Comparison of strain parameters in dyssynchronous heart failure between speckle tracking echocardiography vendor systems

The Markers and Response to CRT (MARC) study was designed to investigate predictors for response on CRT, including several echocardiographic parameters [17]. The study was initiated and coordinated by the six centres within the framework of the Centre for Translational Molecular Medicine (CTMM), project COHFAR (grant 01C-203), and additionally supported by Medtronic (Fridley, Minnesota, USA). Study monitoring was done by Medtronic, data management and validation by the investigators (MR, BG) in collaboration with Medtronic. The study was approved by the institutional review boards of all participating centres. All patients gave written informed consent. The trial was registered at clinicaltrials.​gov: NCT01519908.

Two hundred forty patients eligible for CRT according to the most recent international guidelines were included in the MARC study [18, 19]. In short, MARC study inclusion criteria were: sinus rhythm and optimal pharmacological heart failure therapy, QRS-duration ≥130 ms in patients with left bundle branch block (LBBB) and QRS-duration of ≥150 ms in non-LBBB patients with NYHA class II and QRS-duration ≥120 ms in LBBB patients with NYHA class III. Exclusion criteria were severe renal insufficiency, an upgrade from a bradycardia pacemaker or CRT-P to CRT-D, permanent atrial fibrillation, flutter or tachycardia, right bundle branch block, and permanent 2nd or 3rd degree atrioventricular block. Before and 6 months after CRT implantation, data were recorded at the outpatient department, including electrocardiographic and echocardiographic examination. Patients were excluded for this sub-analysis if frame rate of the apical four chamber (4CH) view was below 35 Hz, in case of irregular heart rhythm, unanalysable images due to technical errors or if image quality was very poor.

LV ejection fraction, LV end-diastolic and end-systolic (LVESV) volumes were measured by biplane Simpson’s method [20]. Volumetric response to CRT was defined as the percentage of change in LVESV between echocardiographic examination before and 6 months after CRT implantation. Patients were classified as responder in case of ≥15% reduction in LVESV.

Echocardiographic 4CH images were subjected to offline speckle tracking analysis (WE, MC). The optimal images for speckle tracking were selected and used for the vendor dependent and independent platform (Additional file 1: Figure S1). All images were scored for quality (poor, average, or high) by two experienced observers. Image quality was categorized as high if the total LV myocardium was visible during the entire cardiac cycle, average if one or two segments were not clearly visible and poor in all other cases. Images were exported to vendor specific software (GE EchoPac 11.3 and Philips QLAB 10.0) in standard formats and exported as DICOM-files for vendor independent software (TomTec 2D Cardiac Performance Analysis (2DCPA) version 1.2.1.2). Speckle tracking was performed with standard settings for all vendors. For each platform, a region of interest (ROI) was placed by user defined markers to incorporate the entire myocardial wall. Repeat adjustments of the ROI were done if tracking quality was insufficient. The myocardial wall was separated into six segments by all platforms (i.e. basal and mid inferoseptal, apical septal, apical lateral and basal and mid anterolateral). Philips QLAB analyses an additional true apical segment (i.e. 17 segment model of the AHA), which was excluded for the septal and lateral wall strain curves, as it was part of both walls [21]. Segments were also excluded if adequate tracking was not achievable. The basal inferoseptal, mid inferoseptal and apical septal segment were averaged into a global septal wall strain curve. The apical lateral, basal anterolateral and mid anterolateral segment were averaged into a global lateral wall strain curve. Results from a single wall were excluded if tracking of more than one segment was unachievable. The entire myocardial wall was used for both Philips and GE analysis. TomTec analysis resulted in separate datasets for the endocardial and epicardial border. The epicardial border at the apical and mid ventricular lateral wall was often outside the echocardiographic window, and was therefore excluded in TomTec analysis. This was done even though differences between endo- and epicardial layers are known [22]. The marker for reference length (L₀) was placed at onset of QRS-complex for GE derived images, both for GE EchoPac and TomTec 2DCPA. L₀ of Philips derived images could not be altered in QLAB and was automatically placed in the QRS-complex. L₀ was manually placed at a similar position for Philips derived images analysed with TomTec 2DCPA. Therefore, both direct comparisons (GE vs. TomTec and Philips vs. TomTec) had similar L₀ positions.

Results of speckle tracking analysis were stored and exported for offline analysis with Matlab 2014b (Mathworks, Natick, MA, USA). Author written Matlab scripts allowed for input of valve closure times and semi-automatic calculation of strain parameters. Results of strain parameters were based on global strain curves. Global strain curves were averages of the segments representing the global LV or the separate septal or lateral wall.

