Background
The aim of this study was to determine the feasibility, accuracy, and reproducibility of a novel tracking-based echocardiographic ejection fraction (EF) assessment method in comparison with traditional methods based on magnetic resonance imaging and echocardiography.
Methods
In a prospective assessment, apical echocardiographic grayscale image loops from 81 patients were read in random order by four experienced readers, blinded to any data of the cases. In three separate sessions, EFs were estimated using biplane tracking-based assessment and according to the modified Simpson’s rule, as well as by visual interpretation in three apical views. Data were compared with a reference EF derived from echocardiography and magnetic resonance imaging.
Results
On average, no significant difference was found between EF estimates of the different methods. Tracking-based EF assessments were possible in 90% of the patients. Tracking-based EF assessments showed slightly higher deviations from the reference EF than the modified Simpson’s rule, while interobserver and intraobserver variability of tracking-based assessment were significantly better. Visual interpretation allowed the fastest EF assessment. Tracking-based EF assessment was approximately twice as fast as the modified Simpson’s rule.
Conclusions
Tracking-based EF assessment is feasible, has lower interobserver and intraobserver variability, and is faster than traditional echocardiographic EF quantification. Its minimal demand on user interaction makes it a favorable alternative to traditional echocardiographic approaches, with a particular clinical advantage when reliable follow-up measurements are needed.
Left ventricular (LV) ejection fraction (EF) is a key parameter in clinical cardiology, guiding the diagnosis, management, and risk stratification of patients. Several imaging modalities, including cine ventriculography, radionuclide ventriculography, echocardiography, and magnetic resonance imaging (MRI), have been used for EF assessment. MRI is often considered a clinical gold standard, with relatively good reproducibility and low interobserver variability, although advanced echocardiography has been shown to deliver comparable results. Furthermore, the feasibility of MRI is limited because of its time demands, difficulty in arrhythmic patients or patients with metallic implants, and its limited availability and high costs.
In routine clinical use, echocardiography has become the most widely used modality for EF assessment, being harmless, time saving, portable, and relatively inexpensive. Using the traditional Simpson’s method, the endocardial border is traced in one or two apical image planes, and the ventricular volume is approximated by a stack of discs fitted in these contours. If repeated in end-systole and end-diastole, EF can be calculated. Although simple and straightforward, this approach to EF assessment has a certain interobserver and intraobserver variability. Therefore, echocardiographic EF assessment depends not only on image quality (which can be in part compensated for by contrast use ) but particularly, as with any other technique, on the training and expertise of the operator.
In routine clinical practice, an experienced user tends to compare the results of Simpson’s approach with the visual impression and to repeat the measurement if not satisfied with its result. As a consequence, an experienced operator may decide on a direct visual assessment of LV EF to save time. Furthermore, this visual EF assessment appears more feasible in difficult-to-scan patients or atypical scan planes. Although visual EF estimation is valid to a certain extent, it depends even more on training and expertise.
One can conclude from the above that a method that requires only limited user interaction would help improve the robustness of echocardiographic EF assessments in the clinical routine. To be widely accepted, such a method not only needs to deliver accurate and reproducible results but should also be feasible and at least as fast as traditional methods.
Modern image analysis algorithms, combined with speckle-tracking technology have recently been suggested for this purpose. The algorithms not only rely on pure border detection but combine models of LV shape and motion with image-based two-dimensional motion estimation of the myocardium. The aim of this study was to test the accuracy, reproducibility, and clinical feasibility of such a new tool.
Methods
Population
We used prospective data from 81 hospitalized routine patients in whom both echocardiography and MRI were performed within 48 hours and who represented a wide range of LV global and regional functional abnormalities. Our study group comprised 75 patients with different locations, extents, and severities of ischemic functional impairment of the left ventricle, five patients with dilated cardiomyopathy, and one with amyloidosis. All patients were in sinus rhythm and in stable hemodynamic conditions. Medications remained unchanged during the study. The study protocol was approved by the ethics committee of our institution, and all patients provided informed consented before inclusion.
Echocardiography and Echocardiographic Data Analysis
Protocol
Patients underwent routine echocardiography in left supine position, including the acquisition of apical four-chamber, three-chamber, and two-chamber views in regular grayscale mode, at a frame rate of 60 to 80 frames/sec. From all views, three cardiac cycles were digitally stored for further offline analysis.
Image loops used in the study were preselected by the study coordinator, who also anonymized and randomized the cases. One experienced sonographer and three experienced cardiologists, blinded to any patient information, read the data using a Vivid Q portable ultrasound machine with a tracking-based EF assessment option (AutoEF; GE Vingmed Ultrasound AS, Horten, Norway).
Identical image loops from all cases were seen three times in changing order. First, LV EF was assessed by visual interpretation of all three apical views. Readers were asked to provide integers of EF values, without further constraints. Second, a biplane EF estimation was done using the modified Simpson’s rule. To avoid bias from visual impression, it was not allowed to run the loops or to repeat the Simpson’s rule EF measurement. Third, the AutoEF algorithm on the ultrasound machine was applied to the apical four-chamber and two-chamber views. For all three readings, the time from having the first image on the screen to writing down the EF in the protocol sheet was documented. Results from quantitative EF assessments were rounded to the nearest integer before being entered in the database.
