Discussion
To the best of our knowledge, this is the first study to describe artefact recognition using a qualitative assessment of strain curves. The main findings were as follows: (1) The presence of curve artefacts matched the presence of noise or other 2D image artefacts in 88% of participants; (2) in the presence of curve artefacts, strain values were systematically reduced, and they were increased by foreshortened long-axis projections; (3) discarding strain-curve artefacts rendered segmental strains with higher mean values and lower variability, whereas global strain was not affected and (4) this study is the first to describes the artefact-free normal ranges for layer-specific segmental strain in a large randomly selected population with normal heart structure and function.
Methodological considerations in curve-artefact detection
Artefacts caused by errors in automatic tracking generally result in low peak strain and low SR results, leading to misinterpretation as false pathological results. The concept of curve artefacts was built on the following considerations:
‘Diastolic mismatch’
Normal and pathological strain curves display distinct patterns, where systolic deformation patterns can change due to pathologies causing reduced or delayed regional or global systolic shortening. In contrast to the systolic and early diastolic strains, the late diastolic strain is expected to display uniform strain curves with parallel stretching patterns. At the time of maximal postsystolic shortening (when present), all segments reached a relaxed state at approximately equal segment lengths. After this time point, all segments stretch in parallel at the end of the early filling phase through diastasis of the ventricle and atrial contraction until reaching the initial segment length at end-diastole. As shown in figure 2-I, curves with deviating (often reduced) deformations of a single segment in late diastole can only represent an artefact. Thus, we called this type of artefact ‘diastolic mismatch,’ describing strong positive deviation of the late diastolic curve from the diastolic strain-curve of other segments.
‘Blunted curves’
Another curve-artefact type that was discovered could be caused by insufficient tracking at the beginning and end of the cardiac cycle. This type manifests as zero or positive initial strain, often mimicking the early onset of segmental stretching and the late onset of shortening. In the presence of true regional pathology, the strain of the segment with reduced contractility is significantly lower than the peak strain of the neighbouring segments, and the artefact-free curve displays post systolic shortening (PSS). Delayed segmental shortening, early onset of segmental stretching before end-systole and missing PSS were the typical characteristics of a ‘blunted’ strain-curve.
‘Floating curves’
The third type of curve-artefact is associated with erroneous myocardial tracking, particularly when the myocardium is not being followed throughout the cardiac cycle. In this case, the strain curves do not follow a physiological pattern and show either oscillations or several positive and negative peaks without the corresponding peak strains of the neighbouring segments. In dyssynchrony or regionally reduced myocardial function, positive and negative peaks of the opposite segments can be displayed simultaneously. However, physiologically, the time points of positive and negative peaks in systole are simultaneous, and the peaks in diastole do not diverge after the time point of PSS. We defined ‘floating curves’ as present when several diverging positive and negative peaks were observed, and when the time points of onset or peak deviated from the onsets and peaks of all other segmental curves.
Influence of artefacts on segmental S/SR
Segmental curve artefacts appeared to be highly correlated with either the presence of artefacts in 2D images, such as noise and reverberations; thus, in most cases, curve and noise artefacts were observed together. This strengthens our hypothesis that a high proportion of 2D image artefacts produce curve phenomena. Most curve artefacts without 2D artefacts were from basolateral segments, where tracking is problematic because of lower lateral resolution. Segmental values with curve artefacts were significantly reduced for both strain and SR in all layers. According to our results, artefacts had a high impact on the segmental S/SR values. Thus, segmental curve artefacts need to be handled before drawing clinical conclusions from strain patterns on the segmental bullseye plot.
Apical foreshortening has already been described as a factor for the overestimation of strain values,13 especially in the apical endocardium, which is consistent with the present findings. However, this overestimation probably does not apply to highly dysfunctional segments. The effect of foreshortening and curve-artefacts on clinical needs to be investigated in future studies.
Effect of artefacts on global S/SR
Discarding artefacts from the average strain calculation reduced the ES strain values by <0.5%. Average GLS and artefact-free strain differed by the same amount. Unexpectedly, subendocardial GLS was also lower than the segmental average with artefacts. This may be explained by a software algorithm that reduces the percentage of apical strain contribution to GLS, which affects endocardial GLS, with the highest apical to basal gradients. The difference between the segmental average of the LV and global values was even more noticeable for SR values. The lower global longitudinal (GL) SR values than the average of segmental peaks are expected because the non-simultaneous time points of segmental peaks do not sum up to one blunted peak of the GL SR curve. Thus, when using SR-E as a clinical parameter, the average of peak values at different time points should be used. Although significant, the detection of segmental artefacts had little effect on GLS values and, accordingly, should not be considered when measuring GLS. To date, several studies on larger populations have suggested normal ranges for GLS or SR.1 12 14–18 Accordingly, the normal ranges were integrated into the 2016 American Society of Echocardiography/European Association of Cardiovascular Imaging chamber quantification recommendations.12 19 Our GLS results are in concordance with results of a recent meta-analysis,18 where the mean normal values of mid-myocardial GLS varied from −15.9% to −22.1% (mean: 19.7%).
