Background
Left ventricular (LV) ejection fraction (EF) is a routine clinical standard to assess cardiac function. Global longitudinal strain (GLS) and global circumferential strain (GCS) have emerged as important LV functional measures. The objective of this study was to determine the relationships of GLS and GCS by speckle-tracking echocardiography and featuring-tracking cardiac magnetic resonance (CMR) to CMR EF as a standard of reference in the same patients.
Methods
A total of 73 consecutive patients aged 55 ± 15 years clinically referred for both CMR and echocardiography (EF range, 8%–78%) were studied. Routine steady-state free precession CMR images were prospectively analyzed offline using feature-tracking software for LV GLS, GCS, volumes, and EF. GLS was averaged from three standard longitudinal views and GCS from the mid-LV short-axis plane. Echocardiographic speckle-tracking was used from the similar imaging planes for GLS, GCS, LV volumes, and EF.
Results
Feature-tracking CMR strain was closely correlated with speckle-tracking strain in the same patients: GLS, r = −0.87; GCS, r = −0.92 ( P < .0001). End-diastolic and end-systolic volumes and EF by feature-tracking CMR were significantly correlated with standard manual tracing of multiple CMR short-axis images ( r = 0.97, r = 0.98, and r = 0.97, P < .0001 for all). GLS and GCS by echocardiography and CMR feature-tracking were closely correlated with standard CMR EF: r = −0.85 and r = −0.95, respectively ( P < .001). Global strain measures (in absolute values) were correlated with EF using the formula EF = 3(GLS) + 8% or EF = 2.5(GCS) + 8%.
Conclusions
GLS and GCS by feature-tracking CMR analysis was a rapid means to obtain myocardial strain similar to speckle-tracking echocardiography. GLS and GCS were closely correlated with CMR EF in this patient series and may play a role in the clinical assessment of LV function.
Highlights
- •
FT CMR is a rapid and reliable means to assess LV volumes and EF and global strain values.
- •
GCS and GLS can be similarly calculated by FT CMR and speckle-tracking echocardiography.
- •
LV GCS and GLS measured by any method correlate closely with EF.
- •
We propose two simplified formulas to estimate EF from global strain values that have potential clinical utility.
Myocardial strain imaging has emerged as a powerful quantitative tool to assess regional and global left ventricular (LV) function and predict patient outcomes. Most clinical strain imaging methods have used either speckle-tracking echocardiography or cardiac magnetic resonance (CMR) imaging with myocardial tissue tagging. Although the image resolution of CMR is regarded as superior to that of echocardiography overall, CMR tagging for myocardial strain requires additional processing beyond routine steady-state free precession imaging, which may be time consuming and inefficient. The standard clinical approach to assess LV function has remained the determination of volumes and ejection fraction (EF). Echocardiographic EF can often be technically challenging, and routine CMR volumes and EF are time consuming, requiring manual tracing of multiple images. A feature-tracking (FT) CMR image analysis approach has recently been developed that can determine myocardial strain without tissue tagging and semiautomated volumetric measures from routine Digital Imaging and Communications in Medicine (DICOM) images. The objectives of this study were (1) to test the hypothesis that GLS and GCS by speckle-tracking echocardiography and FT CMR could rapidly assess LV function comparable with LV volume and EF data from manual tracing of serial short-axis CMR images, which is largely considered the gold standard, and (2) to determine the mathematical relationships of global longitudinal strain (GLS) and global circumferential strain (GCS) to CMR EF in a patient series as a potential means to assess LV function.
Methods
We studied 73 consecutive subjects (26 women; mean age, 55 ± 15 years; age range, 19–88 years) who underwent both echocardiography and CMR to evaluate LV function for clinically suspected heart failure symptoms, typically scheduled for the same day or within 3 to 5 days ( Table 1 ). These patients underwent echocardiography and CMR as requested by the clinician involved in their care, not as part of an independent research study. The FT CMR and speckle-tracking echocardiographic analyses were prospectively applied to routine imaging data acquired by routine acquisition protocols for clinical purposes. All patients were in sinus rhythm at the time of CMR and echocardiography. The protocol was approved by the Institutional Review Board on Biomedical Research.
