Comprehensive evaluation of left ventricular deformation using speckle tracking echocardiography in normal children: comparison of three-dimensional and two-dimensional approaches

In this single center study, we prospectively recruited 105 consecutive pediatric subjects (age 0–18 years) with normal cardiac anatomy and function, who were referred to the Cardiology clinic at Children’s Mercy Hospital, Kansas City, for variety of reasons such as murmur, chest pain and palpitations. Subjects with echocardiographic evidence of congenital heart disease, cardiomyopathy, frequent ectopy hindering the evaluation of consecutive sinus beats, cardiac pacing or previous cardiac surgeries were excluded. Demographics and clinical information were obtained from the echocardiogram reports or the individual medical records. This study was approved by the Institutional Research Board of Children’s Mercy Hospital and informed assent/ consent was obtained from all subjects and their legal guardians.

Transthoracic echocardiography (TTE) was performed with the Philips ultrasound system “EPIQ 7”, using age and weight appropriate 2D and 3D matrix array transducers. Each subject underwent consecutive 2D and 3D TTE image acquisition by the same experienced operator using the same cardiac ultrasound system. Three-dimensional and 2D data sets were archived separately for blinded analysis.

For 2D echocardiography, three-beat clips of the multi-planar LV apical views (four, two and three chamber) were optimized for strain analysis using gain, compression, sector width and depth to maximize frame rate (> 60 frames/sec) and capture optimal myocardial tissue definition. Two-dimensional ejection fraction and end-diastolic volumes were calculated using the 5/6 area-length method [15].

For 3D echocardiography, full volume, multi-beat acquisitions of the LV were obtained from the apical, four chamber LV focused views. Image depth, width and gain were adjusted to achieve a volume rate of at least 30 volumes/sec. Acquisitions were optimized to include the entire LV including the apex and were obtained during breath hold (whenever possible) to avoid any stitch artifact.

Two-dimensional and 3D echocardiographic acquisitions were analyzed using the TomTec software Image-Arena version 4.6 SP3, Unterschleissheim, Germany; 4D LV 3.1 for 3D STE and 2D CPA for 2D STE. For 2D GLS analysis, a single cardiac beat with the best endocardial definition was used. The region of interest was traced in end-systole by delineating the endomyocardial contour from medial to lateral mitral valve (MV). Tracking was automatically performed and manually adjusted if necessary, based on visual inspection of tracking (Fig. 1). As per EACVI task force recommendations [16], GLS was calculated as the change in the “length of the line” in each apical view and LV GLS was calculated as the average of GLS of the three LV apical views.

For 3D analysis, full-volume 3D data sets with minimal amount of dropout and stich artifacts were analyzed on a separate computer workstation by a separate cardiologist to minimize bias. Three-dimensional LV volumes, LVEF and strain were measured using the semi-automated border detection algorithm for four-, two-, and three-chamber views (Fig. 2). After confirming the anatomic landmarks (mitral annulus and LV apex), the software generated end-diastolic and end-systolic endocardial contours of the LV in each of the above cut-planes. These contours were manually adjusted when necessary, to optimize boundary position and tracking throughout the cardiac cycle (Fig. 2). The LV outflow tract, papillary muscles, and trabeculations were included within the LV cavity.

For regional strain analysis, 2D and 3D images of the LV wall were automatically divided into 16 segments and displayed as a bull’s eye figure. Two-dimensional and 3D strain analysis were performed by independent blinded readers (Figs. 1 and 3).

Strain analysis was repeated on 20 randomly selected patients by the same reader after 4 weeks for intra-observer variability and by a second reader for inter-observer variability.

Patients with poor quality 2D images defined as those with foreshortened apical views, significant lung artifacts obscuring the LV apex or LV walls, and or frame rate < 60 frame per second) were excluded from analysis. Poor quality 3D full volume acquisitions defined as those with a significant stitch artifact, poorly visualized MV, or apex, and or volume rate < 30 volumes per second) were also excluded. For both 2D and 3D strain analysis, tracking was automatically performed by the strain analysis software, however manual edits were made as deemed necessary and analysis was accepted only after visual inspection. If tracking was suboptimal, the endocardial border was retraced. If satisfactory tracking was not accomplished after three re-tracings, specifically, if more than three segments had poor tracking, the study was deemed not analyzable and was subsequently excluded. Time required for acquisition and for offline analysis was recorded for the 2 modalities.

