4. Discussion
Beyond a scientific context, how does speckle-tracking analysis using fetalHQ® perform in routine clinical practice and could less experienced operators benefit from this state-of-the-art technique? The present study aimed to scrutinize the applicability of 2D-STE based on the expertise of two operators.
Quantification of fetal myocardial function is challenging, and several different techniques have been used [6,14,25,66]. Already in 2012, Crispi et al. stated that in expert hands and with adequate ultrasound equipment, accurate assessment of cardiac function is feasible in most fetuses [4,8]. With the continuous development of AI algorithms and their implementation in platforms of different ultrasound devices, two softwares – MPI+TM (Samsung Healthcare, Gangwon-do, South Korea) and fetalHQ® (GE Healthcare, Chicago, IL, USA) – have recently been established to examine different cardiac functional parameters of the fetus. So far, only the learning curve for the acquisition of the MPI manually [67] and the benefits of its semiautomatically measurement in clinical routine for less experienced operators have been demonstrated [47]. The additional value of the 2D-STE for less experienced sonographers has yet to be demonstrated, although the significance in a scientific setting is undisputed.
FetalHQ® has shown excellent reproducibility for the analysis of morphometric as well as functional cardiac parameters [2,7,14,16]. Based on our data, we were able to demonstrate good to excellent agreement between both operators for the applicability of 2D-STE for comprehensive analysis of global and segmental cardiac deformation. Although, current studies have recently questioned the initially excellent reproducibility of segmental parameters in particular [2,5,68,69].
Despite all the advantages that 2D-STE can offer in combination with post-processing analysis using fetalHQ®, the technique mainly depends on the skill and experience of the operator and the quality of acquisition of an optimal clip of the 4-CV [14,25,70]. An experienced operator of prenatal ultrasound will certainly undergo a short learning curve to acquire and analyze short image-optimized cine-loop sequences of the fetal cardiac 4-CV by 2D-STE [69,71]. For less experienced operators with initially lack of adaptive skills in visualizing an optimal 4-CV of the fetal heart, a marked learning curve can be assumed. Germanakis et al. summarized that, although speckle tracking technique is algorithm-guided, offline image analysis is not user-independent. The operator must place or adjust the endocardial tracking lines. Therefore, a learning curve is essential prior successful implementation in clinical routine [5,10,15,25,71]. To take these aspects into account and to achieve the same level of standardization as in adult cardiology, DeVore et al. have provided a practical step-by-step approach [16].
The classification of the cardial functional parameters in the pathophysiological context and their subsequent interpretation is highly complex. Due to differences in ultrasound equipment, measurement techniques with non-standardized pre-settings, non-systematic image settings and processing, vendor-specific analysis software with different algorithms tracking speckle patterns and myocardial deformation as well as maternal and fetal characteristics of the study population [2,5,12,14], a wide variability in the quoted endoGLS reference values during pregnancy results to date. Since there is neither a mandatory standardization of the process nor of the semiautomatic algorithms, and there are many pitfalls to deal with, the cardiac function parameters described in the literature vary considerably [2,5,7,8,12,14,25]. It should also be noted that any normal reference ranges of myocardial deformation may need to be adjusted to the corresponding GA [14]. Patey et al. reported on intervendor discrepancies with significant clinical and research consequences. They demonstrated that fetal 2D-STE assessments are reliable when performed on the same ultrasound platform, while ultrasound machines and software from different vendors produce significantly different fetal 2D-STE parameter estimates. This should be considered when interpreting and comparing research data [72]. The review by Kühle and co-workers revealed that the mean value of all cited studies – excluding those with positive values – considered for the right ventricular GLS is reported as -19.36 % ± 4.55 pp, that of the left as -19.21 % ± 3.88 pp. The lowest cited value for the right ventricular GLS was -35.88 %, the highest -12.80 % and the lowest left ventricular GLS -26.01 %, the highest -12.30 % [25]. The mean endoGLS values of both operators in this paper are consistent with those referenced in the literature. It is widely recognized that fetal circulation is right-heart-dominant with higher myocardial shortening during contraction in the right ventricle [11,25,73,74]. The more negative the endoGLS values, the better the contractility of both ventricles [10,14]. It would therefore be expected that longitudinal strain values in the right ventricle are between 1.0 and 1.5 times higher (i.e. more negative) compared to the left ventricle due to the difference in myofiber orientation [40,75], although contradictory measurements can be found in literature concerning this as well [10]. As many other studies, we were unable to confirm this hypothesis based on our data (Table 3, Figure 2 and 3) [25]. Echocardiographic evaluation of ventricular function is challenging, especially