Intervendor Variability of Two-Dimensional Strain Using Vendor-Specific and Vendor-Independent Software


Although two-dimensional (2D) strain is widely used to assess left ventricular mechanics, the strain values derived from vendor-specific 2D speckle-tracking software are different even for the same subjects and are therefore not interchangeable. The aim of this study was to test the hypothesis that vendor-independent software would produce lower intervendor variability between 2D strain measurements and overcome this limitation.


Two sets of three apical images were acquired using two of three types of ultrasound machines (GE, Philips, and Toshiba) in 81 healthy volunteers (GE vs Philips in 26 subjects, Philips vs Toshiba in 31 subjects, and GE vs Toshiba in 24 subjects). Two-dimensional global longitudinal strain (GLS) was measured using vendor-specific software and two vendor-independent software packages (TomTec and Epsilon) in each set of apical images, and GLS values were directly compared with one another.


The upgrades of vendor-specific software yielded different values of GLS compared with the previous versions of the software. The correlations between the GLS values determined using vendor-specific software exhibited a wide range of r values ( r = 0.23, r = 0.42, and r = 0.72), with significant bias, with the exception of one comparison. The vendor-independent software provided modest degrees of correlation (TomTec: r = 0.65, r = 0.65, and r = 0.77; Epsilon: r = 0.65, r = 0.74, and r = 0.77), with limits of agreement (range, ±3% to ±4.5%) that were not negligible.


Although the vendor-independent 2D strain software provided moderate correlations between the GLS values of the ultrasound images obtained from the same subjects using different vendors, relatively large limits of agreement remain a relevant problem. These results suggest that the same ultrasound machine and the same 2D speckle-tracking software should be used for longitudinal analysis of strain values in the same subjects and for cross-sectional studies.

The assessment of regional and global left ventricular (LV) strain and strain rates may be advantageous compared with traditional LV functional parameters, including LV ejection fraction, for the quantification of LV mechanics and the prediction of future outcomes. With the advent of two-dimensional (2D) speckle-tracking analysis software, this method can be used to measure strain in all four cardiac chambers, during both systole and diastole. Myocardial shortening and lengthening can also be measured in a variety of directions, including the longitudinal, circumferential, and radial directions. Two-dimensional strain measurements are widely performed for several cardiovascular disorders. In addition, emerging applications, such as the identification of subclinical dysfunction during chemotherapy, have emphasized the need for a standardized approach to strain values. Published studies aiming to clarify the differences in 2D strain values between specific disease states and normal conditions, however, often use different ultrasound vendors with different analytical software and cutoff values. Recent studies have demonstrated that intervendor peak strain values measured by vendor-specific software in the same subjects are frequently different and incomparable.

In clinical situations, the majority of ultrasound laboratories use several ultrasound machines from different vendors for routine echocardiographic examinations. Therefore, any longitudinal studies evaluating changes in strain values using 2D speckle-tracking software should consider the reliability of results obtained using different vendor-specific software algorithms. To overcome these limitations, several vendor-independent 2D strain software packages have been developed. We hypothesized that vendor-independent 2D strain software would overcome the intervendor variability inherent in strain measurements. Because global longitudinal strain (GLS) is the most widely used parameter and is a robust index for clinical studies, we aimed to assess the intravendor and intervendor correlations among GLS values obtained in three clinically relevant situations: (1) the effect of the upgrade of vendor-specific software on GLS measurements in the same data sets, (2) a comparison between GLS values obtained using vendor-specific software for the same subjects, and (3) a comparison between GLS values obtained using vendor-independent software with images acquired from the same subjects using different ultrasound equipment.



Among 193 subjects who were enrolled to determine intervendor variability in the Japanese Ultrasound Speckle Tracking of the Left Ventricle study, we selected 81 subjects aged ≥20 years (GE vs Philips in 26 subjects, Philips vs Toshiba in 31 subjects, and GE vs Toshiba in 24 subjects). The subjects were enrolled from four hospitals in Japan and were primarily hospital employees, their relatives, and other volunteers recruited via advertising. The eligibility criteria were as follows: (1) no history of hypertension, with normal blood pressure at the time of examination; (2) no history of diabetes mellitus, hyperlipidemia, or cardiovascular disease; and (3) no history of cardiac medication use. All subjects underwent physical examinations and 2D echocardiography to exclude those with wall motion abnormalities or valvular disorders. The ethics committee of each of the hospitals approved the study protocol, and informed consent was obtained from all study subjects.