Statistical analysis was performed using SPSS for Windows version 26.0 (SPSS, Inc., Chicago, IL, USA). Kolmogorov-Smirnov test was used to test for normality. Categorical data were presented as percentages, continuous and normally distributed data were presented as mean ± SD, while data deviating from normality were expressed as median and inter-quartile range. Correlation between 2D and 3D strain was performed using Pearson correlation coefficient (r) and linear regression. As recently recommended by Bunting et al. [17], we used Bland Altman analysis including bias and limits of agreement, intra- class correlation coefficients (ICC); (two-way mixed model, absolute consistency between single measurements) and percent error (100 x difference/mean) to quantify the agreement between the 2 techniques.

Good agreement was indicated by near zero bias and narrow limit of agreement (LOA) relative to the measured values. Guidelines recently recommended by Bunting et al. [17] were used for the interpretation of ICCs. An ICC < 0.40 denoted poor agreement, 0.40–0.59 was fair, 0.60–0.74 was good, and ICC of 0.75–1.00 indicated excellent agreement. Statistical significance was indicated by P value < 0.05.

Baseline demographic and echocardiographic characteristics are summarized in Table 1. The mean age of subjects included was 11.2 ± 5.5 years, 20 of whom were ≤ 5 years old. All subjects had normal LV systolic function as quantified by the 5/6 area-length 2D LV ejection fraction (EF) with a mean of 62 ± 4.1% and by 3D EF with a mean of 61 ± 3.6%.

One hundred five subjects were originally enrolled in the study. Eight subjects were excluded from 2D analysis due to poor quality 2D images. Four additional 2D datasets were excluded due to poor tracking upon strain analysis, with an overall 2D STE feasibility rate of 88.6%. Fifteen subjects were excluded from 3D analysis due to poor quality 3D full volume acquisitions and inadequate tracking, (nine of whom were younger than 5 years old, with heart rate > 120 bpm) with an overall 3D STE feasibility rate of 85.7%. This had a statistically insignificant p value when compared to 2D STE feasibility rate (85.7 vs 88.6%, p = 0.21).

The average temporal resolution of the 2D and 3D datasets included was 78 ± 9 frames per second, and 39 ± 7 volumes per second, respectively. The average acquisition time and average offline analysis time was significantly shorter for 3D STE vs 2D STE (acquisition time = 1 ± 1.2 mins vs 2.4 ± 1 mins; p = 0.03, analysis time = 3.3 ± 1 vs 8.2 ± 2.5 mins; p = 0.001, respectively).

Comparison of the global and regional strain values by 2D and 3D STE for the entire cohort is presented in Table 2. The correlation and measures of inter-technique agreement between 2D and 3D strain measurements are presented in Table 3. There was an insignificant difference, strong and significant correlation (r = 0.73, p < 0.05) as well as excellent agreement (Bias = − 0.3, LOA = − 3.7 -3.6, percent error = 1.4% and ICC = 0.82) between 2D and 3D GLS (Fig. 4). On the other hand, there was a varying degree of correlation (r = 0.21–0.51) and agreement between the different myocardial segments for regional strain (Bias = − 0.2- 3.7, LOA = − 21.7- 19.7, percent error = 2.2–30.3% and ICC = 0.37–0.74) with overall weaker correlation and less agreement in the basal than apical segments (Table 2).

Table 4 displays the degree of intra and inter-observer reproducibility (measured by ICC) for 2D and 3D STE global and regional strain. The intra and inter-observer reproducibility was equally robust for LV GLS between the two techniques (ICC range of 0.81–0.92). There was less degree of reproducibility among the different myocardial segments (ICC range of 0.21–0.73), than with global strain, with a trend towards higher ICCs among the apical than the basal regions. This did not seem to vary between 2D and 3D STE techniques.

Two-dimensional STE is a technique that utilizes frame-by-frame tracking of acoustic speckles to provide a quantitative assessment of myocardial deformation [2]. Since, there are no geometric assumptions or significant influence of loading conditions, this technique is believed to be superior to conventional 2D echocardiographic measures of ventricular function such as EF. However, given the complex spatial arrangement of the LV myofibrils and simultaneous multi-vector contractions, it is not unexpected for strain measurements limited to three one-dimensional planes to suffer planar simplification and out-of-plane motions.

