Abstract
Background Left ventricular (LV) global longitudinal strain (GLS) and circumferential strain (GCS), and left atrial (LA) strain (LAS) are indicators of poor clinical prognosis. However, it is unclear how they relate to each other and to LV and LA geometry. The aim was to clarify these relationships to inform clinical and research applications.
Methods Patients referred for cardiovascular magnetic resonance imaging were retrospectively identified. Univariable and multivariable linear regression models evaluated associations between GLS, GCS, LAS, LV mass, the volumes and dimensions of the LV and LA, and mitral annular plane systolic excursion (MAPSE).
Results In patients (n=66, median [interquartile range] age 62 [53–72] years, 82% male, LV ejection fraction 48 [34–56]%, range 6–69%), GLS associated with both GCS (R2=0.86, p<0.001) and LAS (R2=0.51, p<0.001), and LAS associated with GCS (R2=0.42, p<0.001). GLS, GCS, and LAS were all univariably associated with MAPSE, LV mass, and the volumes and dimensions of the LV and LA (p<0.001 for all). In multivariable analysis, GLS associated with MAPSE and LV length (R2=0.85, p<0.001); GCS with MAPSE, LV end-systolic volume, and LV mass (R2=0.80, p<0.001); and LAS with LA end-diastolic volume and MAPSE (R2=0.67, p<0.001).
Conclusions MAPSE and LV length alone can accurately estimate GLS. GLS and GCS provide similar information. LV and LA strains can be understood as geometrically coupled composite measures of MAPSE, and the size, function, and dimensions of the LV and LA. The composite of these geometrical relationships likely explains the excellent prognostic strength of strain measures.
Introduction
Strain of the left ventricle (LV), measured as global longitudinal strain (GLS) or global circumferential strain (GCS), and strain of the left atrium (LA), measured as left atrial strain (LAS), are increasingly being used for assessment of LV function and for patient prognosis. GLS and GCS have been identified as more sensitive 1 and reproducible 2 parameters of LV systolic function compared to LV ejection fraction (LVEF), and can provide incremental prognostic value beyond LVEF 3. While LVEF is used to characterize heart failure (HF) with reduced LVEF (HFrEF) and HF with preserved LVEF (HFpEF), LV strain has been shown to be a more accurate measure for detecting the presence of LV dysfunction and evaluating disease progression over time 4,5.
In patients with HFpEF, strain is often impaired 6. Mathematical modelling has been used to describe how GLS, GCS, LV wall thickness, and LV end-diastolic volume associate with LVEF 7. Those authors concluded that changes to LV end-diastolic volume and mass explain how LVEF may be preserved despite a reduced GLS. Moreover, LAS, also referred to as LA reservoir strain, has emerged as a promising measure in the assessment of LV diastolic dysfunction. LAS has been shown to correlate more closely with invasive measurements compared to conventional diastolic measures including e’, E/e’, E/A, and left atrial volume index 8,9. Notably, the LV apex and the posterior aspect of the LA remain relatively stationary throughout the cardiac cycle. Consequently, LV and LA strain measures are inherently coupled through the atrioventricular valve plane displacement, often referred to as mitral annular plane systolic excursion (MAPSE). Not surprisingly, GLS and LAS have been shown to be closely correlated, and GLS and LA volume alone have been shown to provide an accurate estimate of LAS 10.
Strain analysis is performed by tracking the myocardial deformation over the course of the cardiac cycle using dedicated state-of-the-art image analysis software. The use of such dedicated software could add perceived complexity to a parameter that is relatively straightforward. Mathematically speaking, strain is no more than a fractional change in length for what is effectively a mid-wall line through the myocardium, which is measured in either the longitudinal (GLS, LAS) or circumferential (GCS) direction. However, an understanding of the geometric relationships associated with these strain measures might provide insight into the clinical interpretation of these measures. Therefore, the aim of this study was to empirically clarify how LV and LA strain measures relate to conventional measures of LV and LA dimensions and function and to each other, in order to inform the understanding and interpretation of LV and LA strain measures in clinical and research applications. The hypothesis was that strain measures are highly correlated with each other and with conventional LV measures, and that these measures can be used to approximate strain measures with high accuracy.
