Background
Myocardial strain is a sensitive measure of ventricular systolic function. Two-dimensional speckle-tracking echocardiography (2DSE) is an angle-independent method for strain measurement but has not been validated in pediatric subjects. The aim of this study was to evaluate the accuracy and reproducibility of 2DSE-measured strain against reference tagged magnetic resonance imaging–measured strain in pediatric subjects with normal hearts and those with single ventricles (SVs) of left ventricular morphology after the Fontan procedure.
Methods
Peak systolic circumferential strain and longitudinal strain (LS) in segments ( n = 16) of left ventricles in age-matched and body surface area–matched 20 healthy and 12 pediatric subjects with tricuspid atresia after the Fontan procedure were measured by 2DSE and tagged magnetic resonance imaging. Average (global) and regional segmental strains measured by the two methods were compared using Spearman’s and Bland-Altman analyses.
Results
Global strains measured by 2DSE and tagged magnetic resonance imaging demonstrated close agreements, which were better for LS than circumferential strain and in normal left ventricles than in SVs (95% limits of agreement, +0.0% to +3.12%, −2.48% to +1.08%, −4.6% to +1.8%, and −3.6% to +1.8%, respectively). There was variability in agreement between regional strains, with wider limits in apical than in basal regions in normal left ventricles and heterogeneity in SVs. Strain values were significantly ( P < .05) higher in normal left ventricles than in SVs except for basal LS, which were similar in both cohorts. The regional strains in normal left ventricles demonstrated an apicobasal magnitude gradient, whereas SVs showed heterogeneity. Reproducibility was the most robust for images obtained with frame rates between 60 and 90 frames/sec, global LS in both cohorts, and basal strains in normal left ventricles.
Conclusions
Strains measured by 2DSE agree with strain measured by magnetic resonance imaging globally but vary regionally, particularly in SVs. Global strain may be a more robust tool for cardiac functional evaluation than regional strain in SV physiology. The reliability of 2DSE-measured strain is affected by the frame rate, the nature of strain, and ventricular geometry.
Strain, a dimensionless measure of myocardial deformation, provides quantitative evaluation of ventricular global and regional function. Strain rate, a time derivative of strain, has been shown to correlate linearly with left ventricular (LV) peak elastance, which is a load-independent global measure of ventricular systolic function. Both strains and strain rates are less affected by tethering, translational artifacts, and traction and are considered to be more sensitive measures of ventricular function than Doppler measurements of myocardial velocities. Tagged magnetic resonance imaging (MRI) of the heart has been used as the standard method to evaluate myocardial strain and strain rate. However, the time-intensive and labor-intensive nature of the tagged MRI method precludes its routine use in clinical pediatric practice.
A novel echocardiographic method of measurement of myocardial strain in two dimensions determines myocardial deformation from continuous frame-by-frame tracking of speckles. Speckles within B-mode images of myocardium result from ultrasonic back-scattering from structures smaller than the inherent spatial resolu-tion of the echocardiographic imaging system. The appearance of these “natural acoustic markers” is considered to be relatively stable between subse-quent image frames; hence, changes in their positions follow tissue motion. A currently available two-dimensional strain echocardiographic system (GE Vingmed Ultrasound AS, Horten, Norway) tracks speckles in a larger number of small regions to measure their motion and compute both global and regional myocardial strains. Different speckle-tracking methods have been also applied in other available systems, but their systematic validation studies in pediatric population are lacking. Systematic validation of two-dimensional speckle-tracking echocardiography (2DSE) in pediatric populations with congenital heart defects and cardiomyopathy may be particularly pertinent because these patients have heterogeneity of myocardial fiber orientation and electrical activation sequence, which determine the magnitude and distribution of strain and can affect regional myocardial backscatter properties and hence the speckle pattern. Furthermore, speckle tracking is partially dependent on frame rate. At present, there is no optimal frame rate for B-mode image acquisition for speckle tracking discerned in the pediatric age group. A systematic validation of the accuracy and reproducibility of 2DSE in pediatric subjects with normal hearts and with congenital heart defects should be a necessary prerequisite to its use in evaluating pathologic changes in cardiac function in this age group.
