Background
Accurate assessment of left atrial appendage (LAA) morphology is crucial in determining an LAA occlusion strategy. The aim of this study was to develop a novel echocardiographic volume-rendered imaging technique to visualize LAA morphology.
Methods
This was a retrospective study. Forty patients with atrial fibrillation who underwent three-dimensional (3D) transesophageal echocardiography (TEE) and cardiac computed tomographic angiography (CCTA) before catheter ablation were enrolled. Full-volume 3D data were acquired and displayed in gray values–inverted (GVI) mode. Threshold segmentation and interactive segmentation were used to create 3D digital replicas of the LAA chambers. The morphologic classification, number of lobes, and dimensions of the LAA were analyzed and compared with the data obtained with CCTA.
Results
LAA morphology and measurements were successfully acquired via CCTA and 3D GVI TEE in all 40 cases. In terms of LAA morphologic classifications, 19 cases of chicken wing, eight of windsock, nine of cauliflower, and four of cactus morphology were determined using 3D GVI TEE, and 20 cases of chicken wing, eight of windsock, eight of cauliflower, and four of cactus morphology were determined using CCTA. The κ value between these two methods was 0.963. Measurements of maximal diameter, minimal diameter, and area of the ostia and the depth of the LAAs were larger when based on the 3D GVI transesophageal echocardiographic data than when using cardiac computed tomographic angiographic data ( P < .01); however, there was agreement between the results. Formed thrombi were well displayed by both computed tomography and 3DGVI TEE.
Conclusions
Three-dimensional GVI TEE can be used to acquire LAA morphologic volume-rendered images that are similar to computed tomographic volume-rendered images, and it shows promise as a feasible and valuable modality for planning individual LAA occlusion procedures.
Percutaneous left atrial appendage (LAA) closure is a promising nonpharmacologic approach for stroke prevention in patients with nonvalvular atrial fibrillation. Because of the significant variations in LAA morphology, an accurate assessment of the LAA is crucial in selecting the appropriate device size during preoperative planning. Generally, preprocedural measurements are performed with transesophageal echocardiography (TEE), and the morphology of the LAA is evaluated with cardiac computed tomographic (CT) angiography (CCTA). Recently, the clinical application of three-dimensional (3D) TEE has made it possible to acquire 3D information about the LAA, which improves the accuracy of measurements. However, unlike CCTA, 3D TEE cannot intuitively show the entire shape of the LAA. In this study, we used a novel technique called 3D gray values–inverted (GVI) imaging to obtain 3D images of the LAA that are similar to CT volume-rendered (CT-VR) images. Because the volume images could not be displayed with the gray values inverted using commonly available software and diagnostic platforms, we used special postprocessing software that has been shown to be valid for 3D echocardiography in previous studies. This method, which was validated on the basis of the use of CT images as references and is referred to as “echocardiographic volume rendering,” might provide additional imaging support for the planning of LAA occlusion procedures.
Methods
Study Participants
This was a retrospective study. Patients with atrial fibrillations who were admitted to Renmin Hospital of Wuhan University between March 2014 and February 2015 and underwent 3D TEE and CCTA before interventional treatment were enrolled in the study. Patients with other forms of arrhythmia, cardiomyopathy, congenital heart disease, or coronary heart disease were excluded. A total of 49 patients who had both transesophageal echocardiographic and cardiac CT images in our hospital database were considered for this study. Four patients were excluded because of the high frame rate (>12 volumes/sec) of their transesophageal echocardiographic studies to ensure adequate spatial resolution of the images that were analyzed. Five cases were excluded because of dislocation and poor splicing of the CT images. A total of 40 patients (age range, 28–72 years; mean age, 56 years) were included in this study. Thirty-five patients had persistent atrial fibrillation, and five had paroxysmal atrial fibrillation. The ethics committee at Renmin Hospital of Wuhan University approved the study protocol.