Methods
Population
We retrospectively screened patients that had undergone clinically indicated cardiovascular magnetic resonance imaging (CMR). The study was approved by the local human subject research ethics committee and all participants provided written informed consent. The population was selected to represent a variable spectrum of LVEF, including 20-25 patients from each of the following ranges: LVEF<30%, 30-40%, 40-50%, 50-60%, and >60%, while excluding those with distinct myocardial pathologies such as hypertrophic cardiomyopathy, amyloidosis, hemosiderosis, Anderson-Fabry’s disease, congenital heart defects, atrial fibrillation during CMR, or insufficient image quality to perform LV strain analysis. Further exclusion criteria included inadequate tracking of the myocardium, defined as visually inadequate tracking of >2 segments in any of the long-axis or short-axis views, and visually apparent foreshortening or the presence of ≥3 of the atrial appendage and pulmonary veins in both 2- and 4-chamber views.
CMR imaging
Clinical CMR scans were performed at 1.5T or 3T MAGNETOM (Aera or Skyra, Siemens Healthcare, Erlangen, Germany) with an 18-channel and 32-channel phased-array body and spine coil, respectively. The exam included retrospectively ECG-gated balanced steady-state free precession cine imaging in short axis (8 mm slice thickness, 1.6 mm slice gap) and long axis 2-, 3- and 4-chamber views (8 mm slice thickness) with 30 phases per cardiac cycle.
Post-processing analysis
All post-processing analyses were performed using freely available software Segment version 3.0 R9405e (http://segment.heiberg.se). LV volumes were assessed by manual delineation of the endo- and epicardium according to guidelines (Schulz-Menger et al., 2020), excluding papillary muscles for LV mass assessment. LV GLS, LV GCS were separately analyzed using a semi-automated feature-tracking module using cine images by a single observer (FF) blinded to clinical data and geometric measures. Endo- and epicardial borders were manually traced in end-diastole and automatically propagated throughout the cardiac cycle. LV GLS was measured as the mean longitudinal strain in two-, three- and four-chamber views, and LV GCS as the mean circumferential strain in an apical, midventricular, and basal short-axis view. Delineations were adjusted manually if ≥2 segments did not track myocardial motion adequately as assessed visually.
LA dimensions and volumes were measured using the bi-plane area-length method in the two- and four-chamber views by a blinded single observer (DS), excluding the left atrial appendage and pulmonary veins. LAS was assessed as the strain measured from end diastole to end systole in a two- or four-chamber view or the mean of both. LV and LA measurements were performed by two different observers.
MAPSE was measured as the mean distance in millimeters traveled by the mitral annular insertion points from end-diastole to end-systole as the average of two manual caliper measurements per view in all three long-axis views. LV length (LVL) was measured as the distance from the most apical point of the epicardium to the midpoint between the mitral annular insertion points in end-diastole. The longitudinal contribution to LV stroke volume was calculated as the maximal LV short-axis epicardial area multiplied by MAPSE as previously described and validated (Carlsson M, et al., 2007).
Statistical analysis
Statistical analysis was performed using RStudio Team (2022). Patient characteristics were summarized as the median [interquartile range] for continuous variables and as counts and percentages for categorical variables. Univariable linear regression was used to assess correlations. Multivariable linear regression was used to identify measures associated with GLS, GCS, and LAS using models with variables selected using variable inflation factors (VIF) <5 to reduce the effects of multicollinearity. Agreements between GCS and GLS were assessed using Bland-Altman plots. A p-value less than 0.05 was considered statistically significant.
Results
Patient characteristics
A flowchart of patient selection is shown in Figure 1. The final study population included 66 patients: 14 (18%) were female, age was 62 [53–72] years and LVEF was 48 [34–57]%, full range 9–69%. Characteristics of the study population are shown in Table 1.
