Predictors of Spontaneous Echocardiographic Contrast Within the Left Atrial Appendage in Cardiac Computed Tomography of Patients with Atrial Fibrillation

Kotaro Ouchi (  kotaro-alex@jikei.ac.jp ) Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku https://orcid.org/0000-0002-02286845 Toru Sakuma Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku Takahiro Higuchi Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku Jun Yoshida Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku Ryosuke Narui Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku Ayumi Nojiri Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku Teiichi Yamane Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku Hiroya Ojiri Jikei University School of Medicine: Tokyo Jikeikai Ika Daigaku


Introduction
Atrial brillation (AF) is the most common cause of cardiogenic embolism, and more than 90% of embolic strokes result from thrombi in the left atrial appendage (LAA) [1,2]. The appearance of spontaneous echo contrast (SEC) in the LAA has been associated with a high risk of thromboembolic events and may represent the stage preceding thrombus formation [3,4]. The CHADS 2 scoring scheme is a validated method for estimating the risk of stroke and the need for anticoagulation therapy in patients with AF [5], and LAA morphology and reduced LAA ow velocity (LAAFV) have been reported as important predictors of SEC and thrombus in the LAA and risk of stroke in these patients [6,7]. The high spatial and temporal resolutions and multiplanar reconstruction capabilities of cardiac computed tomography (CT) allow noninvasive and detailed assessment of the cardiovascular system and good visualization of the LAA, but cardiac CT ndings regarding the relationship between SEC and LAA have not been reported in patients with AF. We therefore sought to de ne predictors of SEC using LAA ndings in cardiac CT of patients with AF.

Materials And Methods
Our institutional review board approved this retrospective observational study and waived the requirement for documentation of informed consent from patients. An opt-out option on our website allowed patients to decline inclusion of their data in the study.

Patients
We identi ed 783 consecutive patients with AF who underwent Transesophageal echocardiography (TEE) evaluation of SEC and ow velocity in the LAA prior to percutaneous pulmonary vein isolation (PVI) at our institution between January 6, 2013 and December 16, 2019. Among these cases, we also assessed data of transthoracic echocardiography (TTE) performed within the 6 months prior to PVI and cardiac CT performed within the 3 months prior to PVI. We excluded 139 patients who did not undergo TTE and/or cardiac CT within that period and 3 patients whose CT data could not be analyzed because of image degradation caused by an artifact (one case), the absence of part of the LAA from imaging range (one case), and inability to recognize the LAA structure (one case). No patient demonstrated LAA thrombus on TEE. The nal study included data of 641 patients (570 men, mean age, 59.35 ± 9.16 years, age range, 26-81 years; 71 women, mean age, 65.17 ± 6.88 years, age range, 44-79 years). When a patient underwent TEE more than once during the study period, we analyzed ndings of only the initial TEE ( Fig. 1). All patients who underwent PVI for AF routinely received anticoagulation therapies, either warfarin or non-vitamin K antagonist oral anticoagulants, prior to echocardiography and CT. For patients receiving warfarin, the international normalized ratio of prothrombin time (PT-INR) was controlled within the therapeutic range between 2 and 3 [8].

Transesophageal Echocardiography
Transesophageal echocardiography was performed at our institution according to standard clinical procedure using one of 4 ultrasound systems: Pro Sound Alpha 10 (multiplane 5.0 MHz transducer) (Aloka, Tokyo, Japan); Pro Sound F75cv (multiplane 5.0 MHz transducer) (Hitachi, Tokyo, Japan); iE33™ (multiplane 2.0 to 8.0 MHz transducer) (Philips Healthcare, Best, The Netherlands); or EPIQ7 (multiplane 2.0 to 8.0 MHz transducer) (Philips Healthcare). All TEE examinations were performed within 3 months before the scheduled PVI procedure (interval between TEE and PVI, 0 to 10 days [mean 2.12 days] in the group with SEC and 0 to 93 days [mean 2.77 days] in the group without SEC). Lidocaine was used for local anesthesia of the hypopharynx. All patients received intravenous midazolam for conscious sedation. Multiple planes of the LAA, including a continuous view through the LAA from 0 to 180 degrees, were examined at the appropriate level within the esophagus. Detailed observations were made of all LAA structures. Flow velocity in the left atrial appendage was assessed using pulsed-wave Doppler interrogation on TEE in the view at 0 and 90 degrees. Peak LAAFV was measured after optimally aligning the pulsed-wave Doppler signal with LAA ow using color ow imaging, with sampling done at the site where maximal ow velocities were obtained. The highest LAAFV between the two values for each patient was applied for further analysis. SEC was de ned as dynamic "smoke-like" echoes characterized by a swirling motion and observed during the cardiac cycle using an optimal gain setting [9]. The observation of echo-dense material acoustically separate from the endocardium within the LAA con rmed a de nite thrombus. Two echocardiographers blinded to the CT results interpreted all images.