TEE
Both two-dimensional (2D) and 3D TEE were performed using the GE Vivid E9 platform (GE Vingmed Ultrasound AS, Horten, Norway) equipped with a 6VT-TEE probe. TEE was conducted according to a standardized protocol with local hypopharyngeal anesthesia. Two-dimensional image loops were recorded with the transducer array rotated through 0°, 45°, 90°, and 135°. Three-dimensional image loops were stored in 4D zoom single-beat mode with the LAA included in the sampling frame. The temporal resolution of the 3D transesophageal echocardiographic images was reduced to 8 to 12 frames/sec for better spatial resolution depending on different sampling volumes. All 2D and 3D images were stored with at least five loops available for offline review.
Measurements on 2D and 3D Transesophageal Echocardiographic Images
The cross-section of the left circumflex coronary artery was used as the anatomic marker to measure the LAA ostial dimension in the 0°, 45°, 90°, and 135° planes, and the maximal value was recorded as the LAA ostial dimension of the 2D echocardiogram. LAA depth was measured on these planes as the distance from the midpoint of the ostium to the tip of the LAA. Three-dimensional image loops were reviewed to identify an optimal frame located at 75% of the R-R interphase of the cardiac cycle. The maximal and minimal diameters and the area of the ostium of the LAA were measured on 3D transesophageal echocardiographic resliced images of the LAA ostium. The depth of the LAA was measured on 3D transesophageal echocardiographic resliced images of the long axis of the LAA.
Acquisition and Segmentation of the 3D and 3D GVI Transesophageal Echocardiographic Images
Volume images in the Digital Imaging and Communications in Medicine (DICOM) format of the same frames measured on 3D TEE were imported into specialized segmentation software (Mimics; Materialise, Leuven, Belgium), which can read and measure the volume of DICOM images acquired via 3D echocardiography. GVI mode was used to display the chambers of the heart instead of the valves and walls. Various segmentation methods were used to distinguish the LAA chamber. First, threshold segmentation was applied to 2D slice images to set the specific grayscale values, by which the blood pool border of the LAA can be distinguished from the endocardium of the LAA. In addition, if the border of the orifice and the apex of the LAA could not be accurately recognized under the same grayscale values, the grayscale value of the orifice was used because the preprocedure landing zone evaluation (the area within the LAA where the occlusion device will be positioned) is usually based on a measurement taken at the LAA orifice. The maximum threshold value was 254, and minimal values were between 110 and 150, depending on the details of the different cases. Pectinate muscles were excluded from the volume images. An interactive segmentation method was then used for both 2D and 3D images to separate the LAA volume images from the background. After segmentation, every image slice was reviewed for the accuracy of the segmented images by comparing the segmented masks of the LAA chamber contours with the original images. Three-dimensional digital replicas of the LAA chambers were then created for assessment and measurement ( Figure 1 ).
CCTA and Segmentation
CCTA was performed using a 64-multidetector computed tomographic system (GE LightSpeed VCT; GE Healthcare, Milwaukee, WI), with a slice thickness of 0.625 mm. DICOM images at the same phase of the cardiac cycle as those acquired for TEE were exported to the Mimics segmentation software to ensure consistency with the echocardiographic images. The segmentation process was similar to that used for the 3D GVI transesophageal echocardiographic images; threshold segmentation, interactive segmentation, and a final review were performed sequentially to create the 3D LAA models.
Analysis of LAA Morphology and Measurements Using 3D GVI TEE and CT-VR
The definition of an LAA lobe refers to its anatomic definition. However, in the examination of the echocardiographic and CT images, the structures observed were the chambers of the LAAs, and it is impossible to distinguish between chambers divided by long, thick pectinate muscles and different lobes. Because lobes and long, thick pectinate muscles may have similar effects in relation to LAA occlusion, deep chambers divided by thick pectinate muscles were also counted as lobes as observed in the 3D GVI transesophageal echocardiographic and CT images used in this study. The classification of LAA morphology using both 3D GVI transesophageal echocardiographic images and CT-VR images followed the common CT standards. LAAs were classified into four morphologic groups: (1) chicken wing (an LAA with an obvious bend in the proximal part of the dominant lobe), (2) windsock (an LAA with a main lobe of a sufficient length as the primary structure), (3) cauliflower (an LAA that has a limited overall length without any forked lobes), and (4) cactus (an LAA with a dominant central lobe and secondary lobes extending from the central lobe). The ostial dimensions and the depth of the LAA were measured using Mimics software. The maximal and minimal diameters and the area of the ostia were measured in the interactive multiplanar reconstruction view of the LAA ostium. The depth of the LAA was measured using the interactive multiplanar reconstruction view of the long axis of the LAA.