| Variable | Value |
|---|---|
| All study patients | n = 73 (100%) |
| Men | 47 (64%) |
| Age (y), mean ± SD | 55 ± 15 |
| Subgroup of patients with systolic dysfunction, EF ≤ 35% | n = 30 (41%) |
| Ischemic cardiomyopathy | 12 (16%) |
| Nonischemic cardiomyopathy | 18 (25%) |
| Subgroup of patients without systolic dysfunction, EF > 35% | n = 43 (59%) |
| Ischemic heart disease | 8 (11%) |
| Hypertrophic cardiomyopathy | 4 (5%) |
| Hypertensive heart disease | 3 (4%) |
| Paroxysmal atrial fibrillation | 9 (12%) |
| No significant cardiac disease identified | 19 (26%) |
CMR Acquisition
CMR imaging was performed with a 1.5-T Magnetom Espree (Siemens Medical Solutions, Erlangen, Germany) with a 32-channel phased-array cardiovascular coil. Steady-state free precession imaging was performed during 5- to 10-sec breath-holds using parallel acquisition acceleration factors (generalized autocalibrating partially parallel acquisition) of three and stored digitally for offline analysis. Three long-axis and an entire stack of short-axis cine loops were acquired using steady-state free precession imaging with the following typical parameters: echocardiography time, 1.22 msec; repetition time, 47.4 msec; flip angle, 60°; slice thickness, 6 mm (4-mm gap in short-axis stack); spatial resolution, 1.8 × 1.5 mm 2 ; and temporal resolution, 30 frames/RR interval. Assessment of LV volumes and EF was performed by manual tracing of the endocardial borders at end-diastole and end-systole in each of the short-axis slices using conventional CMR software (Argus; Siemens Medical Solutions).
CMR FT Analysis
A novel semiautomated FT CMR software package (2D Cardiac Performance Analysis MR; TomTec Imaging GmbH, Munich, Germany) was used as a vector-based analysis tool that mimics echocardiographic speckle-tracking and is based on a hierarchical algorithm that operates at multiple levels using two-dimensional (2D) FT techniques. LV strain, volumes, and EF were analyzed from routine DICOM data sets by investigators blinded to the clinical, echocardiographic, and other CMR data. The following specific tomographic imaging planes were selected from the DICOM CMR imaging data set to correspond to slices similar to echocardiographic views. Longitudinal imaging planes included four chamber, two chamber, and long axis. These longitudinal views corresponded to the standard apical four-chamber, apical two-chamber, and apical long-axis views used for echocardiography. The short-axis plane was selected at the papillary muscle level using the most circular orientation. The mid-LV short-axis view corresponded to the echocardiographic parasternal short-axis view at the mid-LV level. For FT CMR analysis, regions of interest were traced on the endocardium and epicardium for these three longitudinal views and the short axis view at the midpapillary level. Adjustment of the region of interest was done after visual assessment during cine-loop playback to ensure that the LV segments were tracked appropriately. Longitudinal strain was calculated from the four-chamber, two-chamber, and long-axis view imaging planes, and circumferential strain from the mid-LV short-axis plane, similar to the tomographic approach for speckle-tracking image analysis. Time-strain curves from six segments for each view were plotted along with the average time strain curves for GLS and GCS for the respective views ( Figures 1 and 2 ). Three longitudinal LV cine loop planes were selected, similar to routine clinical echocardiographic imaging: four-chamber view, two-chamber view, and long-axis (three-chamber) view. LV end-diastolic volumes (EDVs) and end-systolic volumes (ESVs) were derived using the semiautomated CMR FT method by placing a region of interest on the LV endocardial and epicardial borders, respectively. The FT software combined volume information from these three tomographic planes to rapidly calculate overall LV EDV, ESV, and EF. The mean of EDV ([EDV four-chamber view + EDV two-chamber view + EDV long-axis view]/3) and ESV ([ESV four-chamber view + ESV two-chamber view + ESV long-axis view]/3) were calculated, and the EF was estimated from the mean volumes: [(EDV mean − ESV mean)/EDV mean] × 100%.