Two-hundred-eleven of 240 MARC study patients were included in this sub-analysis, 123 in the GE-cohort and 88 in the Philips-cohort. Nineteen patients were excluded for GE analysis, of which five were excluded from the main study, two had irregular heart rhythm, four had a frame rate below 35 Hz, four had overall low image quality and two had only one analysable segment for the lateral wall. Ten patients were excluded for Philips analysis, of which four were already excluded from the main study, five were stored in a datafile not analysable for STE and one had a frame rate below 35 Hz. There were no significant differences in baseline characteristics between cohorts (Table 1), except for frame rate. Frame rate was higher in the GE-cohort (61 ± 12 Hz) compared to the (Philips-cohort 55 ± 7 Hz, p < 0.001). LV end-diastolic and end-systolic volumes tended to be lower in the Philips-cohort compared to the GE-cohort. LV ejection fraction was comparable, as were conventional electrical dyssynchrony (i.e. QRS duration and morphology) and mechanical dyssynchrony parameters (i.e. IVMD, apical rocking and septal flash). IVMD was above the cut-off value of 40 ms in both groups, septal flash was seen in approximately half of all patients, while apical rocking was observed in around 60% of patients. CRT response rate was non-significantly different in the two cohorts (GE-cohort: 59% vs. Philips-cohort: 65%), with non-significantly differences in ESV reduction (GE-cohort: 20 ± 23% vs. Philips-cohort: 25 ± 26, p = 0.208).

Septal wall cross correlation was significantly lower compared to the LV and lateral wall for GE vs. TomTec (septum: 0.682 ± 0.290, LV: 0.835 ± 0.213, lateral wall: 0.800 ± 0.244, p < 0.05) and for Philips vs. TomTec ((septum: 0.712 ± 0.293, global LV: 0.898 ± 0.156, lateral wall: 0.827 ± 0.226, p < 0.05). There was no apparent statistical difference between the three subgroups based on image quality (Additional file 2: Table S1). Only for the lateral wall in GE vs. TomTec did the high-quality images (0.892 ± 0.123) have significantly higher R² compared to the poor-quality images (0.713 ± 0.312, p < 0.05).

To the best of our knowledge, this is the first study to compare results of different STE software packages, specifically for mechanical dyssynchrony in CRT-candidates. The average differences of peak strain and systolic strain were small and non-significant between vendor dependent and vendor independent STE packages. Nevertheless, TomTec had lower values compared to GE EchoPac and higher values compared to Philips QLAB. Unfortunately, we cannot define the source of discordance, as a gold-standard for deformation imaging (i.e. sonomicrometry) was not available in our study. The relative high correlation for global longitudinal strain between STE platforms is in accordance with earlier publications [16]. There are currently no publications on vendor comparison studies on STE in patients with heart failure and dyssynchrony, besides a small comparative study by our own group [24]. Moreover, comparison to earlier publications on peak strain and timing values is difficult as previous studies implemented older versions of STE software, while we used the most recent versions. STE software is constantly under development, partly due to the STE standardization taskforce of the EACVI/ASE. This task force includes among its members representatives of several vendors. Their efforts resulted in small and acceptable differences between vendors for global longitudinal strain [16, 25]. However, the variability among vendors in more specific longitudinal strain features is not yet elucidated, nor is the exact bias between vendors with respect to regional strain assessment. Furthermore, the cohort studied for standardization consisted of a wide range of subjects (mean LVEF 60%, global longitudinal strain −19.2%) and is therefore not representative for CRT patients with dilated hearts, reduced LV function, and complex deformation characteristics [23]. Moreover, CRT patients can have suboptimal acoustic windows which affects image quality and reliability of strain analysis. In the current study, comparable 4CH peak longitudinal strain values were found in CRT patients, although the limits of agreement of Bland-Altman plots were relatively wide, and results for individual patients varied significantly. The discrepancies between the current study and the publications by Farsalinos et al.. and Yang et al. may therefore be ascribed to the examined populations [16, 26]. A mechanistic modelling study showed higher variability in peak strain among vendors and a higher inter-observer variability in a dilated thin-walled LV [27]. This modelling study suggests a lower level of agreement among vendors in heart failure patients, which might explain the findings in our current observations.

Differences between manufacturers are largely attributed to discrepancies in STE algorithms. Albeit recently thoroughly investigated, [16] the algorithms of the majority of commercially available speckle tracking software have lacked published validation [27]. They are furthermore not open-source. TomTec 2DCPA uses DICOM images and thereby imports images with lower frame rate and lower image quality compared to the raw image files used by the vendor dependent platforms. Lower frame rates influence temporal resolution, which hampers reliable assessment of both strain values and timing indices. The image quality directly influences spatial resolution, decreasing reliable tracking of speckles. TomTec also displays separate endo- and epicardial strain curves for each segment, and mean myocardial wall strain results are not given. The use of endocardial strain data might have caused a slight overestimation of peak strain values [22]. GE EchoPac uses ‘global’ wall myocardial strain by default, although users can choose between endocardial, epicardial or mid myocardial layers. Lastly, the method used by Philips QLAB is unknown, although a global myocardial based approach is likely. Timing of the reference length is also of importance for standardization, as differences in the onset of strain curves directly influences absolute strain values as wells as timing indices. As mentioned, timing of reference length was uniformed for TomTec analysis compared to both vendor dependent platforms.