assessment of the right ventricle function due to its complex anatomy [76]. On the one hand, anatomical structures like the moderator band can complicate the correct placement of the apical caliper for the operator, on the other hand prominent trabeculation can impair the automatic delineation of the endocardial border during a cardiac cycle by the software [5,25,75]. Such inaccuracies in the acquisition of right ventricular deformation parameters with suboptimal contrast between the endocardium and the respective ventricular chamber without precise wall tracking might result in underestimated endoGLS values [10,25]. Furthermore, 2D-STE is only based on 2D 4-CV, which neglects spatial aspects of cardiac anatomy of the right ventricle. This might apply to our data, with lower mean endoGLS value of the left compared to the right ventricle (Table 3, Figure 2 and 3). In addition, the agreement of the right ventricular endoGLS measurements between the two operators in our study is lower compared to left ventricular (Table 4, Figure 4). This assumption of lower agreement for right ventricular parameters was also reported by Huntley et al. [5]. Altogether, all measured functional parameters in our study – global as well as segmental – showed good to excellent agreement between both operators (Table 4, Figure 4 and S7 to S9). The lowest agreement, but still considered as good, was found for the FS. These results are in accordance with recent publications questioning the reproducibility of FS, which was previously reported as excellent [2,5,68]. It was also noticeable in our study that the absolute values of the FS of the individual segments frequently deviated from the percentages defined in the software fetalHQ®. The focus of this study was not the determination of further reference values or the verification of reproducibility of fetalHQ®, but rather the proof of agreement between two operators of different expertise. Nevertheless, this aspect should be mentioned. 2D-STE can be used throughout pregnancy, even as part of first trimester screening [25,77]. However, in clinical routine, we emphasize its best application mid-gestational and advanced GA, as other authors have also elaborated [5,10,14,25,28,29]. Our endoGLS values of both ventricles assigned to gestational age are consistent with the contradictory data in the literature, with slight decrease of right ventricular and almost constant left ventricular endoGLS values with increasing GA (Figure 2). Ohira et al. recently reported on similar findings with their underlying pathophysiological mechanisms [28].
In fact, 2D-STE should have better intra- and interobserver variability than other techniques used to evaluate cardiac function [16]. Apart from routinely acquired 2D B-mode images of the 4-CV, no further imaging is required. In contrast to Doppler-based methods, 2D-STE is supposed to be angle-independent due to the simple acquisition of the 2D image, regardless of the orientation of the heart to the ultrasound beam, and could therefore be performed in any fetal position [15,16,27]. It is controversially discussed that speckle tracking is less angle-dependent or even angle-independent [2,5,10,12,14,16,17,36,40,78]. Most likely – and we agree with this opinion – 2D-STE can be considered as less angle-dependent but cannot be performed in every fetal position with limited clip acquisition of the 4-CV. At least the position of the IVS should be considered when a cine-loop of a 4-CV is recorded. In clinical routine, it is almost impossible to record every fetal heart with the IVS in a perpendicular position, as recommended by deVore et al. [16]. In our study, the most common apex orientation for both beginner and expert was ‘apex oblique up’ (Table 2). Furthermore, due to the combination of small size of the fetal heart with high heart rate, an adequate frame rate to achieve a high temporal and spatial resolution with tracking of the speckles is intensively discussed [2,5,6,8,9,14,16,25]. 2D-STE requires a high resolution 4-CV, suitable for post-processing analysis, with high image acquisition frame rate, depending on fetal heart rate [2,6,15,16,25,78]. Specialized software with an automated sensitive and accurate border recognition can contribute to simplify and standardize the procedure of speckle tracking [6,11]. However, the quality of subsequent speckle tracking analysis is affected by the proper visualization of the 4-CV within the cine-loop with clear delineation of the endocardium [5,24,25]. Precise image acquisition and quality of the 4-CV can be impaired by maternal BMI, anterior placenta, oligohydramnios, fetal anatomy and position, resulting artifacts caused by maternal as well as fetal movements and, not least, from skills of the operator. Since the number of ultrasound frames per second is assumed to have a great impact on the reliability of 2D-STE, DeVore et al. recommend an optimal frame rate of >80 Hz. But there is currently no consensus about the optimal frame rate for strain imaging with recommendations for significantly higher frame rates [25,28,79,80]. On average, we achieved a frame rate of 76 Hz (beginner) and 73 Hz (expert) in our study collective, which represents the clinical routine (Table 2). Gireadă et al. exceeded the 80 Hz frame rate threshold in only 9%, we in 43% (beginner) and 7% (expert) [36]. When interpreting the results of a speckle tracking analysis and assessing the agreement of measurements between operators, angel of insonation (apex orientation) and frame rate must be kept in mind.
