A review of current trends in three-dimensional analysis of left ventricular myocardial strain

3D strain analysis starts by generating a region of interest (ROI), followed by segmentation of the left ventricle into a 16- or 17-segment model. The size of the ROI differs among 3D strain analytical software packages (LV subendocardial myocardium or the entire LV myocardium). Moreover, some software uses a library of various LV shapes from a large volume of patient data, and selects an appropriate LV shape to generate an ROI in each case (artificial intelligence). While 2D speckle tracking analysis pursues areas that contain specific natural acoustic markers, frame by frame within the ROI (pattern matching), 3D speckle tracking analysis tracks cubes with specific 3D patterns of acoustic markers within the ROI (block matching) to calculate global and regional 3D strain components (Fig. 1). Since the endocardial border is tracked throughout the cardiac cycle, the software also provides time domain LV volume curves from which the LV ejection fraction (LVEF) is calculated. In addition to longitudinal strain, circumferential strain, and radial strain, 3D strain software provides new deformation parameters, such as area strain and principal strain (Figs. 1, 2, 3 and 4). Area strain is determined as the percentage decrease in the size of endocardial (or mid-myocardial) surface area, defined by vectors of longitudinal and circumferential deformation at end-systole from its original area at end-diastole [6]. Principal strain represents the major direction and magnitude of the deformation in which no shear strain occurs. It reflects longitudinal, circumferential, and torsional deformation; therefore, it can represent dynamic 3D movements of the left ventricle [7].

It is important to note that each software system has its own proprietary tracking algorithm, with different analytical approaches. Although strain values should be 0 at the start and end of the cardiac cycle, this does not always happen in regional speckle tracking analysis. Some software rejects regional strain values if drift exceeds a certain level, but other vendors use drift compensation to force strain values to 0 at the end of the cardiac cycle. Radial strain is calculated as a change in transmural wall thickness during a cardiac cycle. Since the myocardium is incompressible, some vendors estimate radial strain from segmental area changes, assuming volume conservation [6]. Strain values may be represented as end-systolic strain or peak strain. These differences result in inter-vendor variability of 3D strain values that are significant, even when 3DE images are acquired from the same subjects [8].

2D strain requires acquisition of multiple 2DE images, including three short-axis views to measure global circumferential strain (GCS) and global radial strain (GRS), and three apical views to measure global longitudinal strain (GLS), resulting in longer data acquisition times. For analysis of all strain components by 2DE, adequate quality of both parasternal and apical images is necessary. However, temporal and spatial resolution are higher, and feasibility is usually high. There have been several validation studies, and normal 2D global strain values have been reported from various large-scale studies [9‐12]. Since 2D strain analysis is performed on a fixed 2DE cutting plane, some speckles or features may be lost in the 2DE imaging plane during a cardiac cycle, due to out-of-plane or twisting motion of the left ventricle [13].

In contrast, only a single apical acquisition is required for 3D strain analysis, resulting in shorter acquisition times and the opportunity to measure all 3D strain components from a single cardiac cycle. 3D strain measurements are not impacted by out-of-plane and twisting motion. However, 3D strain suffers from lower temporal and spatial resolution, which adversely affects tracking reliability. Multi-beat acquisition sometimes produces a stitching artifact between sub-volumes, which results in inaccurate speckle tracking analysis.

Reliability and accuracy of 3D strain measurements have been validated in several studies (Table 1) [14‐18]. First, the accuracy of regional longitudinal, circumferential, and radial strain measurements was determined in experimental studies under different loading conditions, using sonomicrometry data as a reference [14]. There is good agreement between 3D regional strain measurements and corresponding values acquired using sonomicrometry. The best results have been observed in circumferential strain. The same authors subsequently reported that endocardial area strain correlated strongly with sonomicrometry data [15]. Area strain values differed significantly with baseline, pharmacologic stress, and acute ischemia, whereas there were no differences in longitudinal and circumferential strains at baseline or with propranolol infusion or dobutamine infusion, indicating that area strain is a more sensitive parameter than longitudinal or circumferential strain for detecting subtle changes in LV deformation. Other authors applied 3D strain analysis in humans to validate its accuracy. However, it is important to note that there is no true “gold standard” for 3D myocardial mechanics [19]. Instead they employed surrogates, such as cardiac magnetic resonance myocardial tagging [16], feature tracking [18], or 2DE/3DE-derived LVEF [17].