Echocardiographic Acquisition
All echocardiographic studies were performed on commercially available echocardiography systems: Vivid 7 (GE Vingmed Ultrasound AS, Horten, Norway) or iE33 (Philips Medical Systems, Andover, MA). Digital routine grayscale 2D cine loops from three consecutive beats were obtained at end-expiratory apnea from the standard apical four- and two-chamber, long-axis, and mid-LV short-axis views at frame rates of 30 to 100 Hz (mean, 65 ± 15 Hz), as previously described. Gain settings were adjusted for routine clinical grayscale 2D imaging to optimize endocardial definition. The DICOM-formatted files containing LV images were then exported to a personal computer for subsequent offline postprocessing.
Echocardiographic Speckle-Tracking Analysis
Routine B-mode grayscale LV images were analyzed to quantify myocardial strain and volumes (2D Cardiac Performance Analysis) on the basis of stable patterns of natural acoustic markers or speckles within the myocardium, as described previously. LV segmental longitudinal strain, GLS, segmental circumferential strain, and GCS were calculated from strain curves during the entire cardiac cycle with user-defined regions of interest, similar to FT CMR analysis. Peak GLS and GCS were calculated for the entire U-shaped (GLS) and circle-shaped (GCS) LV myocardium as follows: global strain (%) = [ L (end-systole) − L (end-diastole)]/ L (end-diastole) × 100. Using the same software approach as FT CMR analysis, regions of interest were traced on the endocardium from echocardiographic apical four-chamber, two-chamber, and long axis (three-chamber) views. The software similarly combined volume information from these three tomographic planes to calculate overall echocardiographic LV EDV, ESV, and EF. As an additional goal, we used CMR EF as the reference standard to determine the mathematical relationships of GLS and GCS to EF, respectively, in our patient series.
Statistical Analysis
Data are presented as mean ± SD unless otherwise stated. Group values were compared using two-tailed Student t tests for paired data. Correlation analysis was performed using linear regression by Pearson correlation coefficient, while agreement between methods was assessed using Bland-Altman plots. The simplified formulas were derived from the regression equations and then tested for agreement with Bland-Altman plots and coefficients of variation. Receiver operating characteristic curves were constructed for each strain parameter individually to determine sensitivities and specificities for binary outcomes. Inter- and intraobserver variability analysis for all FT CMR measurements was performed in 10 randomly selected patients using the identical cine loop for each view. Inter- and intraobserver variability is expressed as the absolute differences divided by the mean value of the measurements. Microsoft Excel (Microsoft Corporation, Redmond, WA) and MedCalc version 12.0.4.0 (MedCalc Software, Inc, Mariakerke, Belgium) were used. Statistical significance was set at P < .05.
Results
Patient Characteristics
The study group consisted of 73 patients, of whom 30 (41%) had LV severe systolic dysfunction, defined as a CMR EF ≤ 35%. ( Table 1 ). This included 12 (16%) with ischemic cardiomyopathy, 18 (25%) with nonischemic cardiomyopathy, and 43 (59%) without severe systolic dysfunction (CMR EF > 35%). Only 10 patients (14%) had LV regional wall motion abnormalities among the 20 (27%) with coronary artery disease.
LV Volume and EF Analysis
Imaging data were suitable for quantitative conventional CMR analysis in 100% (73 of 73). The average conventional CMR EDV, ESV, and EF, respectively, were 208 ± 87 mL, 125 ± 86 mL, and 45 ± 18%. Imaging data were similarly suitable for quantitative FT CMR analysis in 100% of the subjects (73 of 73). The average FT CMR EDV, ESV, and EF, respectively, were 192 ± 86 mL, 112 ± 80 mL, and 47 ± 18%. For speckle-tracking echocardiography, imaging data were suitable for quantitative analysis in 93% of the study subjects (68 of 73. The average speckle-tracking echocardiographic EDV, ESV, and EF, respectively, were 145 ± 74 mL, 90 ± 66 mL, and 44 ± 17%.