In our opinion, there are three pitfalls with significant impact on the accuracy, reliability and validity of the method as well as the agreement between the operators that must be considered when performing 2D-STE: Firstly, the artifact-free image acquisition of a clip of the 4-CV, secondly, the orientation on the heart with recognition of the left and right side, as well as subsequent identification of an optimal cardiac cycle, thirdly, the manual placement of the three calipers with any required adjustment of the endocardial tracking lines. Considering these aspects, a short learning curve is also evident for less experienced operators. We emphasize that 2D-STE cannot be applied in all patients. In contrast to RV-Mod-MPI, it requires a good pre-selection for success. Compared to other methods for the determination of fetal cardiac function, such as the RV-Mod-MPI, the time factor has a significant negative impact on the 2D-STE analysis. The evaluation of the complex measurements is more time-consuming and requires further expertise. The longer the cine-loop, the longer the import into the software for further analysis. Therefore, 2D-STE analysis is less suitable for performing ‘on-the-fly’ in the busy, tightly scheduled clinical routine, especially for less experienced operators. Despite today’s semiautomatic 2D-STE measurement using fetalHQ® and after more than ten years since Van Mieghem et al. concluded that speckle tracking can be performed by a sonographer with limited experience in fetal cardiology on any prenatal ultrasound exam, we still doubt this statement even in 2023 [40]. We agree with the opinion of Germanakis et al. who underline the importance of understanding the basic principles of strain and the resulting limitations of 2D-STE before adopting it into routine clinical practice [15].
An approach to optimize the performance of the 2D-STE method more efficient with shorter time for acquisition was recently published. The quiver technique enables the operator to identify the endocardial borders more precisely and efficiently. It facilitates the identification of the septal and lateral atrioventricular valve annulus by cine-looping two frames before and after the selected end-systolic and -diastolic frame [24]. In our opinion, it will be of great importance in the future to eliminate human manual adjustment for improving reproducibility and increasing efficiency in daily clinical routine by automatically identifying the most optimal cardiac cycle including the accurate selection of endocardial borders by AI algorithms. This was already demanded by Huntley et al. [5].
Although speckles move in three dimensions, most methods of speckle tracking are based on post-processing analysis of only one acquired plane of a 2D-clip of the fetal heart [9,14]. Speckles beyond the insonated plane will not be analyzed [6]. Cardiac mechanics is complex due to special geometry with myocardial three-directional motion including longitudinal, radial, and circumferential contraction [4,8,25,73]. The helical structure of the ventricular myocardium with its three-dimensional (3D) deformation requires to detect speckles in all dimensions [81]. In addition, 2D-STE excludes the right ventricular outflow tract [11]. Theoretically, holistic, spatial concepts might offer an important contribution in measuring more realistic volumetric parameters. It therefore seems natural to combine STE technique with spatial methods such as 3D- or four-dimensional (4D)-spatio temporal image correlation (STIC) for a more precise and realistic measurement and to overcome B-mode imaging limitations of 2D-STE [2,9,70,82].
Once a suitable 2D-clip of the 4-CV was acquired with knowledge of the pitfalls discussed above, there was in many cases nothing to prevent subsequent offline analysis using fetalHQ® and resulted in high agreement between beginner and expert. Nevertheless, some limitations of the present study must be mentioned. Only 136 unselected, normal, singleton, second- and third-trimester fetuses without CHD in healthy women, focusing on the 23th week of gestation on average, were examined. The period in which these clips for 2D-STE analysis were recorded appears to be quite long. The recording of the clips with subsequent 2D-STE analysis was performed by only two operators. However, these two (beginner and expert) were representative as holders of DEGUM level 1 (German Society for Ultrasound in Medicine), the minimum standard for a sonographer, and DEGUM level 3 for gynecology and obstetrics. The former certifies familiarity with the basics of ultrasound diagnostics, the latter characterizes a proven expert far beyond basic knowledge. Representative for the base (1-8), the middle (9-16) and the apex (17-24) of both ventricles, in this study, according to Nogué et al., segments 1, 9 and 17 were selected for comparison of measurements of both operators [2]. It would be possible to select other segments, for example 1, 12 and 24, as in the study of Zhu et al. [11]. Differences in the agreement of both operators in the measurement of segments not examined are feasible, which applies to segmental, but not to global parameters. In this study, we allowed orientations of the IVS that were considered suboptimal for 2D-STE analysis to reflect daily clinical routine. According to our results, this did almost not impair the good to excellent agreement between the two operators.