Two-dimensional (2D) and real-time three-dimensional (RT3D) echocardiography can be used to assess left atrial (LA) size, but their correlation in dogs remains unknown.
Two-dimensional (2D) and real-time three-dimensional (RT3D) echocardiography can be used to assess left atrial (LA) size, but their correlation in dogs remains unknown.
Estimations of LA size differ depending on the echocardiographic technique.
Privately owned dogs; 70 with myxomatous mitral valve disease and 32 healthy control dogs.
Prospective observational study comparing RT3D volume at atrial end-diastole (RT3DLAd) with 4 different 2D methods of estimating LA size: LA diameter and area in short-axis (LAsax and LAarea) and LA diameter in long-axis (LAlax), both as indexed variables and as predictors of LA volume indexed to body weight (BW) using allometric scaling and geometric assumption of sphericity. Furthermore, agreement between indexed 2D based methods was studied using concordance correlation coefficient (ρc) and Bland–Altman plots.
None of the indexed 2D methods of estimating LA size showed good correlation with BW-indexed RT3DLAd volumes. Estimates of LA volumes from 2D measurements using allometric scaling showed better correlation with RT3D volumes than corresponding calculated volume approximations. The best correlation was found between RT3DLAd and estimated LA volumes based on allometric scaling of LAlax (ρc = 0.89) followed by LAarea (ρc = 0.86) measurements. Comparing indexed 2D-based measurements of LA size, best agreement was found between LAsax to aortic diameter and LAsax to expected LA diameter, based on allometric scaling.
Allometric scaling of 2D-based measurements of LA showed good correlation with RT3DLAd, whereas corresponding indexed measurements or calculated volume approximations did not.
aorta, aortic diameter in short-axis
aortic area in short-axis
congestive heart failure
coefficient of variation
left atrial appendage
left atrial area in short-axis
left atrial volume at atrial end-diastole
left atrial expected diameter
left atrial diameter in long-axis
left atrial diameter in short-axis
myxomatous mitral valve disease
Assessment of left atrial (LA) size is crucial in the evaluation of heart disease, because it allows the examiner to estimate the risk for development of left-sided congestive heart failure (CHF). Indeed, LA size has been shown to be one of the strongest predictors of outcome in dogs with myxomatous mitral valve disease (MMVD). LA acts as a reservoir for pulmonary venous return during ventricular systole, serves as a conduit for the passage of stored blood from LA to the left ventricle (LV) during early ventricular diastole, and serves as a contractile pump delivering 15–30% of LV filling during atrial systole. LA size commonly is estimated either in two-dimensional (2D) parasternal short-axis or long-axis views at ventricular end-systole or early diastole. Different methods of measuring LA diameter, circumference, and cross-sectional area have been used, and measurements are commonly indexed to aortic (Ao) diameter, circumference, and cross-sectional area, respectively. Allometric scaling in which linear cardiac measurements are indexed to body weight (BW) raised to the power of approximately 1/3 also is commonly used. It is currently not known which of the weight-dependent or weight-independent measurements of LA size most accurately reflect true LA size.[4, 5] Because LA is a three-dimensional (3D) structure, LA dilatation may occur in all directions (ie, cranio-caudal, medio-lateral, and ventro-dorsal direction) although perhaps not uniformly.[6, 7] Thus, the accuracy of diameter and area measurements in one plane to assess the size of 3D structures may be questioned, and volumetric scanning using 3D echocardiography has the potential to provide more accurate estimates of chamber size. LA diameter and area measurements, especially in enlarged LA, were shown to be poor predictors of 3D LA volumes in a study of human patients.
3D echocardiography was first described by Dekker et al, and real-time 3D (RT3D) echocardiography was first reported in 1990.[10, 11] Presently, technology is available using a matrix transducer generating ultrasonic beams in a phased-array manner, which are automatically aimed in multiple directions, allowing simultaneous visualization of the beating heart. RT3D data sets are usually obtained from several cardiac cycles. During each cardiac cycle, a wedge-shaped subvolume is acquired, and, by combining subvolumes from several cardiac cycles, a full volume dataset is obtained.
Although the RT3D method may be expected to more accurately estimate LA size compared with 2D estimates, it is more time consuming, and off-line analysis using specialized equipment often is needed. 2D estimates of LA size are needed in the clinical setting, and the question may arise on which of the 4 most commonly used 2D methods for estimating LA size shows best agreement with the RT3D method. Whether a monodimensional or a bidimensional measurement reliably can estimate a known volume or not may be investigated using different schemes. Thus, the purpose of this study was to compare 4 different 2D-based echocardiographic methods to estimate LA size with RT3D assessment of LA volume, using indexed measurements, allometric scaling, and appropriate equations to estimate volumes from a linear or area measurement, assuming that the sought volume is a sphere. The 2D-based methods also were compared with each other to evaluate their respective interchangeability.