The aim of this clinical study was to validate the accuracy and reproducibility of 2DSE-measured longitudinal strain (LS) and circumferential strain (CS) by comparison with tagged MRI–measured strain in a cohort of children and adolescents with normal hearts and those with complex congenital heart defects after the Fontan procedure.
Methods
Study Population
A prospective study was undertaken in pediatric cohort of 22 children and adolescents with normal cardiovascular evaluation and 15 children and adolescents with single-ventricle (SV) physiology after the Fontan procedure at St. Louis Children’s Hospital. The children and adolescents representing the normal heart cohort were initially referred by primary care physicians and pediatric cardiologists to undergo echocardiographic evaluations of heart murmurs, chest pain, or syncope. They were included in the study if they had normal structure and function of the heart on transthoracic echocardiography. To ensure a homogeneous cohort of subjects with SV physiology and the similarity of their dominant ventricles to the left ventricle of a normal heart, consecutive subjects with tricuspid atresia after the Fontan procedure who were attending outpatient cardiology clinics were identified and included in the study because they had systemic ventricles of LV morphology, which have anatomic and electrical activation sequence similarity to the left ventricles of normal hearts. In both cohorts, subjects were included in the study if they had normal blood pressure, were in sinus rhythm, and had normal body mass indexes (between the fifth and 85th percentiles) for age and gender. Subjects with contraindications for MRI (subjects with ferromagnetic metallic devices or with claustrophobia) were excluded. Both the two-dimensional speckle-tracking echocardiographic and MRI studies of an individual subject were performed within 2 hours of each other, in resting state, without sedation. The study protocol was approved by the institutional review board of Washington University.
Echocardiographic Evaluation of Myocardial Strain
A standard two-dimensional and spectral or color flow Doppler examination was performed with a commercially available ultrasound imaging system (Vivid 7; GE Medical Systems, Milwaukee, WI) according to the methods recommended by the American Society of Echocardiography. LS was assessed in standard four-chamber, three-chamber (apical long-axis), and two-chamber apical views. CS was measured from standard parasternal short-axis images acquired at the apical (no visible papillary muscles), midventricular (at the papillary muscle level), and basal (at the level of the mitral valve) levels. The images were obtained at end-expiratory phase using a 4-MHz center frequency phased-array probe with second-harmonic imaging. The settings were configured to obtain optimal quality images. Frame rates for these studies were kept between 60 and 90 frames/sec. They were stored in cine loop format for offline analysis by vendor-customized software (EchoPAC PC-2D Strain; GE Medical Systems). LS and CS were measured in 16 LV segments by tracing the endocardial contour on an end-systolic frame that allowed the software to automatically place the contour on subsequent frames by temporally tracking the “natural acoustic speckle” in the B-mode images. Full thickness of myocardium from endocardial to epicardial borders was covered. Adequate tracking for the study was verified in real time, and in segments with poor tracking, the endocardial trace line was readjusted until a better tracking score was achieved. A suboptimally tracked segment was excluded from further analysis ( Figure 1 ). LS and CS curves reflected the average value of all of the acoustic markers in each segment. Studies were repeated at two different ranges of frame rates: 30 to 60 and 90 to 120 frames/sec in half of the (randomly selected) healthy subjects and in all the subjects with tricuspid atresia after the Fontan procedure to evaluate the effect of frame rate on the accuracy of 2DSE-measured systolic strain.