Significantly high correlations with excellent Pearson correlation coefficients ( r ) were noted for all LV EDV, ESV, and EF measurements between conventional CMR and FT CMR ( r = 0.97–0.98, P < .0001 for all; Table 2 , Figure 3 ), between conventional CMR and speckle-tracking echocardiography ( r = 0.91–0.93, P < .0001 for all; Table 2 , Figure 4 ), and between speckle-tracking echocardiography and FT CMR ( r = 0.91–0.93, P < .0001 for all; Table 2 ). Bland-Altman plots showed a small underestimation of volumes by FT-CMR compared with conventional CMR measures and a small bias in EF of −2.5% in EF units. Bias consistent with a systematic underestimation of LV volumes by echocardiography compared with conventional CMR volumes was observed, although comparisons in EF showed no significant bias.
| Varable | r | Linear regression equation | P |
|---|---|---|---|
| Conventional CMR vs FT CMR | |||
| EDV | 0.91 | y = 0.95 x − 6.5 | <.0001 |
| ESV | 0.93 | y = 0.91 x − 1.5 | <.0001 |
| EF | 0.92 | y = 0.95 x + 4.8 | <.0001 |
| Conventional CMR vs ST echocardiography | |||
| EDV | 0.91 | y = 0.76 x − 14 | <.0001 |
| ESV | 0.93 | y = 0.72 x − 0.5 | <.0001 |
| EF | 0.93 | y = 0.84 x + 6.8 | <.0001 |
| ST echocardiography vs FT CMR | |||
| EDV | 0.91 | y = 1.06 x + 40 | <.0001 |
| ESV | 0.93 | y = 1.13 x + 12 | <.0001 |
| EF | 0.92 | y = 1.00 x + 2.6 | <.0001 |
LV Strain Analysis
CMR imaging data by FT analysis was feasible in 100% of patients (73 of 73) for both quantitative circumferential and longitudinal analyses. The average FT CMR GLS and GCS were, respectively, −14.1 ± 6.8% and −16.4 ± 8.4%. Speckle-tracking echocardiographic data analysis was feasible in 97% of patients (71 of 73) for circumferential strain analysis and in 93% (68 of 73) for longitudinal strain analysis. The average speckle-tracking echocardiographic GLS and GCS were, respectively, −12.0 ± 5.6% and −14.9 ± 7.2%. FT CMR GLS and speckle-tracking echocardiographic GLS correlated well with conventional CMR EF ( r = −0.88 and −0.85, P < .0001 for both; Figure 5 ), with an overall bias of 1.9%. Also, FT CMR GCS significantly correlated with conventional CMR EF ( r = −0.95, P < .0001; Figure 6 ) as well as speckle-tracking echocardiography GCS correlated with conventional CMR EF ( r = −0.92, P < .0001), with an overall bias of 1.5%. Similarly, Bland-Altman analysis showed excellent agreement between different software packages for strain data. Receiver operating characteristic curve analyses of GLS and GCS for predicting a conventional CMR EF ≤ 35% showed that cutoff values of −12% and −11% in FT CMR GLS and speckle-tracking echocardiography GLS yielded 91% and 85% sensitivity, with 93% and 89% specificity, respectively, and areas under the curve (AUCs) of 0.965 and 0.955. Cutoff values of −12% and −13% for FT CMR GCS and speckle-tracking echocardiographic GCS, respectively, yielded 95% and 93% sensitivity with 96% and 93% specificity, respectively, and AUCs of 0.966 and 0.963.
Simplified Formulas
After assessing the high correlation between EF and global strain values, different simplified formulas were determined from the regression equation ( Table 3 ), which were subsequently tested for agreement with coefficient of variation and Bland-Altman plots. Because GLS and GCS have negative signs, absolute values were used to convert to positive signs. The formulas with the closest agreement with homogeneous coefficients of variation were as follows: for GLS, EF = 3(GLS) + 8%, with a coefficient of variation of 15.2%, and for GCS, EF = 2.5(GCS) + 8%, with a coefficient of variation of 11.1%. The selected formulas displayed in Bland-Altman plots showed consistent agreement within ±1.96 SDs compared with the conventional CMR EF, with a mean bias of −0.3 EF units (formula by GLS) and 0.5 EF units (formula by GCS). In Table 3 , we present the echocardiographic global strain values for consecutive LV EF values, according to the regression equation followed by the EF results obtained when using the different selected formulas.