A total of 102 privately owned dogs—70 dogs with MMVD and 32 healthy control dogs, presented at Albano Animal Hospital, Stockholm, Sweden, were included in the study. All dogs were examined using the same equipment and the same protocol. Dogs were considered healthy based on normal findings on physical examination, ECG, echocardiography, and Doppler examinations. Diagnostic criteria for MMVD included thickened mitral valve leaflets and mitral regurgitation detected on color-coded Doppler echocardiogram. All examinations were performed and evaluated by 1 veterinary specialist in cardiology (AT).
2D and RT3D examinations were performed with an ultrasound unit1 equipped with 3.0–8.5 MHz phased-array transducers (for 2D) and 7–2 MHz matrix transducer (for RT3D) in all dogs. Dogs were unsedated and gently restrained in left and right lateral recumbency for RT3D and 2D images, respectively, during the examination. Measurements of 2D images of LA and Ao were made directly on the monitor freeze-frame image. Measurements of LA (LAsax) and Ao diameters in early ventricular diastole were made on 2D parasternal short-axis view at the level of the Ao valve at the 1st frame after Ao valve closure. LA (LAarea) and Ao (Aoarea) areas were calculated using the same image by tracing the circumferences and using automated calculation software provided by the ultrasound machine. The LA appendage (LAA) was included only in the LAarea measurement. Measurements of LA (LAlax) long-axis diameters were made in early ventricular diastole on an image obtained on the frame just before mitral valve opening, measuring a line bisecting LA at the level of the fossa ovalis and parallel to a line crossing the mitral valve annulus. Measurements were made on 3 consecutive cardiac cycles, and mean values were used in the statistical analyses.
RT3D images of LA were created using the 7–2 MHz matrix transducer to obtain a pyramidal volume in real time. Real-time volumes acquired from 4 to 7 cardiac cycles were obtained to produce a larger pyramidal volume, producing a full data set. The lowest possible scan line density was used in all dogs to produce the largest possible pyramidal volume of acquisition.
Off-line analyses of RT3D LA volume were made using a software.2 Analysis of LA volume in left atrial end-diastole (RT3DLAd), concurrent with ventricular end-systole, was timed as the frame preceding mitral valve opening. Care was taken to exclude LAA and pulmonary veins. The analysis involved manual definition of 4 reference points (2 points in each view) placed at the endocardial border of the LA side of the mitral annulus in both 4-chamber and 2-chamber views. A 5th reference point was placed at the midpoint of the dorsal LA border in either view. The endocardial border was then traced using an automated detection process to create a cast of the LA cavity (Fig 1). When the volume was computed, the endocardial border detection was verified for accuracy and manually edited when necessary. Three acquisitions were made for each variable, and mean values were used in the statistical analyses. All measurements were made by 1 veterinary specialist in cardiology (AT).
Within-day variability was assessed using 6 dogs, including 3 dogs without heart disease and 3 dogs with MMVD (class BI and BII according to the CHIEF classification of CHF).[15-17] Each dog was examined 6 times on a given day. RT3D-derived volumes of LA in atrial diastole were estimated, and 2D diameters and areas of LA and Ao in short-axis, as well as LA diameters in long-axis, were measured. Each variable was measured on each of the 6 acquisitions for each dog and the resulting mean values and standard deviations were used to determine the coefficient of variation (CV) (Table 1). All CV values were below 15%. There was considerable difference in variability among different methods with the lowest CV value obtained for LAlax (4.2%) and the highest for LAarea/Aoarea (14.2%).
|Variable||SD||CV (%) and Range|
To compensate for differences in body size among included dogs, the 2D estimations of LA diameters in short- and long-axis views were indexed to the diameter of Ao in short-axis view (LAsax/Ao and LAlax/Ao). Furthermore, the LAarea was indexed to Aoarea (LAarea/Aoarea). Finally, the observed LA diameter in short-axis view was indexed to expected normal LA diameter (LA/LAexp). The expected normal LA diameter was calculated according to the following equation: LAexp = 0.76 × BW (kg)0.345. In comparison with other measurements, LAd volume estimated using RT3D echocardiography, using allometric scaling, and using geometric assumptions, was indexed to BW.
The associations between the RT3D volume and the 2D-based methods of measuring LA size (LAsax, LAlax, and LAarea) were investigated using allometric scaling. The RT3D volume was predicted according to the formula Y = aMb, where Y is the RT3D volume (mL), M is the 2D-based measurement, and a and b are constants. The logarithmic form of this equation is log10 RT3D LA = log10 (a) + b log 10 (M), a linear equation, which, in turn, allows estimation of the constants a and b using linear regression analysis, according to Cornell et al.