Transthoracic Echocardiography
Transthoracic echocardiography was performed according to standard clinical protocol. All patients underwent TTE within the 6 months prior to PVI (interval between TEE and TTE, -6 to 178 days [mean  20.29 days] in the group without SEC). The CT scan protocol did not call for the use of beta-adrenergic blocking agents prior to scanning. Patients were administered sublingual nitroglycerine a few minutes before CT scanning to allow simultaneous evaluation of the coronary artery.
During scan acquisition, each patient received intravenous administration of contrast material (Iopamiron®, 370 mg I/mL; Bayer AG, Leverkusen, Germany). Early-phase scanning was controlled by bolus tracking in the ascending aorta and followed by injection of 30 mL of pure saline at the same speed when enhancement was achieved. Delayed-phase scanning began 60 seconds after the start of contrast material injection. The injection speed was computed using the formula: injection speed (mL/s) = body weight (kg) × 0.07 mL/s, and the volume of the iodine bolus was computed as: volume (mL) = injection speed × duration of CT data acquisition (seconds) + 5. Scan parameters were: slice collimation, 2 × 64 × 0.6 mm (SDFlash) or 2 × 32 × 0.6 mm (SD) with a z-ying focal spot; gantry rotation time, 280 ms (SDFlash) or 330 ms (SD); pitch, 0.2 to 0.5; tube voltage, 120 kVp; and tube current, 330 mAs (SDFlash) or 300 mAs (SD). Scanning ranged from the level of the carina to just below the dome of the diaphragm. We retrospectively applied electrocardiography (ECG) gating and ECG-dependent tube-current modulation to reconstruct images. All reconstructed image data were transferred to workstations (MultiModality Workplace, Siemens), and ECG gating of data of transaxial slices (effective slice width, 0.75 mm; increment, 0.4 mm; medium-smooth convolution kernel B 36 F) permitted reconstruction of images. Reconstructed images were obtained during the 1-96% -R interval of the cardiac phase. Heart rates ranged between 39 and 148 beats per minute (bpm) (mean 64.75 bpm).

Image Evaluation
Two radiologists with 9 and 25 years' experience in cardiovascular radiology who were blinded to the patient's history and TEE results retrospectively reviewed the CT examinations in consensus. In the present study, we classi ed the morphology of the LAA into 4 types according to the shape and complexity of the appendage-chicken-wing, wind-sock, cauli ower, and cactus-using 3-dimensional volume-rendered structures based on ndings of a previous study [10] (Fig. 2). The chicken-wing type displays only one lobe (length > 40 mm) and demonstrates bending of less than 100 degrees in the proximal part of the LAA; the wind-sock type shows one dominant lobe (length > 40 mm) with several secondary, or even tertiary, lobes and bending that exceeds 100 degrees; the cauli ower type is characterized by its length less than 40 mm and complex internal structures; and cactus type morphology manifests a dominant central lobe (length < 40 mm) with one or more secondary lobes.
LAA early lling defect was de ned as a clear low attenuating lesion representing incomplete mixing of the contrast agent and blood that appeared only on early-phase images and demonstrated complete homogenous enhancement on late-phase images [11] (Fig. 3).
We quanti ed the volume of the LAA from its contours as depicted in cross-sectional images obtained using SYNAPSE VINCENT® 3-dimensional analytical volume software (Fuji lm Medical Co., Tokyo, Japan) (Fig. 4). The operator contoured and lled the LAA at each slice (Figs. 4a-c, green area), and the computer added the volumes of all of the slices to calculate the LAA volume; volumes of the slices were calculated automatically by multiplying the contoured area and slice thickness in cubic centimeters (cm 3 ) (Fig. 4d). The ostium of the LAA was de ned by the plane that connected between the base of the Coumadin ridge and the proximal left circum ex artery [12]. These procedures were evaluated against the original transverse images and multiplanar reformations that included short-and long-axis views of the heart. LAA volume was indexed for body size by dividing by body surface area calculated using the DuBois formula.
One radiologist with 9 years' experience in cardiovascular radiology who was blinded to the patient's history and the results of TEE and TTE performed visual and quantitative measurement of the LAA, and two assessors independently measured the LAA volume of 100 randomly selected patients to allow examination of inter-rater reliability [13]. The inter-observer agreement regarding LAA morphology was also evaluated using Cohen's kappa [14,15].

