Multiple vendor-specific two-dimensional speckle-tracking echocardiographic algorithms with which to characterize myocardial mechanics are commercially available. The purpose of this study was to compare global longitudinal strain (GLS) results between two independent software vendors using a neutral image platform.
A convenience sample of 100 prospectively collected patients was evaluated. Subjects with more than two left ventricular endocardial segments poorly delineated were excluded. GLS was obtained from the apical four-chamber, three-chamber, and two-chamber views using two independent speckle-tracking echocardiographic software packages (EchoInsight version 1.5.0 and Image-Arena version 4.5). Linear regression analysis and paired t tests were used to compare GLS results. Intraclass correlation coefficients and Bland-Altman plots were used for assessments of reliability.
The “out-of-the-box” mean GLS was −12.99 ± 2.38% using EchoInsight and −16.87 ± 2.84% using Image-Arena (mean difference, 3.87 ± 2.42%; P = .0001). Agreement between the software packages was moderate (intraclass correlation coefficient, 0.43; 95% confidence interval, 0.32–0.55). Using uniform variables to derive GLS (Lagrangian strain measured in systole and diastole at the endocardium and averaging the peak segmental strain curves), EchoInsight GLS was −16.17 ± 2.90% and Image-Arena GLS was −16.87 ± 2.84% (mean difference, 0.70 ± 2.75%; P = .02), with an intraclass correlation coefficient of 0.70 (95% confidence interval, 0.52–0.79).
Image-Arena GLS results were consistently different (more negative) than EchoInsight measures out of the box but became similar when information used to derive GLS was uniform. The evolution of measures of myocardial mechanics into routine clinical practice will require vigilance and standardization of the various techniques, necessitating independent validation of commercially available speckle-tracking echocardiographic products.
Speckle-tracking echocardiography (STE) provides assessment of cardiac chamber function through the largely angle independent quantification of left ventricular (LV) myocardial deformation. STE has shown value in quantitatively identifying global and regional dysfunction in a multitude of cardiac conditions. However, one of the current limitations of strain imaging techniques is the lack of agreement between equipment and proprietary software preventing the adoption of universally accepted thresholds of abnormality. Given differences in image processing and the information used to derive a final strain value among manufacturers, follow-up echocardiography for strain analysis using a different manufacturer’s system may provide misleading results. In a recent expert consensus statement from the American Society of Echocardiography and the European Society of Echocardiography, a recommendation was made that more comparisons between vendor-specific software packages must be made before STE can become routinely adopted in clinical practice. Most reports evaluating differences between STE products report data analyzed from the images directly using echocardiographic equipment and software from the same manufacturer. The limitation of applying these studies to clinical practice is that many echocardiography labs use equipment from multiple vendors for image acquisition and may have speckle-tracking echocardiographic products from one or more vendor. Furthermore, many labs store images in a neutral imaging format, necessitating the use of speckle-tracking echocardiographic programs that analyze vendor-neutral images for patient follow-up. With the advent of neutral imaging formats such as Digital Imaging and Communications in Medicine (DICOM), the ability to compare speckle-tracking echocardiographic products from various vendors head to head is feasible. Few studies, however, have compared products head to head using a neutral imaging format. The purpose of this study was to perform an out-of-the-box comparison of average peak systolic global longitudinal strain (GLS) using vendor-specific software that would analyze images in a vendor-neutral platform. We further sought to determine the differences between manufacturers in the information used to derive GLS.
Echocardiographic images from 100 subjects were prospectively collected after the ethics board approved the study. Patients’ charts were retrospectively reviewed for clinical and demographic variables. Those with atrial fibrillation or flutter were identified before enrollment and excluded from the study. Each subject underwent echocardiography by physician order for various indications. Patient consent was obtained before image collection.
Transthoracic echocardiography were performed using Vivid 7 ultrasound systems with M4S probes (GE Vingmed Ultrasound AS, Horten, Norway). Standard apical two-chamber, three-chamber, and four-chamber views were obtained with special attention to LV endocardial definition. Images were obtained at frame rates of 60 to 100 frames/sec. Three cardiac cycles were obtained for optimal cycle selection in the postprocessing period, and one cardiac cycle from each image was used in the analysis.
Using DICOM-format images offline, both speckle-tracking echocardiographic software programs were used for analysis (EchoInsight version 1.5.0, Epsilon, Ann Arbor, MI [ Figure 1 ]; Image-Arena version 4.5, TomTec Imaging Systems, Unterschleissheim, Germany [ Figure 2 ]). EchoInsight software, which has only recently become commercially available, derives strain values using natural strain algorithms. The DICOM images were stored by default at 30 frames/sec, and therefore both EchoInsight and Image-Arena analyzed images at the same frame rate. Initial GLS analysis was a direct comparison between “out-of-the-box” versions; that is, we determined strain from each program without changing processing parameters or the information used to derive final strain results. Both software programs require manual tracing of the endocardial border. Visual assessment was used to exclude those images with poor visualization of a sum of more than two segments in all three views. EchoInsight divided each view into six segments, for a total of 18 segments analyzed. Out-of-the-box GLS was reported as a percentage using the peak (maximum negative value) of the average strain curve of all 18 segments. EchoInsight used natural strain algorithms out of the box, and values were converted to Lagrangian strain using a known conversion equation. Image-Arena divided the LV chamber into 18 segments, and GLS values were derived using the Lagrangian strain algorithm, using the average of the peak of the endocardial strain curves for each segment over the entire cardiac cycle. GLS was then compared between EchoInsight and Image-Arena.