Global longitudinal strain
A summary of the relationships between strain values and geometric measures is illustrative in representative patients in Figure 2. GLS was -15.5% [-17.2 – -11.6] in the study population. GLS was highly correlated with GCS (R2=0.86, p<0.001). GLS was also highly correlated with conventional LV metrics, including LV mass, EDV, ESV, LV diameter, MAPSE, and LV length, but not with BSA or age. In multivariable regression analysis using GLS as the dependent variable, MAPSE, LV mass, and LV ESV together contributed to the model with the strongest association (R2=0.89, p<0.001). MAPSE modeled together with other parameters of LV size yielded models of similar strength of association (MAPSE and LVEDV, R2=0.87, p<0.001; MAPSE and LV length, R2=0.85, p<0.001; MAPSE and LV maximal diameter, R2=0.85, p<0.001). In these multivariable models, the association with LV mass did not remain statistically significant. LVEDV and LVM were correlated (R2=0.51, p<0.001). Results from univariable and multivariable linear regression analysis are presented in Table 2. GLS was highly associated with MAPSE indexed to LV length (R2=0.83 p<0.001) and with MAPSE indexed to the sum of LV length and LV diameter (R2=0.84, p<0.001). GLS could be estimated with high accuracy from MAPSE indexed to LV length. In our series iof patients, the linear relationtionship between MAPSE/LVL could be expressed using the following formula:
Representative patients illustrating the association between strain and geometric measures. All images are shown in the same magnification. End-diastolic (ED) and end-systolic (ES) long-axis (top row) and short-axis (lower row) images of three representative patients with normal (left), hypertrophied (middle), and dilated (right) geometry shown together with measures of left ventricular ejection fraction (LVEF), global longitudinal strain (GLS), global circumferential strain (GCS), mitral annular plane systolic excursion (MAPSE), LV mass (LVM) and end-diastolic volume (EDV), stroke volume (SV), left atrial strain (LAS), left atrial end-systolic volume (LAESV) and left atrial length (LAL). The white lines denote the LV and left atrial length (LAL) in ED and the white dotted lines LV and LA length in ES. The white double-headed arrows denote septal and lateral MAPSE. Across all patients and magnitudes of LV function, approximately 60% of the LV stroke volume is generated by the longitudinal movement of the atrioventricular plane, also referred to as MAPSE, that simultaneously empties and fills the LV and LA. Thus, MAPSE mechanically couples the function of the LV and LA to each other. Meanwhile, the LV apex and the posterior aspect of the LA are effectively stationary throughout the cardiac cycle. Note that across these representative patients, 1) Intra-individual GLS and GCS are very similar, 2) GLS is a function of both MAPSE and LV volume or length, and 3) LAS is function of MAPSE and LA volume or length.
Agreement between Global Longitudinal Strain (GLS) and GLS estimated as MAPSE*-100/LVL – 3.96.
Associations between GLS and left heart measures.
Global circumferential strain
GCS was -16.2% [-18.75 – -11.6] in the study population. Similarly to GLS, GCS was highly correlated with LV mass, stroke volume, LVEDV, LVESV, MAPSE, LV length, LVEF, and LV diameter, but not with BSA (Table 3). In multivariable analyses, LVESV and MAPSE together produced the best model for estimating GCS.
Associations between GCS and left heart measures.
Left atrial strain
LAS was 24 [20–29] % in the study population. Results from the uni- and multivariable analysis of LAS, LA, and LV parameters are summarized in Table 4. LAS was correlated with GLS (R2=0.51, p<0.001). LAEDV and MAPSE together associated with LAS (R2=0.65, p<0.001). MAPSE was correlated with LAS (R2=0.46, p<0.001), and this correlation increased when MAPSE was indexed to LA length (R2=0.51, p<0.001).
Associations between LAS and left heart measures.
Measured LAS and LAS calculated according to a previously published model defined as LA volume/LV volume times GLS associated with LAS (R2=0.31, p<0.001).
Longitudinal and circumferential contribution to left ventricular stroke volume
The longitudinal contribution to LV stroke volume was 62 [58 – 66]% in the population. The longitudinal contribution to LV stroke volume did not associate with GLS (p=0.68) or MAPSE indexed to LV length (p=0.12). Similarly, GCS was not associated with the short-axis contribution to LV stroke volume (p=0.92).
Discussion
The main findings of the study are that LV strain is a composite measure of both LV function and geometry that is closely associated with MAPSE and LV dimensions, and LA strain is similarly closely associated with MAPSE and LA volume. All strain measures are closely correlated with each other. Insight into and consideration of these associations between strain measures and conventional geometric measures of LV and LA size and function can inform clinical interpretation.
GLS and MAPSE were closely correlated, and the magnitude of the association increased further when MAPSE was indexed to LV length, to LV maximal diameter, or both, reflecting the dependency of GLS on both the atrioventricular plane displacement and LV size. A simple formula comprised of MAPSE and LV length could be used in our sample to estimate GLS with high accuracy. Since strain measures are subject to sizeable inter-vendor variability 12, estimating GLS using MAPSE indexed to LV length provides a potential vendor-independent and accurate estimate of GLS. In line with these findings, MAPSE indexed to LV length has previously been suggested as a surrogate measure for GLS in the case of limited image quality 13, and was recently found to provide similar prognostic accuracy to GLS 14.
LV mass was correlated with GLS, but the association remained significant only in a multivariable model using LVESV (p=0.04), and not in the other multivariable models. This may be explained by a high correlation between LVM and LV size (EDV and LVL). In our study we excluded patients with HCM.