Two-Dimensional Echocardiography

Two-dimensional grayscale harmonic images were obtained using ultrasound systems from two of three vendors (Vivid 7 or Vivid E9 [GE Vingmed Ultrasound AS, Horten, Norway], iE33 [Philips Medical Systems, Andover, MA], and Artida or Aplio [Toshiba Medical Systems, Otawara, Japan]). LV apical four-, two-, and three-chamber views were acquired in the left lateral decubitus position during two or three consecutive cardiac cycles during a breath hold and digitally stored on the hard disk. The gain and compression were adjusted to minimize the dropout of the LV endocardial and epicardial borders. The depth and the sector angle were adjusted to include the left ventricle, but the sector size was minimized to achieve a higher frame rate.

Image Analysis

Two-dimensional speckle-tracking analysis was performed using several speckle-tracking software packages by an experienced observer blinded to the age and gender of each subject. To determine intravendor variability, we used the two most recent versions of their vendor-specific software selected for this study (EchoPAC PC versions 112 and 113 [GE Vingmed Ultrasound AS], QLAB versions 9.0 and 10.0 [Philips Medical Systems], and ACP versions 3.0 and 3.2 [Toshiba]). For intervendor variability, we used the latest version of the vendor-specific software and two vendor-independent software programs (2D CPA [TomTec Imaging Systems, Unterschleissheim, Germany] and Cardio Oncology [Epsilon Imaging, Ann Arbor, MI]) ( Figure 1 ).

Figure 1

Representative longitudinal strain analysis in each vendor. Left columns show apical four-chamber view, and right columns show longitudinal strain curves in each of six segments. Dotted line indicates average strain curve generated from six-segment strain curve. Global longitudinal strain was defined as average peak strain from 18 segments (apical four-chamber, two-chamber, and three-chamber views). (A) GE image and vendor-specific strain analysis; (B) Philips image and vendor-specific strain analysis; (C) Toshiba image and vendor-specific strain analysis; (D) Philips image and vendor-independent strain analysis (TomTec); (E) Philips image and vendor-independent strain analysis (Epsilon).

Detailed speckle-tracking analysis for each vendor has been described elsewhere. Briefly, both the endocardial and epicardial borders were manually traced in each of the apical views. The region of interest was then adjusted to include the entire myocardium. The software algorithm automatically divided the left ventricle into six equidistant segments and tracked the speckle patterns on a frame-by-frame basis using the vendor-specific proprietary speckle-tracking algorithm. Finally, the software automatically generated time-domain LV strain curves for each of the six segments from which the peak strain was measured. If the tracking was judged suboptimal by visual inspection after several adjustments of the endocardial and epicardial borders, the segment was excluded from the analysis. GLS was calculated by averaging the peak strains of the 18 LV segments derived from three apical views (GE, Philips, Toshiba, and TomTec) or by averaging the end-systolic strains obtained from 18 LV segmental longitudinal strain curves (Epsilon). The GLS values were calculated from the entire myocardial wall using the software manufactured by Philips, GE, and Epsilon. The GLS values were measured from the endocardium using the software manufactured by Toshiba and TomTec.

The generally negative value of longitudinal strain can cause confusion when comparing values because deterioration in LV function results in a counterintuitive increase in the arithmetic value of strain. Therefore, we report the absolute values of GLS in the text, figures, and tables to clearly demonstrate that subjects with superior LV function have higher GLS values.

Because all vendor-specific software can store information regarding the regions of interest for speckle-tracking analyses, we were able to retrieve the region of interest at exactly the same area for comparing GLS measurements obtained using different versions of analysis software, eliminating intraobserver variability.

Observer Variability

For observer variability, images from 16 subjects acquired using ultrasound equipment were subjected to 2D speckle-tracking analysis using the latest version of the vendor-specific software. For vendor-independent software, we selected 16 subjects (six subjects whose images were acquired with GE equipment, five subjects whose images were acquired with Philips equipment, and five subjects whose images were acquired with Toshiba equipment). To assess intraobserver variability, one observer (Y.N.) measured GLS values twice at an interval of 2 weeks. Two observers (Y.N. and K.M.) measured GLS values in the same data sets to assess interobserver variability. Intra- and interobserver variability values were each determined as variability percentages, defined as absolute differences in the percentages of mean of the two measurements, the limits of agreements, and intraclass correlation coefficients.