Observer Variability
Intra- and interobserver variability of the 3D GVI transesophageal echocardiographic measurements and morphologic assessments were assessed in 10 randomly selected subjects. For interobserver variability, the same volume frames of each subject were analyzed by two different observers. The two observers analyzed all 3D echocardiographic data independently and were blinded to each other and to the clinical details of the data. For determining intraobserver reproducibility, the observers analyzed the data 1 week apart and were blinded to the final results from their first analysis and to the clinical details of the data. Echocardiographic and CT measurements were also performed by different observers, and observers were blinded to each other and to the clinical details of the data.
Statistical Analysis
Continuous data are expressed as mean ± SD and categorical data as frequencies or percentages. All data were first analyzed to assess the normality of their distribution using Kolmogorov-Smirnov tests. For comparisons of measurements between the echocardiographic and CT methods, paired t tests were used. To determine the agreement of the estimates of the number of lobes and morphologic types, as assessed by the two methods, a κ test was used, which was weighted according to the frequency of the number of LAA lobes and morphologic types. Differences in measurements between the methods are reported as biases and limits of agreement (2 SDs), as determined by Bland-Altman analysis. Interobserver and intraobserver agreement between the measurements of the LAA dimensions and the morphologic assessments was determined on the basis of intraclass correlation coefficients. Statistical analyses were performed using SPSS version 17.0 (SPSS, Chicago, IL) and MedCalc version 11.0.1.0 (MedCalc Software, Mariakerke, Belgium). A P value < .05 was considered to indicate statistical significance.
Results
Comparison of Morphologic Data Obtained Using 3D GVI TEE and CT-VR
Morphologic information and 3D models were successfully acquired from CT and 3D GVI transesophageal echocardiographic images in all 40 cases ( Figure 2 ). It took approximately 25 min to process the images of one LAA. The LAA morphologic types observed via 3D GVI TEE were determined to be chicken wing in 19 cases (47.5%), windsock in eight cases (20%), cauliflower in nine cases (22.5%), and cactus in four cases (10%). These four types were also observed using CT-VR, with chicken wing found in 20 cases (50%), windsock in eight cases (20%), cauliflower in eight cases (20%), and cactus in four cases (10%). In one case, the morphologic type was determined to be cauliflower using 3D GVI TEE and chicken wing using CT-VR. The κ value between these two methods for this assessment was 0.963.
Number of Lobes as Determined Using 3D GVI TEE and CT-VR
LAAs with single lobes, double lobes, or multiple lobes were observed in 15, 17, and eight cases, respectively, using 3D GVI TEE and in 14, 16, and 10 cases, respectively, using CT-VR. The κ value between these two methods for this assessment was 0.884.
Measurements and Agreement between 3DGVI TEE and CCTA
On the basis of 3D GVI transesophageal echocardiographic and cardiac CT angiographic measurements, the maximum diameters of the LAA ostia were 21.5 ± 3.9 and 23.2 ± 4.3 mm, the minimum diameters of the ostia were 16.7 ± 3.1 and 17.9 ± 3.2 mm, the areas of the ostia were 287.0 ± 116.0 and 334.7 ± 121.9 mm 2 , and the depths of the LAAs were 29.3 ± 5.2 and 31.4 ± 5.0 mm, respectively. The estimates of the dimensions of the LAAs determined using CCTA were significantly larger than those determined using 3D GVI TEE for the maximal and minimal diameters and the area of the ostium and for the depth of the LAA. The agreement between the 3D GVI transesophageal echocardiographic and CT estimates, assessed on the basis of Bland-Altman analysis, showed that the limits of agreement for the maximal and minimal diameters and the area of ostium and for the depth of the LAA were −1.0 and 4.6 mm, −2.5 and 4.9 mm, −14.6 and 110.0 mm 2 , and −1.1 and 5.3 mm, respectively, and that 95%, 92.5%, 97.5%, and 95% of data points, respectively, fell within the limits of agreement ( Figure 3 ).
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