The main impediment to acceptance of 3D strain is the lack of proof that it is truly superior to 2D strain for clinical use, beyond its theoretical advantages, resulting in its use mainly for research. A recent European Association of Cardiovascular Imaging (EACVI) survey from 96 echo laboratories in 22 European countries revealed that only 32% of European laboratories routinely use transthoracic 3DE, mainly due to limitations in image quality and operator dependency [20]. We performed a meta-analysis concerning the feasibility of 2D and 3D GLS in studies with sample sizes > 100 patients [21‐29]. Feasibility of 3D GLS was 85% (95% confidence interval (CI): 79 to 89%). Corresponding values of 2D GLS were 91% (95% CI: 88 to 93%). If we compared feasibility of 3D GLS among different ethnicities, values were 91% (95% CI: 84 to 95%) in Asians, 81% (95% CI: 71 to 88%) in Europeans, and 73% (95% CI: 65 to 80%) in American patients. These results showed that feasibility of 3D GLS is lower than that of 2D GLS, and that feasibility is better in Asian patients than in American or European patients. However, there could be a selection bias, since some studies only selected patients with adequate 2DE image quality [23‐25], and patients with poor 2DE image quality (the authors did not mention the number of these patients in the manuscripts) were excluded before strain analysis. Thus, actual feasibility of 2D/3D strain in consecutive series of patients could be much lower than the observed results. All published studies came from echocardiographic laboratories where physicians and sonographers were thoroughly familiar with 3DE data acquisition and analysis. Thus, the results may not be generally applicable.

We performed a meta-analysis to compare global 2D and 3D strain values using the same ultrasound vendor’s 2D and 3D strain software. We selected only publications in which 2D GCS/GRS were measured using three short-axis views, and 2D GLS was measured with three apical views. We found 3846 paired comparisons of GLS between 2DE and 3DE in 36 publications using one of the three vendors (GE Healthcare, TomTec Imaging Systems, Toshiba Medical Systems) [13, 17, 18, 21‐23, 25‐28, 30‐55]. The mean value of 3D GLS was significantly lower than that of 2D GLS, with a mean bias of 1.4% (Fig. 5). The pooled mean value of 3D GLS for GE and Toshiba was significantly lower than that of 2D GLS. However, there were no significant differences when using TomTec software. Regarding GCS, there were no statistically significant differences between 3D GCS and 2D GCS in 1894 paired comparisons [13, 18, 23, 25, 26, 30, 35‐37, 40‐43, 48, 50, 56]. Vendor analysis showed that 3D GCS was significantly smaller than 2D GCS when GE analytical software was used, but 3D GCS was significantly larger than 2D GCS when we used Toshiba or TomTec software. Finally, there were no significant differences between 3D GRS and 2D GRS in 1778 paired comparisons from 14 studies [18, 23, 25, 26, 30, 36, 37, 40‐43, 48, 56, 57]. Although software from all three vendors yielded no differences in GRS between the two techniques, results in each publication showed different trends, such that 2D GRS was larger than 3D GRS in some studies [25, 26, 42, 43] and 2D GRS was smaller than 3D GRS in others [23, 36, 37, 41, 56]. These results showed that differences in global strain between 2DE and 3DE vary according to the ultrasound vendor. Thus, 2D and 3D strain are not interchangeable, and we need vendor-dependent reference values of 3D strain.

