For more than a decade, myocardial strain has been in research use in the evaluation of cardiac ischemia and infarction and myocardial diseases and in the assessment of synchrony. In these settings, strain measurement potentially yields information on tissue characterization as well as the inotropic state of the myocardium. However, this information is based on the underlying concept that myocardial strain is a physical property of myocardium that describes its deformation under an applied force. What is puzzling, therefore, is why there is a literature on the variability of a parameter that represents a fundamental physical property of tissue. In this issue of JASE , two reports build on an existing literature ( Table 1 ) regarding machine-specific aspects of two-dimensional strain that seem likely to be duplicated in the era of three-dimensional strain.

Nelson et al. examined global longitudinal strain in a convenience sample of 100 prospectively selected patients (mean age, 60 ± 17 years; 46 men) with a variety of conditions, most commonly chest pain and coronary artery disease. In contrast to previous studies in which discrepancies may have arisen from the acquisition of two sets of images, these patients had echocardiographic images acquired using one system (Vivid 7; GE Medical Systems, Milwaukee, WI), and strain was measured offline using two types of vendor-independent software (Echo Insight, Epsilon, Ann Arbor, MI; Image Arena, TomTec Imaging Systems, Unterschlessheim, Germany). The results of this study contrast to previous work showing that vendor-independent software produces more homogeneous results than vendor-specific strain packages. The average strain measurements were significantly different (12.99 ± 2.38% vs 16.87 ± 2.84%), and the agreement between the software packages was only moderate, with an intraclass correlation coefficient of 0.43. Of many potential differences, including individual (and therefore potentially nonidentical) tracing within each software package, selection of cardiac cycles, and differences in smoothing, the investigators particularly focused on the evaluation of endocardial strain with one system and both endocardial and epicardial strain with the other (which would explain the lower average strain, as epicardial strain is less than endocardial strain). Indeed, after correcting the measurement of strain to the endocardium using both techniques, the mean difference between the measurements was much smaller, and the intraclass correlation coefficient was 0.70. The use of Digital Imaging and Communications in Medicine (DICOM) standard data in this study, providing images at a reduced frame rate, may also have led to an underestimation of peak strain (see below).

In their study, Risum et al. took a different approach. These investigators studied 30 patients (15 with and 15 without cardiac disease) with echocardiographic machines from two different vendors (Vivid 9, GE Medical Systems; iE33, Philips Medical Systems, Andover, MA) at approximately 60 frames/sec for analysis using vendor-dependent software and then analyzed the DICOM images (30 frames/sec) using vendor-independent software (Cardiac Performance Analysis; TomTec Imaging Systems). The investigators found that strain values obtained using this vendor-independent software were comparable with those obtained using vendor-specific software for longitudinal strain, independent of ultrasound machine. The lower frame rate of DICOM images provided acceptable agreement between the types of software, but with wider limits of agreement. However, poor agreement was observed for circumferential, radial, and transverse strain. The coefficient of variation between the different strain values was smallest with longitudinal strain (5.5%–8.7%) but unacceptably large for circumferential strain (10.7%–20.8%) and largest for radial and transverse strain (15.3%–33.4%).

The variations documented in these reports are one of many reasons that strain analysis has failed to reach the routine clinical arena. There are fundamentally three sources of variation: acquisition, postprocessing, and the hemodynamic status of the patient. Frame rate has traditionally been considered the most important aspect of acquisition, with the optimal frame rate for the calculation of speckle strain being cited as approximately 60 frames/sec. Underestimation of measurements from images stored in DICOM format (30 frames/sec) might be anticipated with parameters that require high temporal resolution, such as the assessment of strain rate and characterization of diastolic properties. The study reported by Risum et al. suggests that the measurement of peak systolic strain is not compromised by undersampling. A less widely appreciated facet is that there is a trade-off between temporal and spatial resolution, so that the acquisition of data at higher frame rates not only poses more challenges for tracking speckles but also potentially may lead to some compromise in spatial resolution because of sacrifices in line density. The use of contrast opacification is currently incompatible with the accurate measurement of strain.

