## Background

Left atrial (LA) enlargement has been acknowledged as a significant predictor of cardiovascular morbidity and mortality.

## Methods

To evaluate the accuracy of two-dimensional and three-dimensional echocardiography for determining LA volume, LA volume measurements by echocardiography were compared with those measured by 64-slice multidetector computed tomography (MDCT) as a reference standard.

## Results

Fifty-seven consecutive patients (mean age, 66 ± 11 years; 59% men) referred to echocardiography and MDCT on the same day were prospectively evaluated. LA volume by three-dimensional echocardiography was correlated closely with that by MDCT ( *r *= 0.95, *P *< .0001), with 8% underestimation. LA volume by two-dimensional echocardiography was correlated less well with that measured by MDCT ( *r *= 0.86, *P *< .0001) and consistently underestimated LA volume by 19%, particularly as the left atrium enlarged.

## Conclusions

LA volume assessment by three-dimensional echocardiography was correlated closely with that measured by MDCT, albeit with an 8% underestimation. Three-dimensional echocardiography is a feasible noninvasive method to evaluate LA volume.

Left atrial (LA) size is a clinically useful measurement because it has been acknowledged as a significant predictor of atrial fibrillation, stroke, heart failure, and death. Moreover, LA size as determined by LA volume expresses the severity of left ventricular diastolic dysfunction. To assess LA volume, two-dimensional (2D) echocardiography has been used, but it has been limited in accuracy because of the irregular geometry of the left atrium.

Currently, real-time three-dimensional (3D) echocardiography has become widely available; this technique does not rely on geometric assumptions. Several studies have demonstrated the improved accuracy of 3D echocardiographic measurements of left ventricular volume and right ventricular volume. Similarly, a number of studies have demonstrated the feasibility of 3D echocardiography for the assessment of LA volumes, which has been validated against magnetic resonance imaging (MRI). However, the accuracy of LA volume measurements using 3D echocardiography has not been entirely validated against multidetector computed tomography (MDCT), which offers several advantages over MRI for cardiac volume measurement. We therefore studied 3D echocardiographic evaluation of LA volume, using 64-slice MDCT as the reference standard for comparison.

## Methods

## Study Patients

With approval from the institutional review board, we prospectively enrolled patients with suspected coronary artery disease who were referred for MDCT for evaluation of the coronary arteries and transthoracic echocardiography for evaluation of cardiac function on the same day. Exclusion criteria were a history of cardiac arrhythmias, including pacemaker or implantable cardioverter-defibrillator implantation; greater than mild mitral stenosis or regurgitation; and prior mitral valve repair or replacement. Also, patients were excluded if they could not undergo MDCT because of renal insufficiency (serum creatinine > 1.5 mg/dL) or known allergies to iodinated contrast. Patients were first examined using MDCT and then immediately examined with echocardiography. Informed consent was obtained from all enrolled patients.

## Echocardiographic Data Acquisition

Transthoracic echocardiography was performed with patients in the left lateral decubitus position using a commercially available echocardiographic system (iE33; Philips Medical Systems, Andover, MA). All echocardiographic images were stored digitally, and measurements were performed offline using the commercially available QLAB software package (Philips Medical Systems). Measurements were obtained for three consecutive beats and averaged for all determinations. LA volumes were indexed to body surface area. All echocardiographic LA volume measurements were performed independently and without prior knowledge of the results of MDCT.

## LA Volume by Standard 2D Echocardiography

LA volume assessments by standard 2D echocardiography were done using an S5-1 probe (2–4 MHz) and determined using the biplane area-length formula: LA volume = (0.85 × four-chamber area × two-chamber area)/length, as recommended by the American Society of Echocardiography. Maximal LA area was measured with a planimeter for four-chamber and two-chamber views by manually tracing the endocardial border at ventricular end-systole, just before mitral valve opening, excluding the confluence of the pulmonary veins and LA appendage. Apical four-chamber and two-chamber views had to be of good quality, with no foreshortening, and the endocardial border had to be well visualized.

