The effects of extracellular contrast agent (Gadobutrol) on the precision and reproducibility of cardiovascular magnetic resonance feature tracking

Myocardial strain imaging allows for objective quantification of cardiac contractility and is increasingly being recognized as a sensitive tool for analysis and diagnosis of various myocardial disorders [1, 2]. Furthermore strain analysis provides important prognostic information useful for predicting outcome of various ischemic and non-ischemic cardiomyopathies [3‐5]. Although myocardial strain values were first derived from cardiovascular magnetic resonance (CMR) tagging (TAG) [6], echocardiography has been established as the clinical standard for the assessment of strain, as it is more widely available [7]. Feature tracking (FT) is a CMR post processing tool which allows for functional wall motion analysis in CMR cine steady-state free precession (SSFP) images and therefore opens the possibility of strain assessment in a standard clinical setting. The value of FT, but also its limitations, for assessment of regional and global systolic and diastolic strain have been demonstrated [8‐10]. The CMR FT algorithm is based on an echocardiographic post processing tool, where strain assessment was first achieved without tagging. Voxels from the endocardial border are ascribed a certain number of features (e.g. brightness and dyshomogeneities of the tissue with respect to a 256-level gray scale) and then tracked from frame to frame [9], which enables a deduction of information about mechanical deformation. Extracellular contrast agents decrease the blood-myocardium contrast in SSFP images and thus might affect FT results. However in routine CMR, SSFP cine scans are frequently acquired after contrast agent injection to save time [11]. Thus the aim of this study was to investigate the effects of an extracellular contrast agent on the precision and reproducibility of the feature tracking algorithm.

The study protocol was completed by all participants. Table 1 summarizes pre- and post-CA results. Figure 1 demonstrates an example of typical Ɛcc curves derived by FT in a subject pre- and post-CA application. Mean HR (67.5 ± 13.6 BPM vs. 69.2 ± 13.9 BPM), mean BP ( 136.1 ± 18.2/72 ± 9 mm Hg vs. 138 ± 20.1/72.9 ± 8.2 mm Hg) as well as mean LVEF( 61 ± 9 % vs. 60 ± 9 %) did not differ significantly pre- and post-CA application. The blood-myocardium contrast quotient significantly decreased post CA application (4.03 ± 0.6 vs. 2.16 ± 0.3 p < 0.0001). No significant correlation was found between the degree of contrast reduction and the degree of strain reduction following CA application (r = 0.32). Assessment of endocardial shortening revealed no significant difference between pre-CA and post-CA images (27.97 ± 8.02 % vs. 27.63 ± 8.2 %).

10–15 min after CA application FT derived midventricular PSCS (-24.8 ± 6.4 % vs. -20.4 ± 6.3 %), apical PSCS (-28.67 ± 6.5 % vs. -24.06 ± 8.5 %), basal PSCS (-24.42 % ± 6.5 vs. -20.68 ± 7.1 %), PSLS (-19.57 ± 3.3 % vs. -17.24 ± 4.1 %) and midventricular EPSCS (-9.84 ± 3.4 % vs. -8.13 ± 3.4 %) were significantly reduced in comparison to baseline strain analysis. Furthermore, midventricular PSCSR(-1.52 ± 0.4 vs. -1.28 ± 0.5) and PDCSR(1.4 ± 0.5 vs. 1.05 ± 0.5) were also significantly reduced after CA application (Tables 2 and 3). Correlation for pre- and post-CA derived midventricular PSCS was r = 0.81 (Fig. 2).

Bland Altman analysis of pre-CA intra-observer reproducibility yielded a better agreement (0.21 ± 1.5 with limits of agreement between -2.8 and 3.2) than post-CA intra-observer reproducibility (-0.11 ± 2.5 with limits of agreement between- -5.1 to 4.9). Correspondingly interobserver variability (0.64 ± 1.75 with limits of agreement between -2.8 and 4.1 vs. 4.93 ± 3.7 s with limits of agreement between- -2.4 to 12.2) was superior pre-CA (Fig. 3). Correlation coefficients for pre-CA derived strain were excellent (Intraobserver: r = 0.97) (Interobserver: r = 0.95) and superior in comparison to post-CA derived strain (Intraobserver: r = 0.91) (Interobserver: r = 0.91). The intra- and interobserver coefficients of variation were 4.5 and 5.4 %, for pre-CA strain assessment, 22.8 % and 20.9 % for post-CA strain assessment respectively.