Statistical Analysis

Continuous data are expressed as mean ± SD or as medians and interquartile ranges according to the data distribution. All statistical analyses were performed using commercially available software (JMP version 11.0; SAS Institute Inc, Cary, NC). Two groups were compared using paired or unpaired t tests. Because the linear regression used to assess the various methods of determining GLS has some limitations, we also assessed the biases and limits of agreement by Bland-Altman analysis. P values < .05 were considered to indicate statistical significance.


Study Subjects

The anthropometric characteristics of this study subjects are summarized in Table 1 .

Table 1

Baseline clinical characteristics in the study subjects

Clinical characteristic Value Range
Age (y) 32 ± 7 23–52
Male 78 (93%)
Height (cm) 171 ± 6 155–185
Weight (kg) 67 ± 9 45–105
Body mass index (kg/m 2 ) 22. 6 ± 2.9 17.1–33.9
Body surface area (/m 2 ) 1.78 ± 0.14 1.10–2.27
Heart rate (beats/min) 63 ± 10 43–88
Systolic BP (mm Hg) 121 ± 10 100–138
Diastolic BP (mm Hg) 72 ± 8 57–88

BP , Blood pressure.

Data are expressed as mean ± SD or as number (percentage).

Effect of Upgrade of Vendor-Specific Software on GLS

Even when using the same ultrasound vendors, the upgrades to the vendor-specific speckle-tracking software significantly affected the GLS values in the same images ( Table 2 ). Compared with the previous versions of the software, the software upgrades produced significant reductions in the GLS values obtained using GE and Philips equipment and slight but significant increases in the GLS values obtained using Toshiba equipment. The correlations between the GLS values obtained with the two types of software ranged from excellent to fair, depending on the ultrasound vendor ( Figure 2 ). Because upgrading the software was expected to improve tracking, which would in turn affect the calculation of GLS, we recalculated GLS only from the segments for which both types of software demonstrated acceptable tracking. The recalculation improved the correlation value for the Philips data but had no effect on the correlations for the GE and Toshiba data ( Table 2 ). To evaluate how the software upgrade would affect clinical diagnosis (normal or abnormal), we calculated how many of the normal patients would have been reclassified as abnormal by the newer software package ( Figure 3 ). The reclassification rate was 7% for GE, 19% for Philips, and 0% for Toshiba.

Table 2

Effect of upgrade on GLS for each vendor

GLS (previous version) No. of analyzable segments GLS (current version) No. of analyzable segments Bias P 95% LOA r
GE 19.41 ± 1.62 17.4 ± 0.9 18.36 ± 1.45 17.3 ± 1.1 −1.06 <.0001 −2.37 to 0.25 0.91
19.53 ± 1.58 17.2 ± 1.1 18.43 ± 1.38 17.2 ± 1.1 −1.10 <.0001 −2.36 to 0.16 0.91
Philips 19.40 ± 1.86 16.7 ± 1.8 17.09 ± 1.96 17.8 ± 0.9 −2.31 <.0001 −5.29 to 0.68 0.68
19.40 ± 1.86 16.7 ± 1.9 17.35 ± 1.95 16.7 ± 1.9 −2.04 <.0001 −4.78 to 0.70 0.73
Toshiba 15.75 ± 1.46 15.9 ± 1.4 16.39 ± 1.52 16.5 ± 1.5 0.61 <.0001 −1.11 to 2.33 0.83
15.92 ± 1.37 16.0 ± 2.6 16.48 ± 1.55 16.0 ± 2.6 0.56 <.0001 −1.19 to 2.31 0.82

LOA , Limits of agreement.

For each vendor, the first row includes direct comparison of GLS derived from analyzable segments, and the second row includes comparisons of GLS assessed by segments that could be analyzed by both software versions.

Upgrade from version 112 to version 113.

Upgrade from version 9 to version 10.

Upgrade from version 3.0 to version 3.2.

Figure 2

Linear correlation ( top ) and Bland-Altman plots ( bottom ) for global longitudinal strain between previous version of the software and current version of the software for each ultrasound vendor. Global longitudinal strain was expressed as absolute values. Dotted line in the linear correlation analysis represents line of identity. Dotted lines in the Bland-Altman analysis represent bias and 95% limits of agreement. (A) GE, (B) Philips, (C) Toshiba.

