SSFP) or balanced SSFP cine images (Online Table S1). The main reasons for this are: 1) a routine CMR technique; 2) high contrast-to-noise ratio at the endocardial–blood interface; 3) little blood flow dependency; 4) higher temporal resolution; and 5) short acquisition time (8). The other sequence used was fast gradient echo cine (8). Their strain values were similar to those from SSFP but required a longer acquisition time.
Most of studies included had a small sample size (n ≤ 50), and only 3 studies had sample size ≥100, with a maximum of 150, with limited overlap among their normal ranges. This may be because CMR requires more resources than other noninvasive cardiac modalities. LVGLS was minimally affected by the sample sizes. LVGCS from small-sized studies seemed to show smaller values, although this was not a significant determinant of LVGCS in the meta-regression.
More than one-half of the studies derived LVGCS and LVGRS from a single slice at the papillary muscle level only, and the rest used 3 short-axis planes although their normal ranges overlapped (Figures 3 and 4). Similarly, most of the LVGLS were calculated from the 4-chamber view only, whereas only 3 papers used 3 apical views. They are also overlapped (Figure 2). Nevertheless, the use of a single-plane method should depend on the underlying disease: if a heterogeneous distribution of disease process can be anticipated, the 3-level method would reflect more accurate deformation of the whole heart.
Although the normal values of LVGLS and LVGCS fluctuated in a narrow range, LVGRS had a wider range of CIs. Online Figure S3 also reflects this variation, where LVGRS for different sample sizes yielded an unpredictable pattern. Unfortunately, our meta-regression analyses did not identify the source of this variation. The same issue has been reported in STE (19), and its causes remain contentious. We speculate that through-plane motion could be a partial explanation for this problem.
Calculation of global strain
Another point to be mentioned is that there are 2 ways of calculating global strain: 1) an average of the peaks of individual strain curves; and 2) peak of the mean curve. Only 1 paper clarified which method was used to obtain global strains. Although a high correlation between the 2 methods was reported (16), this ambiguity caused difficulties and bias in this meta-analysis. In this context, only results from the second method were included. In addition, a few studies measured LVGLS using 3 apical views and LVGCS and LVGRS from 3 short-axis planes.
Heterogeneity among studies
In our meta-analysis, each of the strain metrics had large I2 values. The interpretation of I2 is discussed in Online Appendix F. Most of our hypothesized confounding factors were not fully explanatory for this. This problem could be further explained by 3 potential reasons: 1) Population. Normal strain values among different populations may slightly differ. 2) Interobserver variability. Interobserver variability could be another source of variation among studies. Differences defining myocardial contours may result in between-study heterogeneity. Table 2 summarizes interobserver and intraobserver variability of included studies. The coefficients of variation of the interobserver variability ranged from 3.7% to 32.2%, whereas those of intraobserver variability were 2.7% to 43.5%. LVGCS seems to be the most robust. 3) Different software vendors also could be a source of heterogeneity. However, our sensitivity analysis including studies using TomTec only (Online Tables S4a and S4b) did not show substantial reduction in I2.
Comparisons with normal ranges among various methods
Normal ranges of STE, MT, SENC, DENSE, and MRI-FT are summarized in Table 3. Although MRI-FT-derived LVGLS and LVGCS yield quite similar ranges to those of STE, there are discrepancies in LVGRS and RVGLS.
MT is an MRI technique that generates grid patterns or parallel lines on the magnitude-reconstructed images (spatial modulation of magnetization [SPAMM] or complementary SPAMM [CSPAMM]), which are then analyzed, or by extracting information about myocardial tags in k-space (harmonic phase imaging [HARP]). The strength of MT is its insensitivity to through-plane motion (17). However, MT suffers a low temporal resolution, low signal-noise ratio, and requires some extra sequences, prolonged image acquisition, and a longer breath-hold. The agreement between MRI-FT and MT still remains controversial. Some reported a poor agreement (20), others showed high correlations (17,21). In this study, LVGLS and LVGCS by MT were somewhat less negative than those by STE and MRI-FT. This could be due to its insensitivity to through-plane motion.
DENSE is another technique for strain calculation, first introduced in 1999 (22). The phase-reconstructed images by DENSE have high temporal and spatial resolution, and direct extraction of motion data (23,24). However, this technique requires some specific sequences, which may prolong scan time. So far, applications of DENSE in clinical practice have been still limited by small sample size and only applied for short-axis images. The pooled normal ranges of DENSE were lower than MRI-FT. Due to a few studies that reported LVGRS by DENSE, 95% CI varied in a very wide range.