Because of initial out-of-the-box differences between GLS values in the cohort, values of GLS using EchoInsight were than derived using different variables, including systole only versus systole and diastole, the peak of the average global waveform versus the average of the peak of the strain curves from each segment, natural versus Lagrangian strain, and endocardial versus a combination of endocardial and epicardial strain. We chose to manipulate the variables used to derive a final measure with the EchoInsight package because of the ease of measuring different variables with the EchoInsight software. The final measurements with each software package were derived using the average of each endocardial segmental strain curve peak using Lagrangian analysis during the entire cardiac cycle. Interobserver and intraobserver variability was assessed on 10 randomly chosen image sets with each vendor. Readers were blinded to their own or the second readers’ prior measures and allowed to choose the cardiac cycle analyzed within each image. In the case of intraobserver variability, the repeated measures occurred on different days.
Paired t tests were used to compare GLS results between strain analysis software. Pearson’s regression analysis was used to determine correlation. Intraclass correlation coefficients (ICCs) and Bland-Altman plots were used for agreement. Statistical significance was accepted at a P value of <.05.
After exclusion for suboptimal images (defined as more than two LV endocardial borders poorly delineated), 88 of 100 subjects (mean age, 60 ± 17 years; 52% men) were used in the final analysis ( Table 1 ). In summary, the average age was 60 ± 17 years, and LV ejection fractions were normal in the population (mean, 63 ± 6%). Atrial size was mildly enlarged (mean left atrial volume index, 33 ± 10 cm 3 /m 2 ), and mean right ventricular systolic pressure was 33 ± 12 mm Hg. Indications for echocardiography are shown in Table 2 . The majority of echocardiographic studies were performed for arrhythmias, known or suspected coronary artery disease, and chest pain. A significant number of patients had conditions known to lower GLS values, including coronary artery disease, amyloid, heart failure, and hypertrophic cardiomyopathy. Twelve of 100 subjects were excluded because of a lack of visualization of more than two cardiac segments. The interobserver ICCs were 0.903 (95% confidence interval [CI], 0.087 to 0.954) for EchoInsight and 0.791 (95% CI, 0.165 to 0.958) for Image-Arena. The intraobserver ICCs were 0.976 (95% CI, 0.965 to 0.996) for EchoInsight and 0.623 (95% CI, −0.488 to 0.924) for Image-Arena.
|Age (y)||60 ± 17|
|LV ejection fraction (%) (range)||63 ± 6 (32–78)|
|Indexed left atrium volume (cm 3 /m 2 )||33 ± 10|
|Systolic blood pressure (mm Hg)||132 ± 21|
|Diastolic blood pressure (mm Hg)||76 ± 11|
|Heart rate (beats/min)||67 ± 11|
|Deceleration time (msec)||206 ± 44|
|E/e′ ratio||11 ± 7|
|Isovolumic relaxation time (msec)||85 ± 22|
|Right ventricular systolic pressure (mm Hg)||33 ± 12|
|Coronary artery disease||15 (17)|
|Chest pain||10 (11)|
|Kidney transplantation||8 (9)|
|Shortness of breath||8 (9)|
|Valvular heart disease||7 (8)|
Initial out-of-the-box results showed a significant difference between vendor GLS results. Image-Arena GLS was −16.87 ± 2.84% and EchoInsight GLS was −12.99 ± 2.38% in the overall cohort (mean difference, −3.87 ± 2.41%; P = .0001). Agreement between the two measures showed an ICC of 0.73 (95% CI, 0.59–0.82; R 2 = 0.33; P < .001; Figure 3 ). The Bland-Altman plot is shown in Figure 3 .
Information used to derive GLS results differed in three of four major ways between EchoInsight and Image-Arena. Image-Arena used Lagrangian analysis to derive strain curves, whereas EchoInsight used natural strain. Further differences between vendors included the evaluation of endocardial strain with Image-Arena, while EchoInsight used a combination of endocardial and epicardial strain. Both vendors used the entire cardiac cycle in which to use the most negative portion of the strain curve as opposed to reporting GLS in systole only. Image-Arena used the average value of the peaks (most negative) of the strain curves from each wall segment, whereas EchoInsight used the peak of the global average curve.
Peak GLS values changed significantly with EchoInsight as information used to derive the final GLS result was changed ( Table 3 ). Out-of-the-box EchoInsight GLS was −12.99 ± 2.38% ( Table 3 ). When the location of strain was changed from endocardial/epicardial to endocardial only, GLS became −13.99 ± 2.61% (mean difference, −0.99 ± 0.60%; P = .0001; R 2 = 0.95; ICC, 0.95 [95% CI, 0.93–0.96]). Using the new GLS value with endocardial strain of −13.99 ± 2.61% ( Table 3 ) and using Lagrangian strain in place of natural strain, GLS became −15.05 ± 2.99% (mean difference, −1.06 ± 0.38%; P = .0001; R 2 = 0.99; ICC, 0.96 [95% CI, 0.94–0.97]). Using this new GLS value of −15.05 ± 2.99% ( Table 3 ) and using the average of the peak segmental strain curves, GLS became −16.17 ± 2.90% (mean difference, −1.11%; P = .0001; R 2 = 0.80; ICC, 0.91 [95% CI, 0.88–0.93]). This final EchoInsight GLS value of −16.17 ± 2.90% ( Table 3 ) used the same variables to derive the final GLS result as Image-Arena, which was −16.87 ± 2.84% (mean difference, −0.70 ± 2.75%; P = .02; R 2 = 0.29; ICC, 0.70 [95% CI, 0.52–0.79]; Figure 4 ). The Bland-Altman plot is shown in Figure 4 .