Making the assumption that LA is a sphere, the diameter (d) of LA in short-axis and long-axis was used to estimate LA volume, using the following equation: volume = .3 Likewise, the LA area (A) in short-axis was used to estimate LA volume using the following equation: volume = .
A computer program3 was used for all statistical analyses. The Kruskal–Wallis test was used for testing equality of medians between dogs with and without MMVD for breed, sex, age, BW, and heart rate. The agreement between body size-indexed values of the 4 different 2D-based methods (LAsax/Ao, LAlax/Ao and LAarea/Aoarea and LA/LAexp) and the RT3D LA volume corrected for BW was evaluated by calculating the concordance correlation coefficient (ρc).[18, 19] In brief, this correlation coefficient is a measurement of the agreement between 2 variables, and it evaluates the degree to which pairs of observations fall on the 45° line through the origin. A ρc of 1 indicates perfect fit. Estimates of ρc, their 95% confidence intervals, and their P-values were used to compare the different measurements of LA size with each other.
Linear regression analysis was used to investigate the relationship between log10 transformed values of 2D-based measurements of LA size (LAsax, LAlax, and LAarea) and RT3D LA volumes. The values, their standard errors of estimates of the intercept, and slope of fitted lines were used to define the constants in the allometric equation as previously described by Cornell et al.
In addition to the concordance correlation coefficient, Bland–Altman plots were used to assess agreement between the indexed 2D based methods (LAsax, LAlax, LAarea and LAexp), by plotting the mean value of the 2 methods by absolute difference. The presence of bias between the 2 methods compared was investigated by the intercept and slope (Pearsons correlation coefficient [ρ]) of the fitted line using linear regression. Level of significance was set at P < .05.
A total of 102 dogs of 43 different breeds were included in the study: Cavalier King Charles Spaniel (10), mixed breed (9), Labrador Retriever (6), Norfolk Terrier (6), Bichon Frise (5), Flat-Coated Retriever (5), Miniature Schnauzer (5), and <5 dogs each of 36 other breeds. Seventy dogs were diagnosed with MMVD, and 32 dogs were healthy controls. According to the CHIEF classification,[15-17] 46 dogs were classified without CHF (45 in class BI and 1 in BII), and 24 dogs were classified with CHF (18 in CII and 6 in CIII). Sixty-two dogs (61%) were males and 40 dogs (39%) were females. BW ranged from 2.8 to 39 (median, 11 kg). Age ranged from 1 to 14 (median, 4 years) for healthy control dogs, and from 3 to 17.6 (median 9.7, years) for MMVD dogs. Heart rate ranged from 68 to 190 (median, 120 beats/min) in all dogs. Subvalvular aortic stenosis and pulmonic stenosis were diagnosed in 2 MMVD dog, respectively. There were no significant differences between dogs with and without MMVD with respect to breed, sex, BW, or heart rate, except for age at the time of examination (P < .001).
None of the 2D-based methods of estimating LA size indexed to Ao or to expected LA diameter showed good agreement with the RT3D method of estimating LA volume at atrial end-diastole indexed to BW. Using the concordance correlation coefficients to evaluate the agreement between methods, the best correlation with the RT3D method was found for the LA/LAexp method, followed by the LAsax/Ao and the LAlax/Ao methods. In comparison with the RT3D method, the LA/LAexp, LAsax/Ao, and LAlax/Ao methods underestimated LA size at higher values of LA size, whereas the LAarea/Aoarea method overestimated LA size at all values (Fig 2a–d).
Comparisons between the 2D-based methods showed correlations, as indicated by the concordance correlation coefficient, in the following ranking order (from greatest agreement to the lowest): LAsax/Ao−LA/LAexp, LAsax/Ao−LAlax/Ao, LAlax/Ao−LA/LAexp, LAlax/Ao−LAarea/Aoarea, LAsax/Ao−LAarea/Aoarea, and LAarea/Aoarea−LA/LAexp. (Table 2). Only the LAsax/Ao−LA/LAexp showed good agreement using Bland–Altman plot (Table 3, Fig 3), whereas the other 2D-based monodimensional indices showed moderate agreement in the comparisons, and the LAarea/Aoarea showed poor agreement with any of the other methods.