GCS was also highly correlated with MAPSE and, similarly to GLS, indexing MAPSE to LV length or LV diameter further increased the strength of this correlation, which indicates that GLS and GCS both provide similar information. Indeed, GLS and GCS were also highly correlated with each other. While the direct correlation between GLS and GCS has not previously been reported, both GLS and GCS have been found to be highly linearly correlated with LVEF, indirectly indicating a high correlation (Onishi et al., 2015). Beyond both being measures of myocardial shortening, GCS and GLS are physically coupled through the endocardial inward movement that results from longitudinal shortening, comparable to the coupling of MAPSE to fractional shortening (Carlsson 2007). GLS is further mechanistically coupled to GCS through the radial inward movement resulting from circumferential shortening.
The median longitudinal contribution to LV stroke volume in our study was 62%, and this was comparable to findings in previous studies showing that longitudinal function is similar across a broad range of values for LVEF 16,17. MAPSE or GLS were not associated with the longitudinal contribution to stroke volume and GCS was similarly not correlated with the short-axis contribution to stroke volume. Consequently, while MAPSE and GLS are measures of longitudinal shortening, they are not necessarily appropriate to describe longitudinal function when defined as the longitudinal contribution to stroke volume.
LAS was, similarly to LV strain, highly correlated with both MAPSE and LA dimensions. Consequently, LAS also associated with GLS, in agreement with previous findings 10,18,19, as well as with GCS. A previous study found that LAS could be calculated from GLS by correcting for the LA to LV volume ratio. Similarly, our results suggest that LAS is mainly determined by LA size and LV function, although the correlation between LAS and LAS calculated from GLS and the LA/LV volume ratio was not as strong in our study as compared to previously presented models (R2=0.31 vs 0.65)10.
LAS has been found to identify patients with increased filling pressures with incremental accuracy compared to conventional echocardiographic diastolic parameters 18,20,21. LA enlargement is a known consequence of longstanding elevated filling pressures, and MAPSE has been found to be correlated with echocardiographic measures of diastolic dysfunction (Willenheimer et al., 1999). It is also known that diastolic and systolic dysfunction often co-occur. Naturally, as the myocardial shortening during systole must be equal to the myocardial lengthening during diastole, MAPSE must, mathematically speaking, be a composite measure of both. These associations may consequently explain the correlation between LAS and diastolic dysfunction. However, the majority of studies comparing LAS to invasive pressure measurements have omitted LV functional indices from performance analyses. Moreover, one study found that LAS only associated with diastolic dysfunction among patients with reduced GLS (≤18%) 23, supporting the notion that GLS is closely associated with LAS. However, one study did include GLS and found LAS provided prognostic value independently of GLS and LAVi in a population with reduced LVEF 19.
Notably, that study did not consider that the correlation between GLS and LAS is biased by the relative sizes of the LV and the LA 10.
Limitations
Female patients were underrepresented in our study, and this may be seen as a limitation. However, while females are known to have smaller hearts than males on a group level, there is no indication that female and male hearts of the same size exhibit different proportions or geometric scaling. Hence, the results should be reasonably applicable across both sexes. Also, the majority of patients included in our study had some degree of focal scarring as we included patients referred for CMR and excluded several other common clinical indications for CMR assessment, such as those with hypertrophic cardiomyopathy, valvular, and congenital heart disease. Focal scarring might bias strain measurements differently depending on the location in the myocardium. However, the exceptionally high correlation between GLS and GCS indicates that this potential effect is unlikely to have a large magnitude of effect on the findings. Moreover, the magnitude of decrease in MAPSE has been found to associate with infarct size but not infarct location24. LA and LV measurements were performed by two different observers, which may have contributed to a lower correlation between LA strain and MAPSE. All images used for strain analyses were optimized for the LV and not the LA. This could potentially affect LA length and volume calculations, and may have, in part, conferred a source of bias as there is a known correlation between an increasing LA-to-LV angle with LA and LV remodeling 25. This could potentially cause increasing measurement error with the presence of cardiac disease. However, both volume measurements and strain measurements were performed in the same images, which should have limited the impact of such potential measurement error.
Conclusions
LV and LA strains can be understood as geometrically coupled composite measures of MAPSE, and the size, function, and dimensions of the LV and LA. MAPSE and LV length alone can be used to accurately estimate GLS, which in turn provides similar information to GCS. These highly correlated composite geometrical relationships between strain and geometric measures of known prognostic significance likely explain the excellent prognostic value of strain measures.
Data Availability
All data produced in the present study are available upon reasonable request to the authors