Speckle tracking is not fundamentally a different modality of imaging but rather a modality of postprocessing, and this is the second phenomenon that is potentially very different among vendors and may influence strain. This processing involves algorithms that anticipate the contour of the ventricle as well as its direction of contraction. These algorithms vary among vendors. For example, tracking (which is initially performed over 2-cm segments and refined down to 5-pixel bands) with Velocity Vector Imaging (Siemens Medical Solutions USA, Inc., Mountain View, CA) is based on an assumption of geometric shape, on the basis of following reference points, which inform snake contours of the left ventricular border. In contrast, acoustic markers suitable for tracking with Automated Function Imaging (GE Medical Systems) are selected within a region of interest, but again, the assumed shape of the ventricle is important because data fitting is weighted according to correlations between the original markers and the tracked markers. For this method, regional Lagrangian strain is calculated at about 50 points distributed at 3-mm intervals along the left ventricular circumference. A variety of other technical variables influence the measurement of strain, including whether this is evaluated in the endocardium, epicardium, or both; smoothing of the data (which varies among machines and can now even be modified by the user, adding an additional variable to an already complicated system); calculation of natural versus Lagrangian strain; and average regional peaks versus taking the average of peak strain.

The third source of variation is physiology. As an ejection phase index, strain is highly sensitive to loading conditions, particularly afterload, and in studies in which two sets of images are obtained, it is important to know that they were obtained under the same circumstances. The impact of hemodynamics on “adaptive scaling” that defines the clinical relevance of intervendor variation is important; in two patients studied on two occasions, a 10 mm Hg difference in systolic blood pressure may far exceed the importance of intervendor variation of global longitudinal strain.

The clinical performance of a quantitative parameter is dependent on two critical aspects: accuracy and precision. Much effort has been directed toward validating echocardiographic assessment against “true” measures of strain, such as microcrystals and magnetic resonance. Although the overall correspondence between these measures has been good, examination of the scatter between these measurements explains how two validated tests may differ. Moreover, comparisons of these validation studies are difficult, because variations may be caused by differences in the reference standard just as much as problems with strain. For example, cardiac magnetic resonance comparisons are influenced by the relatively low temporal resolution (15–20 frames/sec) of cardiac magnetic resonance, differences in regional measurements may be due to problems with coregistration of images, and the tests are inevitably not performed at the same time. Problems with test-retest variation are every bit as important as accuracy; especially in the follow-up setting, this variation also influences the “adaptive scaling” that defines how much intervendor variation is important.

The variations documented in these and previous studies have important implications for the clinical application of strain. First, the study by Risum et al. confirms that DICOM images (and potentially archival images) can be measured reliably. This is an important observation, because in a multivendor laboratory, vendor-independent software is a viable means of reducing heterogeneity among vendor-specific software packages. The proviso needs to be added that it is extremely important to consider exactly what is being measured, because low–frame rate measurement of strain is potentially a problem in a number of circumstances, including the assessment of timing, precise measurements of magnitude and other high-frequency events such as maximum strain rate, and the details of the strain waveform. Although it may be possible to obtain global longitudinal strain in archived images, as used in the Multicenter Automatic Defibrillator Implantation Trial–Cardiac Resynchronization Therapy, this does not mean that all physiologic questions can be addressed from archival images. Second, global longitudinal strain is the most robust of the strain parameters, and variation may be less among machines than for the other measurements of strain. Indeed, messages about vendor-based variation may have inappropriately labeled all types of strain as being variable, when in fact changes in systolic blood pressure may very well have a greater impact on global longitudinal strain than does choice of equipment. Nonetheless, in patients who are undergoing serial evaluation, for example, those undergoing chemotherapy, use of the same platform is preferable if serial studies are planned.

The assessment of strain has much to offer to clinicians as a marker of subclinical disease, among other situations. For this reason, the assessment of strain must move away from the laboratories of expert researchers into general use. However, the use of speckle strain, although deceptively simple, has a number of important drawbacks. Hopefully, the pitfall of equipment variability will not be an enduring problem. The American Society of Echocardiography and the European Association of Echocardiography have set up an expert group, comprising interested researchers and industry members, to seek concordance on the details of what is measured. The development of Doppler and color Doppler is a precedent for such a process, which is achievable. The initial phase, which involves the development of phantoms that will at least allow measurement of the magnitude of the problem, has progressed well, and further progress is awaited.