|Compared Methods||ρc||95 CI||P-value|
After log transformation, the LAlax showed best linear fit when regressed against log RT3D LA volumes, as indicated by the highest model R2 (Table 4). The LAsax and LAarea showed lower, but similar, linear fits to the 3D method. Using concordance correlation coefficients, good correlation was found between BW-indexed values of RT3DLAd and BW-indexed 2D-based estimates of LA volume using allometric scaling. The concordance correlation coefficients of RT3DLAd with estimated LA volumes based on allometric scaling indexed to BW were as follows: LAsax 0.82, LAlax 0.89, and LAarea 0.86 (Fig 4a–c).
|Measurement||Log (a)||a||STd Error of Estimate||b||Std Error of Estimate||R2|
Using concordance correlation coefficients to assess agreement between methods, none of the 3 different 2D-based methods using LA diameter or area to calculate LA volume indexed to BW showed good correlation with the RT3D method of estimating LA volume at atrial end-diastole. The concordance correlation coefficients of RT3DLAd with estimated LA volumes based on volumetric calculation were as follows: LAsax 0.68, LAlax 0.57, and LAarea 0.65 (Fig 5a–c). Except for the LAsax based estimate, the 2D-based methods overestimated the LAd.
In this study, none of the indexed 2D-based methods of using LA diameter or area showed good correlation with RT3DLAd corrected for BW. However, estimations of LA volume indexed to BW using allometric scaling of 2D-based LA diameters and area showed good correlation with RT3D-based LA volumes indexed to BW. Calculated volume approximations of the same measurements showed a lesser degree of correlation. The best correlation was found between RT3LAd and estimated volumes based on allometric scaling of LAlax followed by LAarea measurements.
From a practical point of view, 2D linear or area measurements are easy to perform with any echocardiographic system, whereas the RT3D method requires specific software and off-line analysis. In a clinical situation, where immediate results of measurements are needed for medical decision-making, the RT3D method is impractical, and a quick and easy 2D method is needed. As the LAlax method using allometric scaling showed especially good agreement with the RT3D method, the equation used needs to be incorporated in the software of echocardiography machines to be clinically useful. Most importantly, it is essential to be aware of the inherent inaccuracy of 2D methods when estimating LA size without the use of allometric scaling equations.
LA diameter and area often are indexed to Ao, but indexing to predicted normal LA measurements of diameter or area obtained by allometric scaling of measurement and BW also is used. However, LAsax/Ao ratio often is provided by the software in the ultrasound machine and, as opposed to allometric scaling, does not involve additional equations. It also may be argued that the aortic diameter would be expected to change less over time than BW. In this study, LAsax/Ao method showed excellent agreement with the method using allometric scaling indexing LA to BW, indicating that these methods can be used interchangeably to estimate LA size. The other 2D-based methods to estimate LA volume using mono- or bidimensional measurements did not show good agreement with each other. These findings imply that different 2D methods of estimating LA size may yield different results for a given LA volume, and results of studies using different techniques may not be comparable.
Repeatability was considered good (ie, <10%) or acceptable (ie, <15%) for all measurements in this study. The LAlax method showed the lowest (4.2%) within-day variability of all 4 methods, whereas the LAarea/Aoarea method showed the highest (14.2%) within-day variability in this study (Table 1). These results show that measurements of 1D structures are more reproducible than tracing the circumference of 2D structures to create area measurements, which may indicate that the former is the method of preference.
There are several limitations of this study. The LAarea method includes the LAA, whereas the other 4 methods do not. However, the error in estimating LA size resulting from the exclusion of LAA in these methods is probably negligible. Time points when measurements were made did differ slightly among methods, because the LAsax and the LAarea methods use the 1st frame after Ao valve closure, and the LAlax method and the atrial end-diastolic RT3D measurements use the frame just before mitral valve opening. However, because these time points are very close to each other, only minimal variation in LA size is expected. Because RT3D estimations of LA volumes were not evaluated against a gold standard technique such as magnetic resonance imaging (MRI), this new technique could not be evaluated for accuracy. The RT3D method has been shown to systematically underestimate LA volumes compared with MRI in human patients.
In conclusion, none of the indexed 2D-based methods of LA diameter and area showed good correlation with the RT3D method of LA volume corrected for BW. Good correlation was found between RT3DLAd and estimated LA volumes based on allometric scaling of LAlax measurements indexed to BW, which thus may be considered to be the preferred technique. In addition, LAlax method showed the smallest within-day variability. Regardless of the chosen method, it is essential to be aware of the inherent inaccuracy of 2D techniques compared with RT3D techniques for estimation of LA size without the use of allometric scaling equations.
iE33; Philips Ultrasound, Bothell, WA
QLAB advanced quantitation system version 5.0; Philips Ultrasound, Bothell, WA
JMP, v.5.1, SAS Institute Inc, Cary, NC