## LA Volumes by Real-Time 3D Echocardiography

LA volumes by real-time 3D echocardiography were collected in full-volume mode over four cardiac cycles during a breath hold using an X3-1 matrix-array transducer. Acquisition was triggered to the electrocardiographic R wave. Care was taken to ensure that the entire left atrium was included within a pyramidal 3D data set. The mean frame rate was 19 ± 1 frames/sec. LA volumes by 3D echocardiography were derived from semiautomated tracing of the LA endocardium at ventricular end-systole in perpendicular apical long-axis planes. This was performed by marking 5 points in the atrial surfaces of the mitral annulus: at the anterior, inferior, lateral, and septal annuli, and the fifth point at the apex of the left atrium. Once this was complete, the endocardial surface was automatically delineated, a mathematical model of the left atrium could be visualized from different points of views, and the LA volume calculation was obtained ( Figure 1 ). Manual modification was made to correct the automatic tracings if needed.

## LA Volumes by 64-Slice MDCT

All contrast-enhanced cardiac multidetector computed tomographic examinations were obtained using a 64-slice scanner (Aquilion 64; Toshiba Medical Systems, Otawara, Japan), with patients in a resting supine position and a single breath hold, with retrospective electrocardiographic gating. Patients with heart rates > 70 beats/min received 20 to 60 mg metoprolol orally 1 hour before the scan (29 of 57 [51%]). Image acquisition was obtained using an injection of iodine-based contrast media (iohexol 350 mg I/mL): 50 mL of intravenous contrast agent injected through an antecubital vein at 4 mL/sec, followed by 30 mL of saline solution chaser. Scanning was triggered once contrast material was seen in the ascending aorta and when the contrast was higher in the aorta than in the pulmonary artery trunk. Scan parameters included 400-msec gantry rotation time, pitch of 0.175 cm, slice thickness of 0.5 mm, tube voltage of 120 kV, and tube current–time product of 160 mAs. Data sets were transferred to a dedicated workstation for further analysis using commercially available software (AZE Virtual Place version 3.00; AZE Ltd., Tokyo, Japan). Initially, images were reconstructed throughout the cardiac cycle every 4% of the R-R interval (a total of 25 phases). Views of the mitral valve were reviewed throughout the reconstructed phases to determine the timing of end-systole, which was defined as the earliest reconstructed phase before mitral valve opening. Reconstructed slice thickness, using a soft tissue convolution kernel, was 1 mm. Manual data analysis was performed on the stack of continuous slices ( Figure 2 ). The mean number of slices used for LA volume calculation was 28 ± 8. The borders of the left atrium were defined as the plane of the mitral valve and the visually apparent juncture of the left atrium with the pulmonary veins. The pulmonary veins and LA appendage were not included in the measurements. LA volumes were indexed to body surface area. Multidetector computed tomographic images were analyzed independently and without knowledge of the echocardiographic results.

## Intraobserver and Interobserver Variability

Interobserver variability was assessed from 10 randomly selected images by two independent observers, each blinded to the results obtained by the other. Intraobserver variability was assessed by repeated measurements from 10 images by the same observer 1 week after the first analysis. Interobserver and intraobserver variability was calculated as the absolute difference between repeated measurements in percentage of their mean.

## Statistical Analysis

Continuous variables are expressed as mean ± SD. To measure the strength of the relation of LA volumes between any two methods, linear regression analysis with Pearson’s correlation coefficient ( *r *) was performed. Linear regression models were used to calculate the predicted LA volume value on MDCT from 2D or 3D echocardiographic LA measurements. LA volume by 2D or 3D echocardiographic LA measurement was set as the independent variable, and LA volume by MDCT was set as the dependent variable, and a regression equation was derived using the slope of the relation and the intercept. To evaluate the bias and limits of agreement between any two methods, as well as for systematic variation with increased measurement of LA volumes, the Bland-Altman method was used. *P *values < .05 were considered statistically significant.

## Results

We prospectively evaluated 57 consecutive patients referred for MDCT and echocardiography on the same day. The mean age of the study patients was 66 ± 11 years, and 59% were men. The clinical characteristics of study patients are detailed in Table 1 .