Statistical analysis
We tested normal distribution for the continuous variables to describe the frequencies and distributions of these factors and used Fisher's exact test to compare categorical data between the groups with and without SEC and Wilcoxon rank sum test and Student's t-test to compare continuous data between the 2 groups. Categorical variables were presented as number and percentage of cases, and continuous variables were presented as mean (± standard deviation [SD]) or median (interquartile range).
We prepared multivariable logistic-regression models to estimate the incidence of SEC with potential predictors that we selected based on previous studies and clinical perspectives associated with the appearance of SEC or LAA thrombus [5][6][7]. Before the multivariate logistic regression analysis, we analyzed statistics using the variance in ation factor (VIF) to check for multicollinearity and selected AF type (paroxysmal/persistent), CHADS 2 score, LVEF, LAAFV, indexed LAA volume, LAA morphology, and LAA early lling defect as independent variables. We assessed the odds ratios (OR) for each variable with 95% con dence intervals (CI).
A CHADS 2 score ranging from 0 to 6 was calculated for each patient at the time of TEE, assigning one point each to congestive heart failure (CHF), hypertension, or age above 75 years and two points each to history of stroke, transient ischemic attack (TIA), or systemic embolism [5,16].
We classi ed AF as either paroxysmal, de ned as recurrent AF terminating spontaneously or with intervention within 7 days of onset, or persistent, de ned as AF failing to self-terminate within 7 days and considered longstanding when it failed to resolve after more than 12 months [17].
We also computed a receiver operator characteristic (ROC) curve for the completed model, choosing a threshold value at which the likelihood of SEC could be predicted based on the indexed LAA volume and estimating the resulting sensitivity and speci city.
We used intraclass correlation coe cients (ICC) to calculate inter-rater reliability for measurements of LAA volume (ICC 2, 1), and based on the 95% con dence interval of the ICC estimate, we judged reliability to be poor (values below 0.5), moderate (between 0.5 and 0.75), good (between 0.75 and 0.9), and excellent (above 0.90) [18].
We used κ-statistics to measure inter-rater reliability regarding the assessment of LAA morphology, with values below 0.2 representing slight agreement, between 0.21 and 0.40, fair agreement, 0.41 and 0.60, moderate agreement, 0.61 and 0.80, substantial agreement, and above 0.81, almost perfect agreement [19].
A P-value below 0.05 was considered statistically signi cant.

Page 7/19
The baseline characteristics of the patients are summarized in Table 1. As Compared to patients without SEC, those with SEC demonstrated signi cantly larger mean indexed LAA volume, decreased mean LAAFV and LVEF, a trend toward greater incidence of LAA early lling defect, and greater likelihood of history of persistent AF. There was no statistically signi cant difference in the other variables between the 2 groups.  We generated a ROC curve to identify threshold values for indexed volume at which SEC was most likely to appear (Fig. 5a) and observed the curve's in ection points at 8.04 cm 3 /m 2 (area under the curve (AUC), 0.642; sensitivity, 75.0%, speci city, 48.7%) for indexed LAA volume. We also determined if the addition of LAAFV parameter to a logistic-regression models that included indexed LAA volume improved SEC prediction using the DeLong method. When LAAFV was added to indexed LAA volume, the AUC increased from 0.642 to 0.724 (P < 0.001) (Fig. 5b).