MRI Evaluation of Myocardial Strain, Ventricular Volumetric, and Functional Indices
Tagged MRI
Tagged MRI was performed with a 1.5-T magnetic resonance scanner (Magnetom Vision; Siemens Medical Systems, Iselin, NJ) using a phased-array cardiac coil by methods previously reported by us. Briefly, applying an electrocardiographically triggered spatial modulation of magnetization cine fast gradient-echo sequence (repetition time, 58 msec; echo time, 2.8 msec; flip angle, 30°; acquisition matrix, 256 × 256), five to eight contiguous stacks of short-axis images were prescribed from base to apex, and four long-axis slices were prescribed radially every 45°. The images were acquired during breath holds at end-expiration at 29-msec intervals until the completion of the entire cardiac cycle at each imaging plane.
Ventricular Volumes
Cine bright-blood retrogated steady-state precession images (repetition time, 3.1 msec; echo time, 1.5 msec; flip angle, 70°) in short axis were acquired with 6-mm contiguous slices from the base to the apex of the heart. Endocardial borders of all end-diastolic and end-systolic phases of the cardiac cycles were traced on a standard workstation, and Simpson’s method was applied to calculate the LV end-diastolic volume (EDV) and end-systolic volume (ESV). They were standardized to body surface area. The LV ejection fraction was calculated as the percentage ratio of (EDV − ESV)/EDV.
Strain Analysis by Finite Element Model Construction
Five to eight short-axis image sets and four radially oriented long-axis image sets were analyzed using custom-designed analysis software developed in our laboratory. An initial spline representation of the tag lines on the end-diastolic images was constructed and located on successive images using a computer-based, automated algorithm based on a method reported by McVeigh and Zerhouni. Spline curves from corresponding tag lines were used to construct a spline surface representation of the tag surfaces. Three-dimensional systolic displacements were computed along the intersection curves of individual short-axis and long-axis tag surfaces using a previously described and validated method. Analysis of the displacement data was carried out using the finite element software package StressCheck (ESRD, St. Louis, MO). Predicted displacements at any point within the domain of the model were obtained from a least squares fitting of the measured displacement data using the finite element basis functions. The regional LS and CS values were computed from the results of this fitting in 16 LV segments corresponding to segments for which strain was measured by 2DSE. For consistency in temporal analysis, the time sequence was normalized to end-systole in all modalities. End-systole was determined as the point of time corresponding to the end of the T wave on electrocardiography during the echocardiographic examinations and from the smallest cavity observed in the MRI studies.
The precision of measurements of strain components is important to quantify to assess measurement variability. We (and others) have previously estimated precision on the basis of the smoothness of strain measurements among consecutive time frames, because the computed strain at each time frame was an independent measurement. This temporal variability was quantified by taking the root mean squared deviation of the sequential strain measurement about a cubic fit of strain versus time and expressing it as a percentage of maximal value of the fitted curve. This variability measurement was averaged among all hearts. The results of our previous study showed that the precision of relative magnitude of the three-dimensional displacement reconstruction averaged 0.087 ± 0.002 mm, or 2.4 ± 0.1% of the of the maximum value.
Statistical Analysis
Data were analyzed using statistical software (SPSS version 14.0; SPSS, Inc., Chicago, IL). All values are expressed as mean ± SD. The relationship between the strains measured by the two methods was evaluated with linear regression. Agreement between strain measurements on MRI and 2DSE was evaluated by Bland-Altman analysis by calculating the bias (mean difference) and the 95% limits of agreement (two SDs around the mean difference). The statistical significance of a correlation, r , measured the possibility that r = 0. The significance of the biases and differences in myocardial strain and ventricular ejection fraction were tested by two-tailed paired t test. A P value < .05 was considered significant.
Interobserver and intraobserver variability in strain measurements by 2DSE were determined by measurements of LV strain by a second observer and the same observer, respectively, in half of the (randomly selected) normal subjects and in all 12 subjects with tricuspid atresia after the Fontan procedure and were analyzed by the Bland-Altman method.