Variable | Value |
---|---|

Age (y) | 66 ± 11 |

Men | 35 (59%) |

Body mass index (kg/m ^{2 }) |
24.1 ± 3.3 |

Body surface area (m ^{2 }) |
1.63 ± 0.17 |

Systolic blood pressure (mm Hg) | 138 ± 20 |

Diastolic blood pressure (mm Hg) | 82 ± 11 |

Coronary artery disease | 11 (19%) |

Systemic hypertension | 40 (70%) |

Diabetes mellitus | 10 (18%) |

Dyslipidemia | 28 (49%) |

Left ventricular ejection fraction (%) | 67 ± 5 |

The mean heart rate was 61 ± 8 beats/min during MDCT, 62 ± 6 beats/min during 2D echocardiographic image acquisition, and 62 ± 7 beats/min during 3D echocardiographic image acquisition, with no significant difference between modalities. Acquisition time for 2D or 3D echocardiography was <5 min per patient, and offline LA volume calculation times were <5 min by 2D echocardiography and 5 to 10 min by 3D echocardiography, depending on image quality. All recordings were sufficient quality for interpretation.

The average LA volume was 32.2 ± 7.6 mL/m ^{2 }(range, 18.5–60.2 mL/m ^{2 }) by MDCT, 29.7 ± 7.5 mL/m ^{2 }(range, 17.8–59.4 mL/m ^{2 }) by 3D echocardiography, and 26.2 ± 5.8 mL/m ^{2 }(range, 15.1–44.4 mL/m ^{2 }) by 2D echocardiography. LA volume by 3D echocardiography was correlated closely with that measured by MDCT ( *r *= 0.95, *P *< .0001) as shown in Figure 3 ( *top left *). Bland-Altman analysis showed that 3D echocardiographic methods slightly underestimated LA volume by 8% compared with MDCT, with a mean difference of 2.5 mL/m ^{2 }and 95% limits of agreement of ±3.5 mL/m ^{2 }( Figure 3 , *bottom left *).

LA volume by 2D echocardiography was correlated less well with that measured by MDCT ( *r *= 0.86, *P *< .0001), as shown in Figure 3 ( *top center *). Bland-Altman analysis showed that 2D echocardiographic methods, however, significantly underestimated LA volume by 19% compared with MDCT, with a mean difference of 6.0 mL/m ^{2 }and relatively wide 95% limits of agreement of ±6.3 mL/m ^{2 }( Figure 3 , *bottom center *). LA volume by 2D echocardiography showed a directional measurement bias, with a linear regression coefficient of *R *= 0.36 ( *P *< .0001), indicating increasing underestimation at larger LA volumes.

LA volume by 2D echocardiography was correlated less well with that measured by 3D echocardiography ( *r *= 0.84, *P *< .0001), as shown in Figure 3 ( *top right *). Bland-Altman analysis showed that 2D echocardiographic methods underestimated LA volume by 12% compared with 3D echocardiographic methods, with a mean difference of 3.5 mL/m ^{2 }and relatively wide 95% limits of agreement of ±6.5 mL/m ^{2 }( Figure 3 , *bottom right *). Also, LA volume by 2D echocardiography showed a directional measurement bias, with a linear regression coefficient of *R *= 0.31 ( *P *< .0001), indicating increasing underestimation at larger LA volumes.

The regression formula to obtain the predicted LA volume value by MDCT from 3D echocardiographic LA volumes can be expressed as follows: corrected 3D echocardiographic LA volume = 2.94 + 0.99 × LA volume by 3D echocardiography. After application of the regression formula, the mean difference was eliminated, with 95% limits of agreement of ±3.5 mL/m ^{2 }( Figure 4 , *left *). The regression formula to obtain the predicted LA volume value by MDCT from 2D echocardiographic LA volumes can be expressed as follows: corrected 2D echocardiographic LA volume = 0.08 + 1.23 × LA volume by 2D echocardiography. After application of the regression formula, the mean difference was eliminated, and the limits of agreements slightly narrowed (95% limits of agreement of ±5.7 mL/m ^{2 }; Figure 4 , *right *). There was no significant directional measurement bias ( *P *= .14).