Figure 3

Reclassification rate between the previous version and the updated version of software to determine normal or abnormal. Cutoff value was determined as mean − 2 SDs for each previous version’s vendor-specific software. This cutoff value was applied when GLS was measured by the upgraded software, and reclassification rate was calculated.

Comparison of Vendor-Specific Strain Values

As expected, the GLS values analyzed using vendor-specific software were significantly different for each two pairs of images acquired using two ultrasound machines from different vendors ( Table 3 ). The correlations of the values of GLS obtained using the two vendor-specific software packages ranged from fair ( r = 0.72, GE vs Philips) to poor ( r = 0.23, GE vs Toshiba), with a statistically significant bias and a wide range of limit of agreement ( Figure 4 ).

Table 3

GLS between vendor-specific software

Comparison P 95% LOA r
GE vs Philips
GLS (GE) GLS (Philips) Bias
18.69 ± 1.72 17.88 ± 1.84 −0.80 .0060 −3.41 to 1.81 0.72
GE vs Toshiba
GLS (GE) GLS (Toshiba) Bias
18.62 ± 1.30 16.77 ± 1.11 −1.89 <.0001 −4.85 to 1.01 0.23
Philips vs Toshiba
GLS (Philips) GLS (Toshiba) Bias
16.48 ± 1.84 16.09 ± 1.73 −0.39 .2823 −4.17 to 2.29 0.42

LOA , Limits of agreement.

Figure 4

Linear correlation and Bland-Altman plots for global longitudinal strain between vendor-specific software. (A) GE versus Philips, (B) GE versus Toshiba, (C) Philips versus Toshiba.

Comparison of Vendor-Independent Strain Values

The GLS values analyzed using vendor-independent software for the same subjects using paired images obtained from different ultrasound vendors exhibited a modest degree of correlation ( Table 4 ). No significant differences in the GLS values were observed among the three pairs of comparisons using TomTec software. However, significant differences were noted in the GLS values among two of three pairs of comparisons using Epsilon software. The limits of agreement of the differences (bias) ranged from ±3.12% to ±4.53% for TomTec ( Figure 5 ) and from ±3.05% to ±3.29% for Epsilon ( Figure 6 ). We also performed a head-to-head comparison of TomTec and Epsilon for the same ultrasound image with different ultrasound vendors ( Table 4 and Figure 7 ) and observed a modest correlation. When studies from all three vendors were included in the analysis, the correlation value did not improve considerably.

Table 4

GLS by vendor-independent software between different images by different ultrasound vendors

Software Bias P 95% LOA r
GE: 20.44 ± 2.95 Philips: 19.74 ± 2.67 −0.70 .0806 −4.46 to 3.06 0.77
GE: 20.86 ± 1.83 Toshiba: 20.21 ± 1.94 −0.65 .0815 −3.77 to 2.47 0.65
Philips: 19.20 ± 3.01 Toshiba: 19.36 ± 2.26 0.16 .7138 −4.37 to 4.69 0.65
GE: 18.50 ± 2.29 Philips: 17.33 ± 1.85 −1.17 .0010 −4.23 to 1.89 0.74
GE: 18.68 ± 1.71 Toshiba: 17.43 ± 1.97 −1.25 .0012 −4.24 to 1.79 0.65
Philips: 17.09 ± 2.29 Toshiba: 17.63 ± 2.57 0.55 .0889 −2.72 to 3.81 0.77
TomTec vs. Epsilon
GE: 20.51 ± 2.36 GE: 18.50 ± 2.03 −2.01 <.0001 −6.09 to 2.07 0.56
Philips: 19.43 ± 3.00 Philips: 17.20 ± 2.09 −2.23 <.0001 −7.31 to 2.85 0.53
Toshiba: 19.65 ± 2.71 Toshiba: 17.27 ± 2.17 −2.38 <.0001 −6.94 to 2.18 0.56
All: 19.92 ± 2.70 All: 17.69 ± 2.27 −2.23 <.0001 −6.67 to 2.22 0.60

Only gold members can continue reading. Log In or Register to continue

Stay updated, free articles. Join our Telegram channel

Apr 21, 2018 | Posted by in CARDIOLOGY | Comments Off on Intervendor Variability of Two-Dimensional Strain Using Vendor-Specific and Vendor-Independent Software

Full access? Get Clinical Tree

Get Clinical Tree app for offline access