The current gold standard for myocardial strain imaging in CMR remains tagging [13‐15]. Several techniques such as displacement encoding with stimulated echoes(DENSE) [16], complementary SPAMM (CSPAMM) [17], harmonic phase (HARP) [18] and strain encoding (SENC) [19] have been developed to expedite and optimize the tagging derived strain analysis. These techniques are based on the creation of a pattern of magnetization saturation, from which movement throughout the RR cycle can then be quantified [20]. In comparison, the FT algorithm does not deduct strain information from a created strain pattern. The advantage of FT is that additional complex imaging and postprocessing is no longer necessary for strain analysis. However, the FT algorithm is highly dependent on a high contrast between blood and myocardium [15] and in comparison to tagging techniques FT has been found to be less robust [8, 10]. For comparison purposes, the identical sequence parameters, including identical flip angles, were used for pre- and post-CA imaging. In order to have identical sequences, the flip angle was not adapted for post CA imaging. In this study, reducing contrast by acquiring SSFP images after CA injection delivered significantly different strain-results as well as inferior reproducibility in comparison to strain analysis based on pre-CA images. Circumferential apical-, midventricular- and basal- PSCS, PSLS as well as midventricular systolic and diastolic strain rates were significantly reduced when derived from post-CA SSFP images, while neither heart rate nor blood pressure varied. Post CA midventricular EPSCS showed less severe reduction following CA application in comparison to midventricular PSCS, a possible explanation may be that the epicardial fat-myocardial interface is less affected by CA administration than the endocardium blood pool interface.

The CA Gadobutrol, is not known to cause changes in heart rate, blood pressure or to affect the cardiac conduction system [21]. An average of 15.9 ± 3 ml of CA followed by 25 ml of saline flush were injected per patient for late enhancement imaging. This small volume is not likely to have affected preload or contractility, furthermore increased preload will typically increase cardiac contractility and strain, contrary to the reduction that was found. Although to date the different effects of the various gadolinium based contrast agents on the blood pool-myocardium contrast have not been evaluated, it is to be expected that in case of application of gadopentetate dimeglumine (0.5 M) the effect on contrast will be similar or even worse compared to those of Gadobutrol (1 M), especially as gadopentetate dimeglumine exhibits a weak, transient interaction with serum albumin possibly decreasing the blood pool-myocardium contrast even more [22]. Previous studies assessing FT have found mixed results for reproducibility, with several studies reporting considerable intra- and interobserver as well as interstudy variability especially for regional, as well as apical and basal derived FT derived strain [8, 10, 23, 24], underlining its restricted robustness in comparison to tissue tagging. In the current study pre-CA reproducibility was comparable to results from previous studies [9, 23], while post-CA derived strain reproducibility showed inferior results. Post-CA FT derived strain was reduced in comparison to pre-CA FT derived strain in almost all cases, with corresponding results for intra- and interobserver repeated analysis indicating that contrast reduction leads to an underestimation of strain. Increased intra- and interobserver variability as well as an increased coefficient of variation post CA application however indicate that the error caused by contrast reduction is not truly systematic, as a systematic error should not influence reproducibility. Furthermore the degree of strain reduction did not correlate with the degree of contrast reduction further indicating that CA application leads to an unsystematic underestimation of FT derived strain. The FT algorithm calculates strain based on the detection and tracking of contrasts and dyshomgeneities of a cluster of voxels from frame to frame throughout the RR-cycle. When the initially characterized cluster cannot be re-detected in the following phase, apparently a different cluster of voxels is tracked leading to a false result. As the FT software does not provide information on tracking quality, it is essential to employ FT only in cases with good image quality and myocardium-blood contrast. It is important to note that several cardiac diseases, which can potentially be detected by CMR strain analysis, are often characterized by only mild systolic or diastolic strain impairment [25, 26]. Thus, the highest possible precision is demanded from clinically employed strain assessment tools. A higher variability of strain measurements, as can be found when deducting strain from post CA images, means that mild systolic or diastolic dysfunction could either be overlooked or even falsely diagnosed.

The following limitations apply to this study: Both post-contrast and pre-contrast SSFP were acquired with the same imaging parameters. Thoughtful adaptation of pulse sequence parameters (e.g.flip angle) may increase blood myocardium contrast post CA injection. However, in this study pulse sequences were chosen to reflect the clinical routine. As discussed above endocardial shortening and ejection fraction were used as surrogate markers to rule out an effect of Gadobutrol on contractility, however, the gold standard for strain imaging -namely myocardial tissue tagging- was not employed.

FT has become an established tool for rapid SSFP based strain analysis, however it is dependent on a high contrast to allow for precise voxel tracking from frame to frame. Assessing strain from post CA SSFP cine images, which leads to a reduction in the myocardium-blood contrast, leads to an unsystematic underestimation of strain with increased intra- and interobserver variability. The strain based evaluation of many ischemic- and non-ischemic- cardiomyopathies demands the highest possible precision and robustness, therefore FT based strain analysis should only be performed on pre CA SSFP cine images.