Absolute values of mechanical dyssynchrony indices were significantly lower for Philips, compared to TomTec. Whereas the results on dyssynchrony parameters obtained from GE images displayed higher values for GE compared to TomTec. Although the source of discordance is unknown, dyssynchrony seems underestimated by Philips QLAB speckle tracking algorithms. Underestimation of dyssynchrony is exemplified by results for the SL-delay obtained by Philips QLAB (median 0 ms, interquartile range 0 – 0 ms). The discrepancies of Philips QLAB with both other vendors are remarkable, as both GE EchoPac and TomTec 2DCPA displayed large variation for SL-delay. Moreover, as the Bland-Altman plot of SL-delay in Fig. 4 shows, the discrepancy between Philips QLAB and TomTec 2DCPA, a large number of results are on a line (y = −0.5*x), indicating a large variation in SL-delay for TomTec, while Philips values were mainly close to zero. Philips’ derived septal and lateral wall strain curves were often quite similar, as can be appreciated in the example in Fig. 1. It seems that segmental strain curves are more smoothened by Philips QLAB. While no gold-standard for deformation imaging was applied, the relative absence of dyssynchrony obtained with Philips STE software is striking.

Although intra-observer agreement of strain pattern categorization is relatively good, strain pattern categorization showed apparent variations among vendors. Strain patterns were earlier found to be more robust between vendors [24]. This discrepancy could be attributed to changes in STE algorithms, as there were almost no LBBB type 2 patterns found by Philips. LBBB type 2 is the most distinctive septal deformation pattern, with predominant stretch almost in completely opposite direction to the lateral wall. Higher percentages of LBBB type 2 were observed in the same (i.e. Philips imaged) patient with TomTec. The cohorts of GE and Philips were not significantly different, and conventional dyssynchrony parameters such as apical rocking, septal flash and IVMD were comparable. Therefore, the relative absence of LBBB type 2 patterns is likely caused by the inability to detect dyssynchrony using QLAB. Given the above-mentioned differences in both continuous and categorical dyssynchrony parameters, one might postulate that STE with Philips QLAB is less suitable for detection of dyssynchrony in a CRT population. However, despite the lower values, the predictive value of Philips QLAB derived dyssynchrony parameters is at least comparable to the vendor independent analysis of TomTec 2DCPA. Although Philips QLAB and TomTec 2DCPA were able to predict volumetric response to CRT with the implemented dyssynchrony parameter, the cut-off values were different. Even though cut-off values for GE EchoPac derived parameters were higher, the values are different from earlier published values [3, 9]. These differences may be ascribed to the used software versions or the examined populations. Vendor specific cut-off values should therefore be used for each STE platform.

Echocardiographic analysis in the current study was restricted to 4CH images, as speckle tracking analysis of these images is relevant for dyssynchrony in patients with left bundle branch block and has higher reproducibility [15]. However, echocardiographic 4CH images with adequate image quality can be difficult in patients with dilated hearts. Patient anatomy and cardiac size both complicate echocardiographic acquisition, as can be observed from the number of echocardiograms with poor image quality. Although this is not reflected in our results, lateral wall acquisition can be difficult in heart failure patients. It was surprising that lateral wall cross-correlation values were significantly higher compared to septal wall values for all vendors. Lower septal wall cross-correlation values can be explained by the higher complexity of septal strain patterns (i.e. LBBB type 1 and 2, Fig. 2). This in contrast to strain patterns of the lateral wall, which often had a similar shape between patients, also seen in the agreement on TTP_max for lateral wall strain. Complex septal deformation pattern can more easily be misinterpreted, resulting in lower correlations and wider limits of agreement in Bland-Altman plots. The poor agreement in septal strain analysis was also seen in the low Cohen’s kappa values of septal strain pattern categorization.

Although this study consists of relatively large subgroups, it is a sub-analysis with inherent limitations. However, images were prospectively collected for analysis with STE software. Nonetheless, patients underwent echocardiographic examination by a single vendor, which was assigned dependent of the centre of implantation and therefore non-randomized. Ideally patients would undergo echocardiographic examination by both vendors, making a direct comparison between GE and Philips possible. Moreover, there was no gold-standard used in this study, making it impossible to determine the source of variability. Test-retest variability was not part of the imaging protocol, although consecutive measurements are subject to variation [16]. Nevertheless, the large and comparable subgroups permitted a reliable comparison of vendors, and large differences were seen. This is in contrast to previous studies, in which most echocardiographic dyssynchrony parameters (i.e. SRSsept, SSI, and SL-delay) were tested solely on one vendor (i.e. GE EchoPac).

Although global LV peak strain correlates reasonable between vendor systems, the results of individual patients between vendors may vary, as indicated by the wide limits of agreement in Bland-Altman plots. This variation hampers translation of deformation parameters obtained by STE to clinical practice. All three STE vendors were capable to predict response to CRT, using the implemented dyssynchrony parameters. Although the diagnostic value of GE EchoPac derived parameters is well validated in prior studies, [10, 28] further work is needed to confirm the predictive value of these parameters in clinical practice. Differences between vendors can be large, hampering direct translation from pre-clinical work to the clinical implementation of speckle tracking derived dyssynchrony parameters and patterns in all echo laboratories, with a myriad of echo-machines. We recommend that in patients eligible for CRT, clinicians should use reference and cut-off values specific to the STE vendor.