Global S/SR based on age and sex
Studies regarding the effects of age on LV longitudinal S/SR-A are limited. Therefore, the effect of age on LV myocardial deformation remains controversial. In this study, strain values were not significantly different between the age groups of 40 and 69 years, which is consistent with the findings of Nagata et al.11 However, other studies reported a significant effect of ageing on systolic longitudinal myocardial velocities but not on the longitudinal S/SR,12 14 20 which might be due to decreased systolic deformation from higher arterial stiffness with increasing vascular resistance and afterload. Accordingly, our results demonstrated a gradual decrease in SR-E and a gradual increase in SR-A with increasing age.
The lower global and segmental S/SR values in men than in women are consistent with those of previous studies.15 16 21–23 Many features, such as lower systolic blood pressure; lower wall stress and afterload; and smaller body surface area, ventricular size and LV mass in women than in men, have favourable effects on LV systolic functional parameters. These composite factors seem to be a physiologically plausible reason for higher S/SR parameters in women than in men. Due to the significant differences in these factors between sex, we described the normal ranges separately for men and women.
Normal reference values
Only a small number of previous studies have yielded information regarding population-based normal segmental S/SR values.19 22 24 25 Since segmental S/SR is highly affected by artefacts, the presented segmental systolic and diastolic layer S/SR ranges might serve well as a reference for segmental normalcy. Analysing strain by layers, a consistent gradient for segmental strains from the epi—towards the endocardial strain was found, which might explain the curvature of the ventricle, where the inner layer shortens as a sum of longitudinal shortening and radial displacement of the curved line towards the ventricular cavity.
LV segmental S/SR gradients with gradually increasing values from the basal to apical segments was also observed. The basal-apical gradients appear to be a physiological phenomenon, as they persisted after apical foreshortening, discarding artefacts and the exclusion of the hypertensive population.
Limitations
This study has several limitations that must be acknowledged. First, the data in this study were obtained from a single reader and the ultrasound systems and software were provided by a single vendor. Therefore, strain variability may be substantially lower than with a multireader approach, which can be overcome with improvements in automated border detection. Second, the ages of the participants ranged from 40 to 69 years, indicating that the results cannot be extrapolated to older or younger participants. Third, although this study used data from two population studies conducted in different countries, the majority of participants were European Caucasians, indicating that the results should be extrapolated to other ethnic groups with caution. Fourth, we cannot exclude the possibility that some participants in the normal study population had undiagnosed coronary artery or other cardiac diseases. However, a possible effect of these conditions on the heart structure is unlikely to be significant in the context of our study. Fifth, the artefact-based approach presented in our study requires further evaluation. Sixth, the description of the curve artefacts is a subjective approach, which needs comparison to independent noise detection. In the near future, the authors intend to correlate automatically detected 2D grey-scale imaging artefacts with curve-artefacts for further validation. Finally, the Russian population did not include APLAX views in their echocardiograms, and strain-analysis was only performed on 21% of the APLAX views of the Tromsø population. However, APLAX segments in the Tromsø7 population showed the same range of values as the 4-CH view, and the inclusion or exclusion of APLAX segments did not change the results of the segmental values in samples from Tromsø7.
Clinical implications
Our results showed that the presence of curve artefacts lowers segmental systolic S/SR values, regardless of segment localisation. Because of their similarity to pathology, these artefacts must be identified to avoid misinterpretation of the biased low values as evidence of regionally reduced heart function. In contrast, foreshortening artefacts systematically increased S/SR values, which did not change the interpretation of the presence or absence of pathologies during clinical examination. Pathologically reduced strain curves have a curve with reduced segmental peak strain is followed by PSS, and after the PSS peak, all segments display parallel stretching at the end of the early diastolic and atrial filling phases.
In this study, we described an approach for defining strain-curve artefacts using subjective visual assessment. If this subjective approach is robust as compared with objective 2D imaging artefact detection, curve artefact algorithms can potentially be integrated into automated reading programmes. Timely manual or automated detection of artificial curve shapes facilitates the correct interpretation of strain results without repeated assessment of tracking. Finally, during real-time strain recordings, curve artefact detection might be performed to provide immediate feedback on imaging quality. Other algorithms can be developed taking both, curve artefacts, noise, reverberations, or missing segments into account to help reconstruct discarded segments on the base of the ventricular shape, size and strain curves of artefact-free segments.