Results
Demographic Information
Two of 22 subjects with normal hearts and three out of 15 subjects after the Fontan procedure were excluded from the study because of inadequate echocardiographic images for offline analysis. Subject demographics and hemodynamics are presented in Table 1 . Subjects with normal hearts and those who underwent Fontan procedures were anthropometrically matched except for gender. The subjects with tricuspid atresia had normally related great arteries in nine (seven with intact septa and two with small muscular ventricular septal defects) and d-transposition of great arteries with paramembranous ventricular septal defects in three. They underwent Fontan procedures (cavopulmonary shunts through the lateral tunnel in seven and extracardiac conduits in five) a mean of 11.7 years before the study. They were in New York Heart Association class I ( n = 9; median age, 13 years) or class II ( n = 3; median age, 16 years). None had cardiac arrhythmias. Hemodynamics were similar between the groups at rest.
| Subjects | Total number | Age (years) | Male/female | BSA (m 2 ) | HR (beats/min) | BP (mmHg) | Oxygen saturation (%) | Rhythm/QRS duration (ms) | EF (%) |
|---|---|---|---|---|---|---|---|---|---|
| Normal | 20 | 12.9 (9–17) | 9/11 | 1.22 (1.12–1.62) | 75 (60–104) | 92–110/55–68 | 99 (98–100) | NSR/92 ± 5 | 63 ± 2 |
| Patients with SV after Fontan procedure | 12 | 14.2 (12–17) | 7/4 | 1.27 (1.14–1.69) | 85 (72–110) | 98–121/60–78 | 94 (91–97) | NSR/102 ± 11 | 55 ± 7 |
There were no significant changes in hemodynamics (heart rate, blood pressure, and oxygen saturation) between image acquisitions by the two modalities. Eleven of 320 segments (all apical [3.4%]) in subjects with normal hearts and 17 of 192 segments (14 apical and three mid [8.9%]) in subjects who underwent Fontan procedures were excluded from analysis because of suboptimal echocardiographic images due to reverberations and image dropouts. Thus, of those included in the study, it was feasible to fully analyze 96.6% of the segments of the left ventricles in normal hearts and 91.1% of segments of the SVs in subjects who underwent Fontan procedures. A total of 484 of 512 segmental strain data (94.5%) measured by 2DSE in both cohorts entered in the analyses. After excluding 20 segments because of inadequate tagging (14 basal and six apical [3.9%]), a total of 492 of 512 segmental strain data (96%) measured by MRI in both cohorts entered in the analyses.
Validation of Myocardial Strain Measurements by 2DSE in Subjects with Normal Hearts
Strain measurements from 16 segments were averaged to assess a ventricular global parameter. The average of LS and CS values (global strain) measured by 2DSE in all cardiac segments correlated closely ( r = 0.99, P < .001 for both; Table 2 ) with corresponding strain values measured by MRI ( Figures 2 A and 2 B). Figure 3 shows the results of Bland-Altman analysis of the average values of both strains by the two methods. Bias represents the mean difference. Bias of 0% indicated complete agreement, and a difference of >1.5% strain was considered significant. Measurements of LS by 2DSE resulted in a small but significant overestimation relative to the MRI reference values, as reflected by bias of +1.6% and agreement ranging from +0.0% to +3.12%, with some measurements falling outside the 95% limits of agreement (mean ± 2 SDs). CS measured by 2DSE modestly underestimated CS values obtained by MRI, as reflected by nonsignificant bias of −0.7%. However, compared with LS, the limits of agreement were wider, ranging from −2.48% to +1.08%, and a significant number of measurements were outside the 95% limits of agreement for CS. Thus, better agreement between the two methods for global strain measurements was obtained for LS than for CS.