Truong et al. [58] recently performed meta-analysis to determine normal ranges of LV 3D GLS, 3D GCS, 3D GRS, and 3D GAS in 2346 subjects from 32 studies. The authors reported that the mean value of 3D GLS was 19.1%, ranging from 15.8 to 23.4% among the studies. However, the majority of selected studies had small sample sizes (n = 20–50), and the authors were forced to exclude one large study because median and interquartile range were presented in the manuscript, rather than means ± standard deviations [59]. When we searched publications in which at least 100 healthy subjects were evaluated, there were seven independent datasets from six publications (Table 2) [59‐64]. Overall, the mean value of 3D GLS was 18.1% (95% CI: 16.1 to 20.1%). Mean values of 3D GLS ranged from 15.2 to 21.0% among the studies, and if we defined the lower limit of normal (LLN) as the mean – 1.96SD, LLN ranged from 11.1 to 15.9%. The mean value of 3D GLS in male subjects (n = 696) was 18.4% (95% CI: 16.5 to 20.3%) in four studies that addressed gender-based strain analysis [59, 60, 62, 63]. The corresponding value in female subjects (n = 818) was 19.8% (95% CI: 17.7 to 21.8%). It is interesting that 3D GLS in female subjects was consistently and significantly higher than in male subjects, irrespective of the software used for the analysis, with a mean bias of 1.3% (95% CI: 0.9 to 1.7%). This finding suggests the need for establishment of gender-dependent reference values. Finally, as in a previous 2D GLS study [12], there were inter-vendor differences in 3D GLS reference values. The lowest values were observed when Toshiba software (15.4 ± 1.4%, n = 634) was used, followed by GE software (18.7 ± 2.6%, n = 265). The highest values were obtained with TomTec software (20.6 ± 3.0%, n = 946). This is partly because ultrasound vendors use different ROIs for speckle tracking (both TomTec and Toshiba software use subendocardial tracking [14], but GE software uses full myocardial tracking). Different 3D dataset characteristics in each vendor is another cause of discrepancy [65]. The World Alliance Societies of Echocardiography Normal Values Study plans to address whether racial differences in 3D strain are observed [11].

Since clinical applications of 3D strain are still limited, we focus here on its usefulness in several clinical scenarios.

Multimodality imaging allows fusion imaging [102‐104]. Mor-Avi and colleagues performed fusion imaging using computed tomography coronary angiography (CTCA) and 3D longitudinal strain [103, 104]. Resting 3D regional longitudinal strain was color-coded to detect regional abnormalities in a 3D map. Stress CT perfusion was also color-coded to detect regional perfusion abnormalities. Fractional flow reserve (FFR) was also measured using CTCA. The presence of resting strain abnormalities had 71% sensitivity and 81% specificity for detecting > 50% stenosis in CTCA and stress-induced myocardial perfusion defects within specific coronary vascular beds. It had a sensitivity of 83% and a specificity of 81% for detecting stress-induced myocardial perfusion and FFR < 0.80. These results suggest that fusion imaging with 3D strain and CTCA provide valuable information to detect functionally significant coronary stenosis, even at rest. Additional large-scale studies are required to validate whether 3D regional longitudinal strain abnormalities at rest reflect solely reductions of coronary blood flow due to functionally significant coronary stenosis.

Like the 2D strain standardization taskforce [105‐108], the American Society of Echocardiography/ EACVI and instrument partners should work together to identify causes of inter-vendor variability of 3D strain measurements with the objective of reducing it. Current 3D LV strain analysis is predominantly performed to evaluate global LV function. Segmental 3D strain abnormalities may represent specific pathologies. However, accurate 3D segmental tracking requires clearer 3D images with fine spatial resolution and high temporal resolution. Although one-beat acquisition of 3DE full-volume datasets using smaller 3DE transducers is now possible, spatial and temporal resolution are still not adequate to perform reliable segmental deformation analysis. This limitation also causes some underestimation of 3D global strain values, compared with corresponding 2D global strain values. Non-segmental (e.g., coronary artery territories) analysis of deformation is one solution to overcome current problems for segmental analysis [109]. Myocardial curvature analysis or LV shape analysis is another fruitful field of research, because it will provide useful information regarding disease severity and prognosis [99]. Since 3D strain simultaneously provides multidirectional strains, extraction of shear strain components which combine deformation in two different directions is also an interesting field for research [7, 110]. 3DE could be more useful than 2DE in complex and irregular heart chambers, like the right ventricle. 3D strain software aimed at other cardiac chambers is just starting and is now commercially available. It will soon be determined whether 3D RV global strains are more robust parameters than 3D RVEF to predict future outcomes, and whether 3D RV global strains detect subtle RV dysfunction in patients whose 3D RVEF is still normal. Finally, fully automated 3D strain analytical software will eliminate observer variability of 3D strain measurements, and will facilitate the use of 3D strain for routine adoption in the clinical arena.