Tracking-Based EF Assessment
The AutoEF function is an extension of a speckle-tracking–based function analysis package. In short, the endocardium at both sides of the mitral ring and at the apex needs to be indicated by the operator with a mouse click both in the four-chamber and two-chamber views. The software automatically detects the endocardial contour with the option of manual correction. If no intervention is made, tracking of the endocardial contour throughout the entire cardiac cycle starts automatically after a set time. The highest and lowest cavity volume estimates during the cardiac cycle are considered end-diastolic and end-systolic and are displayed together with the calculated EF and other derived parameters ( Figure 1 ).
Influence of Image Quality on EF Estimates
Image quality was defined depending on the segmental visibility of the endocardium, which was classified as 1 (good), 2 (fair), or 3 (poor). Segmental scores were then averaged per patient and compared with the absolute difference between the EF estimate and the EF reference.
MRI and Data Analysis
Patients underwent a standard cardiac MRI protocol in a 1.5-T scanner (Intera; Phillips Medical Systems, Best, The Netherlands). Steady-state free precession cine MRI with retrospective triggering at 30 frames/RR interval was performed to obtain short-axis slices at a distance and thickness of 8 mm.
Data sets were transferred to a workstation for offline analysis (Phillips Medical Systems) and analyzed by one radiologist and one cardiologist, both experienced in cardiac MRI. LV global function was analyzed by contouring the short-axis cine magnetic resonance images at end-diastole and end-systole. Papillary muscles and trabeculations were included in the blood volume.
Definition of the EF Reference
The reference LV EF was calculated as the average of the two MRI readings and the four visual echocardiographic EF assessments. In all cases in which the means per modality differed by >10%, a consensus between readers was sought after unblinded repeat analysis of both MRI and echocardiographic data.
Statistical Analysis
Continuous parameters are expressed as mean ± SD. Grouped data were tested for normal (Gaussian) distribution and equality of SD (Kolmogorov-Smirnov) and compared using two-tailed paired t tests. For more than two groups, repeated-measures analysis of variance with Bonferroni’s correction in post hoc tests was used. P values < .05 were considered statistically significant. Pearson’s correlation coefficient was used to assess correlations. Bland-Altman analysis was used to evaluate the variability between two measurements techniques. Differences between methods are represented in absolute units.
After two months, 20 echocardiographic data sets were reassessed by one reader to determine the intraobserver variability by correlation and Bland-Altman analysis. The reading was performed on the same data and with the same rules as for the first reading. Correlation and Bland-Altman analyses were also used to determine the interobserver variability among all four echocardiographic readers. Averaged interobserver variability parameters were used to compare EF assessment methods.
Analyses were performed using Statistica version 6.1 (StatSoft, Inc., Tulsa, OK) and MedCalc version 9.2.1.0 (MedCalc, Mariakerke, Belgium).
Results
Feasibility
Visual assessment of EF was deemed possible by the four echocardiographic readers in 76 to 79 patients (94%–98%). They succeeded with EF quantification according to the modified Simpson’s rule in 75 patients (93%). Tracking-based EF assessment was possible in 73 patients (90%). Failures to apply the latter method were due to poor image quality ( n = 3), inadequate electrocardiographic tracing ( n = 2), or too low a frame rate ( n = 3).
Time Needed for Analysis
Visual inspection proved to be the fastest method for EF estimation. The average time needed to determine an EF from all three apical views was 25 ± 13 sec. Conventional EF quantification according to the modified Simpson’s rule was the slowest approach (104 ± 22 sec, P < .001 vs visual assessment). EF estimation using the tracking-based EF assessment software took on average 54 ± 22 sec, twice as fast as conventional EF assessment using the modified Simpson’s rule ( P < .001) but slower than visual assessment ( P < .001) ( Figure 2 ).
Reference EF
Analysis of variance revealed no significant difference between the average EF results from the four different methods (MRI, visual, modified Simpson’s rule, and tracking; Figure 3 ).
EF estimates from the two MRI readings correlated well ( r = 0.89, P < .05). Bland-Altman-analysis showed minor bias (0.7%) but considerable variance (1.96 SDs, ±11.0%; Figure 4 ).
From all three echocardiographic EF estimates, the average visual EF assessment correlated best with the average MRI assessment (visual assessment vs Simpson’s rule: r = 0.82 vs r = 0.79; tracking-based assessment: r = 0.70; all P values < .01). Despite low bias, Bland-Altman analysis revealed considerable variance (1.96 SDs, ±16.5%; Figure 5 A).
To obtain a reliable EF reference for our study, an expert consensus had to resolve EF differences of > 10% between modalities in 11 cases. The MRI results were corrected in six cases, and the echocardiographic results were adapted in five cases, which improved the correlation between modalities ( r = 0.89, P < .05) and led to lower variance on Bland-Altman analysis (1.96 SDs, ±9.4%; Figure 5 B).
The average of MRI and visual echocardiographic EF assessment was then used as the reference EF for the study.
Echocardiographic EF Estimation Methods
Comparison to Reference EF
Quantitative echocardiographic EF estimation according to the modified Simpson’s rule was correlated closely with the reference EF ( r = 0.80, P < .01). Bland-Altman analysis revealed an average bias of –0.66% and a variance (1.96 SDs) of ±10.4%. This approach tended to mildly overestimate function in hearts with low EFs and to underestimate it in hearts with high EFs ( Figure 6 A).
