Summary
Background
Speckle tracking is a relatively new, largely angle-independent technique used for the evaluation of myocardial longitudinal strain (LS). However, significant differences have been reported between LS values obtained by speckle tracking with the first generation of software products.
Aims
To compare LS values obtained with the most recently released equipment from two manufacturers.
Methods
Systematic scanning with head-to-head acquisition with no modification of the patient’s position was performed in 64 patients with equipment from two different manufacturers, with subsequent off-line post-processing for speckle tracking LS assessment (Philips QLAB 9.0 and General Electric [GE] EchoPAC BT12). The interobserver variability of each software product was tested on a randomly selected set of 20 echocardiograms from the study population.
Results
GE and Philips interobserver coefficients of variation (CVs) for global LS (GLS) were 6.63% and 5.87%, respectively, indicating good reproducibility. Reproducibility was very variable for regional and segmental LS values, with CVs ranging from 7.58% to 49.21% with both software products. The concordance correlation coefficient (CCC) between GLS values was high at 0.95, indicating substantial agreement between the two methods. While good agreement was observed between midwall and apical regional strains with the two software products, basal regional strains were poorly correlated. The agreement between the two software products at a segmental level was very variable; the highest correlation was obtained for the apical cap (CCC 0.90) and the poorest for basal segments (CCC range 0.31–0.56).
Conclusions
A high level of agreement and reproducibility for global but not for basal regional or segmental LS was found with two vendor-dependent software products. This finding may help to reinforce clinical acceptance of GLS in everyday clinical practice.
Résumé
Contexte
Le speckle tracking est une technique relativement nouvelle, largement indépendante de l’angle, utilisée pour l’évaluation du strain myocardique longitudinal. Toutefois, des différences significatives ont pu être mises en évidence entre les valeurs de strain longitudinal obtenues par speckle tracking avec les premières générations des logiciels de différents vendeurs.
Objectifs
Comparer les valeurs de strain longitudinal obtenues avec les 2 plus récentes versions de logiciels de 2 vendeurs.
Méthodes
Une échocardiographie sans modification de la position du patient était réalisée chez 64 patients par 2 échographes de 2 vendeurs différents pour obtention en déporté du strain longitudinal (Philips QLAB 9,0 et GE EchoPAC BT12). La variabilité interobservateur de chaque logiciel était étudiée sur un échantillon aléatoire de 20 patients issus de la population de l’étude.
Résultats
Les coefficients de variation de GE et Philips pour le strain longitudinal global étaient de 6,63 % et 5,87 % respectivement, indiquant une bonne reproductibilité. La reproductibilité était très variable au niveau segmentaire, avec des coefficients de variation variant de 7,58 % à 49,21 % pour le strain longitudinal avec les 2 logiciels. Le coefficient de concordance pour le strain longitudinal global était haut à 0,95, ce qui indique une bonne concordance entre les 2 méthodes. Tandis qu’une bonne concordance était observée pour le strain apical et médian, les valeurs de strain régionales basales étaient peu corrélées. La concordance entre les 2 méthodes était très variable au niveau segmentaire, les meilleures corrélations étant obtenues pour l’apex (coefficient de concordance à 0,90) et les plus mauvaises étant obtenues pour les segments basaux (coefficients de concordance allant de 0,31 à 0,56).
Conclusions
Une bonne concordance et une bonne reproductibilité sont retrouvées pour le strain longitudinal global de 2 vendeurs, mais pas au niveau régional basal ou segmentaire. Les données de cette étude pourraient aider à renforcer l’importance du strain longitudinal global en pratique clinique.
Background
Longitu dinal strain (LS) describes myocardial deformation, i.e. the fractional change in length of a myocardial segment. Speckle tracking is a relatively new, largely angle-independent technique used for the evaluation of myocardial LS that has been experimentally validated against sonomicrometry . Global LS (GLS) is the average longitudinal component of strain in the entire myocardium, which can be approximated by the averaged segmental strain components in individual myocardial wall segments . Clinical studies have demonstrated the major additional diagnostic and/or prognostic contribution of GLS compared with conventional indices of left ventricular (LV) systolic function in various clinical settings, such as heart failure, valvular heart disease or cardiomyopathies . However, previous reports have demonstrated significant differences between LS values obtained by speckle tracking with the first generation of software products released by various manufacturers . Post-processing appears to be the most important determinant in intervendor variation, while acquisition appears to have only a limited effect . However, speckle tracking standardization among manufacturers is essential, as clinicians must be able to interpret data generated by different devices, irrespective of the vendor . The present study was therefore designed to compare GLS and segmental LS values obtained with the most recent releases from two different manufacturers. To address this issue, systematic scanning with head-to-head acquisition was performed in patients with equipment obtained from two different manufacturers (Philips and General Electric [GE]), with subsequent off-line post-processing for speckle tracking LS assessment.
Methods
Study population
Patients referred for echocardiography during a 2-week period were screened for the following characteristics: good visualization of all LV segments allowing speckle tracking and measurement of LV GLS, sinus rhythm and consent to participate. Ninety-six patients were initially screened for inclusion in the present study. Twenty-three patients were excluded for poor echogenicity, as speckle tracking was not possible in at least one LV segment. Nine patients were also excluded for suboptimal echogenicity. The final study cohort consisted of 64 patients.