Discussion
Our ndings support persistent AF and decreased FV, increased volume, and early lling defects of the LAA as independent predictors of SEC. Our ROC curve suggested an increased likelihood of SEC when the indexed volume exceeded 8.04 cm 3 /m 2 . The combination of LAA volume and LAAFV signi cantly improved the accuracy of SEC prediction.
Decreased LAAFV, increased LAA volume and persistent AF were independent predictors of SEC. The presence of SEC in the LAA has been associated with a high risk of thromboembolic events and may be the preceding stage to thrombus formation [3,4]. Its appearance is thought to re ect increased erythrocyte aggregation caused by low shear rate due to altered atrial ow dynamics and uncoordinated atrial systole [20,21]. Virchow's triad of factors related to thrombus formation includes abnormal changes in blood ow, constituents, and vessel walls. SEC may ful ll the rst two components for thrombogenesis [22]. Spontaneous echocardiographic contrast in patients with AF has been associated with reduced LAA ejection velocity and ndings of left atrial enlargement on transesophageal echocardiography (TEE) [23,24]. Reduced LAAFV measured during TEE is a well-established risk factor for the development of SEC and thromboembolism. Magnetic resonance imaging (MRI) studies have shown larger mean LAA volumes in patients with a history of AF and stroke than those without stroke [25]. As well in our study, increased LAA volume was independent predictors of SEC and combination of LAA volume and LAAFV strengthened the accuracy of SEC prediction. Restriction of contractile function and ow velocity within the LAA can lead to thrombus formation [26]. It is reasonable to consider that abnormal blood ow affected by persistent AF could more likely lead to the development of SEC in the LAA in the same way as the thrombus develops, and patients with persistent or permanent AF have been more likely to occur embolization than those with paroxysmal AF [27][28][29][30]. Chicken-wing LAA has been reported to have the highest LAA-emptying ow velocity and lowest risk for the development of SEC, transient ischemic attack, and stroke [6,7]. In contrast, we did not demonstrate the type of LAA morphology as a signi cant predictor for SEC. LAA morphology was originally divided into four types to help in the practical planning of placement of transcatheter LAA closure devices [31], but the anatomical features of LAA are actually more complicated. More rigorous de nitions for the types of LAA might be required to assess the relationship between SEC and LAA morphology.
Depiction of lling defects in the LAA is challenging in early-phase CT. It has not been fully comprehend their signi cance in the absence of thrombus. It has been considered that inadequate blending of the contrast medium with blood or the encumbrance of contrast lling by the bulky pectinate muscles contribute to lling defects in the LAA [32,33]. Our results showed that the lling defects observed re ect slow-ow states that correlate with SEC. The incidental discovery of early LAA lling defects has increased with the use of cardiac CT, and physicians should be mindful that such ndings might indicate the presence of SEC.
Our ndings generate several hypotheses with potential implications for clinical decision-making. Identi cation of predictors of SEC on cardiac CT ndings might allow noninvasive estimation of the risk of SEC which is associated with risk for thromboembolism. Although CT does not permit assessment of hemodynamics and function, CT has proven a safer, faster, and less invasive tool for the visualization and measurement of the LAA than TEE, the modality commonly used to assess LAA function [34]. Current guidelines for the management of atrial brillation recommend the scoring of CHADS 2 or CHA 2 DS 2 -VASc to estimate their risk of stroke and determine whether they would bene t from anticoagulation therapy [5,35]. However, the prediction of stroke risk can be clinically perplexing in patients with lower CHADS scores. Although, to date, no guidelines have been put forward to guide the treatment of patients with SEC, the presence of SEC in the LAA has been associated with a high risk of thromboembolic events [3,4]. Our results did not show the CHADS2 score alone to be a signi cant predictor of SEC, which suggests that patients with SEC could have high risk of LAA thrombus independently of CHADS 2 score. Various clinical factors independently of the current CHADS 2 /CHA 2 DS 2 -VASc scoring, including SEC could determine the risk of thromboembolic events in patients with AF. The primary limitation to detect the presence of SEC in the LAA is that it requires TEE which is semi-invasive and requires conscious sedation.
We expect that in patients with cardiac CT ndings that predict SEC in the LAA, subsequent TEE evaluation is needed for its de nitive identi cation and to assess the need or bene t of anticoagulant therapy. Validation of this hypothesis will involve prospective studies of to evaluate the relationship between SEC and actual thromboembolic events.
In this study, we measured LAA volume by manual segmentation of CT data and cursor tracing and observed excellent inter-rater reliability regarding LAA volume and morphology, though results using manual methods can be highly variable and create measurement bias. Indeed, the gold standard for image segmentation is that of manual labeling by a human expert [36]. However, automated segmentation to optimize LAA volume measurement is expected to address the shortcomings of this time-consuming method.

Limitations
Our study was limited by its retrospective and observational approach. In addition, the make-up of our study population, patients with relatively low risk of stroke (mean CHADS 2 score, 0.88), re ects selection bias and does not permit the adaptation of our ndings to a population with high stroke risk. However, our data may indicate factors to be considered in risk strati cation in patients with AF with low CHADS 2 scores. As well, anticoagulant therapy in patients who underwent PVI for AF led to thrombus resolution and a lower incidence of stroke. However, the use of anticoagulants does not affect the presence of SEC because it does not change red-cell aggregation [37,38]. In this study, the speci city of threshold for the indexed LAA volume for SEC was not high. It's because the lack of uni cation of the cardiac phase of the reconstructed images might affect measurement results of LAA volume, especially in patients with paroxysmal AF. The retrospective study design did not allow us to evaluate the relationship between factors associated with SEC and actual thromboembolic events, nor did it allow us to investigate correlation between cardiac phase in reconstructed images and its effect on LAA volume in patients with paroxysmal AF. Future prospective studies should address these shortcomings and examine these relationships.

Conclusions
We observed LAA ndings of cardiac CT including increased volume and early lling defects as independent predictors of SEC in patients with AF. Cardiac CT ndings might allow the noninvasive estimation of SEC and additional information for risk strati cation and management of thromboembolic events in patients with AF.   Methods of measuring the volume of the left atrial appendage (LAA) and representative case, a 53-yearold man with an LAA lling defect. LAA volume was quanti ed by contouring cross-sectional images using 3-dimensional analytical volume software, SYNAPSE VINCENT® (Fuji lm Medical Co., Tokyo, Japan). The operator contoured and lled the LAA at each slice (a-c, green area), and the computer calculated the LAA volume in cubic centimeters (cm3).