SENC is developed on the concepts of myocardial MT, but it uses tag planes parallel to the image plane (25). In other words, LVGLS is obtained from short-axis views, and LVGCS from long-axis views. In general, normal ranges of SENC were similar to those by MRI-FT. Interestingly, SENC can provide right ventricular global circumferential strain (RVGCS). Between-study variability of both LVGCS and RVGCS by SENC could be explained by the proportions of males. Additionally, that of RVGCS could be also explained by age, sex, and software vendors.
Several factors merit consideration in the interpretation of our results. First, like all meta-analyses, this work is limited by variations in the original studies and publication bias, although we followed standard approaches to detect this. Likewise, the constituent observational studies may be limited by biases in the recruitment process. Second, we have assumed that all of the measurements were performed by experts, but the levels of expertise among individuals who have actually measured the strain are uncertain. Third, significant heterogeneities among studies were identified. Thus, we performed subsequent meta-regression analyses and stratifications to elucidate the sources of the variations. Fourth, as mentioned earlier in the text, most papers had sample sizes of <50, and studies with larger sample sizes are needed for a more accurate estimation of normal ranges in MRI-FT, especially in RVGLS. Fifth, our study may not have enough power to test vendor differences because only 3 studies reported non-TomTec software data. Only 1 MRI-FT study performed a head-to-head comparison (12). Further studies on this issue should be warranted because this could be a modifiable issue as shown in STE (26). Sixth, the high intrastudy and interstudy variability, and the systematic differences between studies cause difficulties in deriving clear normal ranges. Finally, strain is affected by loading conditions, but we had insufficient data to analyze this.
The pooled means of MRI-FT strains are similar to those of STE. Differences in sequence and software were attributed to variations of LVGRS and LVGCS, respectively. LVGLS and RVGLS variations seemed likely not to be attributed to any of age, sex, software, field strength, or sequence.
COMPETENCY IN MEDICAL KNOWLEDGE: Reported 95% CIs from the 2 largest studies on normal ranges of MRI-based feature tracking strain have no overlaps. The pooled means of myocardial strain in normal subjects are demonstrated via systematic review and meta-analysis in 4 different MRI-derived methods (MRI-based feature tracking, strain encoding, displacement encoding with stimulated echoes, and myocardial tagging).
TRANSLATIONAL OUTLOOK: Although MRI-based feature tracking strain has substantial potential, our results warrant the need for a larger-scale study determining normal rages, preferably with some methodological standardizations including the number of views used. In the meantime, estimated means in MRI-based strain methods would serve as a reasonable guide for end users.
The authors thank Ms. Elizabeth Seymour for her support in systematic review.
Dr. Negishi is supported by an award from the Select Foundation, which had no role in the preparation of this paper. The authors have reported that they have no relationships relevant to the contents of this paper to disclose. Raymond Kim, MD, served as the Guest Editor for this paper.
- Abbreviations and Acronyms
- confidence interval
- cardiac magnetic resonance
- displacement encoding with stimulated echoes
- left ventricle/ventricular
- left ventricular global circumferential strain
- left ventricular global longitudinal strain
- left ventricular global radial strain
- magnetic resonance imaging
- magnetic resonance imaging feature tracking
- myocardial tagging
- right ventricle/ventricular
- right ventricular circumferential strain
- right ventricular global longitudinal strain
- strain encoding
- steady-state free precession
- speckle tracking echocardiography
- Received September 29, 2016.
- Revision received December 6, 2016.
- Accepted December 9, 2016.
- 2018 American College of Cardiology Foundation
- Kempny A.,
- Fernández-Jiménez R.,
- Orwat S.,
- et al.
- Ohyama Y.,
- Ambale-Venkatesh B.,
- Chamera E.,
- et al.
- Habibi M.,
- Lima J.A.,
- Khurram I.M.,
- et al.
- Moody W.E.,
- Taylor R.J.,
- Edwards N.C.,
- et al.
- Hor K.N.,
- Gottliebson W.M.,
- Carson C.,
- et al.
- Mangion K.,
- Clerfond G.,
- McComb C.,
- et al.
- Orwat S.,
- Kempny A.,
- Diller G.P.,
- et al.