Standard echocardiography and workflow
Transthoracic echocardiograms were acquired by using two commercially available ultrasound transducers and equipment (X5-1 probe, iE33, Philips, Andover, MA, USA; M5S-D probe, Vivid E9, GE Medical, Milwaukee, WI, USA), both located in the same echocardiography room. Image acquisition was performed by three experienced sonographers (S.M., A.-L.C., and F.D.). Each participant first underwent comprehensive assessment of cardiac anatomy and function with one of the ultrasound systems. The order of examination on the two machines was randomized. Acquisitions with the two systems were performed during the same echocardiographic examination, with no modification of the patient’s position. Sector size and depth were adjusted to achieve optimal visualization of all LV segments at the highest possible frame rate. The same frame rate was used with the two machines. Acquisition was obtained at the end of expiration. At least three video loops of one cardiac cycle were obtained for apical views. All echocardiographic examinations were stored as raw data in a picture archiving and communicating system (PACS) for subsequent off-line analysis on dedicated Xcelera and EchoPAC workstations.
Speckle tracking strain echocardiography
QLAB software 9.0 was used for images obtained from Philips IE33 and EchoPAC BT12 software was used to analyze those obtained with the Vivid E9. The three apical views obtained on each ultrasound machine were measured with each software package to obtain peak systolic LS (%). All acquired apical views were available for off-line quantification. LS values were computed after having determined the onset of aortic valve closure using Doppler recordings or visual inspection of the kinetics of the aortic valve in long-axis views.
The automatic tracking of the endocardial contour was performed in end-systole with EchoPAC software and in end-diastole with the QLAB software. Tracking was carefully verified and the region of interest was manually corrected to ensure optimal tracking and to cover the entire thickness of the LV myocardium. Longitudinal two-dimensional speckle tracking strain values were analyzed off-line by a single investigator (A.-L.C.) blinded to the patient’s identity during image post-processing. The left ventricle was divided using the 17-segment American Society of Echocardiography model to derive segmental LS values . GLS was obtained as the average of regional strains. Basal regional strain was calculated as the average of basal peak strains measured in the three apical views. Midwall regional strain was calculated as the average of midwall peak strains measured in the three apical views and apical regional strain was calculated as the average of apical peak strains measured in the three apical views. Territorial strain was calculated from the perfusion territories of the three major coronary arteries in a 16-segment LV model by averaging all segmental peak systolic strain values in each territory . Details of this territorial segmentation are shown in Fig. 1 . Interobserver variability of each software was tested on a randomly selected set of 20 echocardiograms from the study population.
Statistical analyses
Data are presented as means ± standard deviations or absolute numbers and frequencies. Variabilities between the two software products were evaluated by the mean of the coefficient of variation for each strain measure (CV). CVs were obtained for GLS, for each view (two-, three- and four-chamber), for basal, midwall and apical ventricular segments, for each segment of the 17-segment model and for the territories of the three major coronary arteries. All strain values were compared between two software products using different methods. Pearson’s correlation coefficient ( r) , which measures the extent to which each value differs from the best-fit line, is a measure of precision. The concordance correlation coefficient (CCC) was calculated as the product of r and the bias correction factor (Cb), which measures the extent to which the best-fit line differs from the 45° line through the origin, and is a measure of accuracy . Bland-Altman plots of differences between GLS values by GE and Philips and mean GLS values were obtained to detect a potential bias and to obtain limits of agreement (LOA) . Statistical analyses were performed using PASW 18.0 (IBM Inc., Bois-Colombes, France) and MedCalc for Windows version 12.5.0 (MedCalc Software, Mariakerke, Belgium).
Results
Patient characteristics
The baseline characteristics of the study population are summarized in Table 1 . The clinical indication for echocardiography was heart failure in 16 (25%) patients, ischemic heart disease in two (3%) patients and chest pain in eight (13%) patients. In addition, three (5%) patients had hypertrophic cardiomyopathy and 11 (17%) patients had significant valvular heart disease. Echocardiography was performed for other reasons in 17 (27%) patients (hypertension, diabetes, preoperative evaluation for non-cardiac surgery, diabetes or stroke). Among the 64 patients included in the present study, seven control subjects (11%) free of any cardiovascular disease, without any electrocardiogram abnormality and with a normal echocardiogram, were recruited from the hospital staff.
| Variables | |
|---|---|
| Men | 38 (59) |
| Age (years) | 62 ± 17 |
| Body mass index (kg/m 2 ) | 26.2 ± 5.4 |
| Diabetes | 13 (20) |
| Hypertension | 28 (44) |
| Smoking | 12 (20) |
| Dyslipidemia | 25 (39) |
Echocardiography
The mean frame rate for GE images was 62 ± 3 frames/s for apical four-chamber views, 62 ± 4 frames/s for two-chamber views and 63 ± 4 frames/s for apical long-axis views. The mean frame rate for Philips images was 61 ± 7 frames/s for apical four-chamber views, 61 ± 7 frames/s for two-chamber views and 61 ± 7 frames/s for four-chamber views. Mean LV ejection fraction was 52 ± 17%, mean LV end-diastolic volume index was 81 ± 35 mL/m 2 and mean LV end-systolic volume index was 43 ± 36 mL/m 2 . LV ejection fraction was < 50% in 18 (28%) patients.
Reproducibility of longitudinal strain with each software
As shown in Fig. 2 , GE and Philips interobserver CVs for GLS were 6.63% and 5.87%, respectively. Interobserver variability was substantially higher for basal strain than for midwall and apical strain, with higher CVs with the QLAB software compared with the EchoPAC software. In addition, interobserver CVs were substantially higher for each apical view with the two software products than for GLS ( Fig. 2 ). CVs for each perfusion territory and segmental level are shown in Figs. 2 and 3 , respectively.