To circumvent this limitation, 3D STE has emerged as a newer, potentially more physiologically sound technique to quantify ventricular deformation, which by nature is a 3D phenomenon. Therefore, 3D STE is not affected by the out- of- plane or twisting motion or apical foreshortening. Additionally, and in contrast to 2D STE, a single apical full volume acquisition can provide comprehensive data about 3D LV global and regional strain, EF, volumes, and sphericity index. Moreover, 3D STE can provide novel myocardial deformation parameters such as area strain [18], principal strain and tangential strain [19].

Akin to adult reports, our study noted good agreement and strong correlation between 2D and 3D LV GLS with overall slightly lower 3D GLS values. A recent meta-analysis performed to compare global 2D and 3D strain values in adults, with 3846 paired comparisons of GLS included from 36 publications, demonstrated insignificant difference when TomTec software was used [19]. Interestingly, the pooled mean values of 3D GLS for other vendors such as GE and Toshiba were significantly lower than that of 2D GLS. While on the pediatric side, we are only aware of a single study where the two techniques were compared. In this study, analysis was performed using an older version of TomTec, in a smaller, mixed cohort of patients with congenital heart disease wherein the ventricular geometry was likely more abnormal and varied than in our study [20]. Nevertheless, strong correlation (r = 0.92 p < 0.001) and minimal difference were still appreciated between 2D and 3D analyis, although (and unlike our results) 3D GLS tended to have higher mean values than 2D GLS.

The variability between 2D and 3D strain analysis could be due to number of factors, such as the discordant frame/volume rates between the two techniques. The two techniques also have different tracking and analysis algorithms with 3D strain being calculated from cubes with specific 3D patterns of acoustic markers (block matching) vs 2D strain which is calculated based on areas that contain specific natural acoustic markers, frame by frame, within the region of interest (pattern matching) [13].

3D STE is capable of providing quantitative regional strain analysis. The role of 3D regional strain is still evolving, with studies demonstrating useful applications in the assessment of myocardial ischemia and viability, although reliable measurement of regional strain remains challenging [19, 21, 22]. In our study, there was a varying degree of correlation and agreement between 2D and 3D peak longitudinal strain values among different myocardial segments. Correlation was weaker, and limits of agreements were wider in the basal compared to apical segments, which is likely since basal segments are farther in the field and have the most active excursion, hence are the hardest to track by 2D and 3D STE.

The overall modest agreement and correlation between 2D and 3D regional strain is consistent with the currently available clinical adult studies [23]. This is thought to be secondary to difficult matching of corresponding LV segments between 2D and 3D STE. Regional strain also tends to be significantly more sensitive to noise and measurement error than global strain as it does not benefit from the favorable influences of averaging [24]. Therefore, individual 2D and 3D regional strain values may not be interchangeable. Instead, grouped segments (for instance by coronary territories), might suffer less from the aforementioned challenges and potentially provide better agreement and more clinically relevant data; an approach that we intend to investigate in the future.

In this study, we report a feasibility of 88.6% for 2D STE and 85.7% for 3D STE. While our 2D STE feasibility data seem to be in good match with the previously published pediatric and adult data (average 80–97%) [13, 25, 26], our 3D STE feasibility rates seem to be significantly higher than the average reported rates (average 63–83%) [19, 27, 28].

The accuracy of 3D STE largely depends on optimal image quality with sufficient frame rate which require a remarkable amount of dedicated training and practice to obtain. In our echocardiography lab, we utilize 3D imaging on a regular basis and are currently reporting 3D LV EF on all our standard function-focused pediatric echocardiograms. For this goal to be finally achieved, our sonographers received extensive one-on-one training and have consequently developed the skills needed for optimal 3D volumetric imaging. Our results confirm that with adequate training, 3D strain analysis can be nearly as feasible as 2D in the pediatric population [26, 29‐31].

Our results also demonstrate a significantly higher efficiency with significantly shorter acquisition and analysis time for 3D vs 2D STE. This is somewhat expected given the fact that a single 3D apical dataset is required for both 3D GLS, while 3 different apical planes need to be optimized and acquired for 2D GLS. This is of utmost importance, as high-volume echocardiography laboratories strive for an accurate, reproducible as well as efficient quantitative method of ventricular function to adopt into clinical routine on a wider scale.

Reproducibility of 3D STE has been reported as acceptable to excellent in several studies with intra-observer variability ranges from 1 to 13% and inter-observer variability ranges from 2 to 14% [12, 32, 33]. As with 2D STE, our data showed higher variability among 3D regional than LVGLS. The excellent reproducibility of global data observed in our study could be reflective of the role of semiautomatic/ automatic tracking and segmentation in minimizing variability during strain analysis.