| LS | CS | |||||||
|---|---|---|---|---|---|---|---|---|
| Strain | MRI | 2DSE | r | P | MRI | 2DSE | r | P |
| Normal left ventricles | ||||||||
| Average | −17.1 ± 1.4 | −17.7 ± 1.5 | 0.99 | <.0001 | −21.2 ± 2.2 | −20.5 ± 1.8 | 0.99 | <.0001 |
| Basal | −15.1 ± 2.5 | −16.3 ± 3.3 | 0.85–0.99 | <.0001 | −19.1 ± 4.6 | −18.5 ± 4.8 | 0.81–0.99 | <.0001 |
| Mid | −17.5 ± 2.9 ∗ | −17.9 ± 2.6 ∗ | 0.97–0.98 | <.0001 | −21.6 ± 3.8 ∗ | −20.8 ± 4.0 ∗ | 0.97–0.98 | <.0001 |
| Apical | −18.8 ± 2.6 † | −19.4 ± 3.3 † | 0.97–0.98 | <.0001 | −23.6 ± 3.3 † | −23.2 ± 2.0 † | 0.96–0.98 | <.0001 |
| Left ventricles of patients with tricuspid atresia after Fontan procedure | ||||||||
| Average | −15.7 ± 3.2 ‡ | −14.2 ± 3.3 ‡ | 0.93 | <.001 | −18.7 ± 4.5 ‡ | −17.5 ± 4.4 ‡ | 0.94 | <.001 |
| Basal | −15.4 ± 2.8 | −14.8 ± 4.5 | 0.85–0.98 | .003 | −15.7 ± 2.2 | −15.2 ± 4.8 | 0.95–0.98 | <.0001 |
| Mid | −15.0 ± 3.7 § | −13.6 ± 4.7 § | 0.89–0.97 | .006 | −16.9 ± 4.8 § | −15.5 ± 5.2 § | 0.90–0.97 | .005 |
| Apical | −14.9 ± 2.1 ‖ | −13.8 ± 6.5 ‖ | 0.84–0.95 | .001 | −15.3 ± 5.6 ‖ | −16.2 ± 5.9 ‖ | 0.87–0.95 | .005 |
∗ P < .05, mid regional strain versus basal regional strain in the same group.
† P < .05, apical regional strain versus mid regional strain in the same group.
‡ P < .05, average (global) strain in the left ventricles of normal heart versus those in SVs.
§ P < .05, mid regional strain in the left ventricles of normal hearts versus those in SVs.
‖ P < .05, apical regional strain in the left ventricles of normal hearts versus those in SVs.
Analysis of regional strains by linear regression showed close correlation between the methods for both LS and CS ( Table 2 ) in each of six segments of the basal ( n = 6 × 20, r = 0.81–0.99) and mid (n = 6×20, r from 0.97 to 0.98) regions and four apical ( n = 4 × 20 minus 11 excluded segments, r = 0.96–0.98) regions ( P < .001 for all regions). However, there was regional variation in agreement between the values of the measurements of strain by the two methods. Bland-Altman analysis revealed that the variability of bias was larger for strain in apical segments than in basal segments (2.06 ± 2.72% vs +1.25 ± 2.96% for apical vs basal LS and −1.00 ± 2.60% vs −0.60 ± 3.13% for apical vs basal CS, respectively) and the 95% limits of agreement were wider (+1.25 ± 2.96% for LS and −0.60 ± 3.13% for CS) ( Figure 3 ). Compared with LS, a significant number of regional CS measurements fell outside the 95% limits of agreement ( Figure 3 ). Thus, a better agreement in the measured value of strain between the two methods was obtained for regional LS.
Validation of Myocardial Strain Measurements by 2DSE in Subjects with Single Left Ventricles of Tricuspid Atresia after the Fontan Procedure
In subjects with single left ventricles who underwent Fontan procedures, average values of LS and CS (global strain) measured by 2DSE in all cardiac segments correlated closely ( r = 0.93 and 0.94, respectively, P < .001 for both) with MRI-measured values ( Table 2 ). However, Bland-Altman analysis of measurements of average strain of all cardiac segments revealed significant ( P < .05) underestimation by 2DSE compared with the MRI values of LS (bias, −1.7%) and nonsignificant underestimation of CS (bias, −0.9%), with wide 95% limits of agreement for both (−4.6% to +1.8% and −3.6% to +1.8%, respectively) ( Figure 4 ).
