Abstract

Background:

Myocardial strain imaging is a robust tool for evaluating extra-valvular remodelling in aortic stenosis (AS). Multi-phase cardiac computed tomography (CT) angiography acquired for transcatheter aortic valve implantation (TAVI) planning enables a novel three-dimensional (3D), geometry-independent strain assessment beyond two-dimensional (2D) transthoracic echocardiography (TTE). This study evaluates the agreement and reproducibility of CT- and TTE-derived longitudinal strain and examines its association with pulmonary hypertension (PH) in significant AS.

Methods:

Left ventricular global longitudinal strain (LV-GLS), left atrial global longitudinal reservoir strain (LA-LS), right ventricular global longitudinal strain (RV-GLS), and RV free-wall longitudinal strain (RV-FWLS) were determined using 2D TTE and CT-based 3D motion tracking in patients with severe AS undergoing TAVI evaluation. Patients with PH were defined using guideline-directed TTE criteria for high probability PH (H-PH: n = 43, 46.2%) and compared with those at low probability (L-PH: n = 50, 53.8%).

Results:

Agreement between CT and TTE was strong for LV-GLS (r = 0.837), RV-GLS (r = 0.853), and RV-FWLS (r = 0.780) and moderate for LA-LS (r = 0.677) (all p < 0.001). Peak longitudinal strain on both TTE and CT was significantly reduced in H-PH compared with L-PH (p < 0.001). Optimal strain cutoff values for identifying H-PH were lower on CT than on TTE (LV-GLS: −16.6% vs. −17.9%; LA-LS: 10.2% vs. 14.4%; RV-GLS: −15.3% vs. −20.1%; RV-FWLS: −18.0% vs. −21.1%). In an inter-modality comparison, it was found that TTE-derived LV-GLS was superior to CT-derived LV-GLS for detecting H-PH [AUC: 0.94 [95% CI 0.89–0.99] vs. 0.85 [95% CI 0.78–0.93], p = 0.013], whereas differences for LA-LS, RV-GLS, and RV-FWLS were non-significant (all p > 0.05). TTE- and CT-derived strain measurements showed excellent reproducibility (ICC > 0.9).

Conclusion:

TAVI CT is a promising tool for 3D longitudinal strain assessment and a valuable adjunct to TTE for quantifying extra-valvular remodelling associated with AS progression to PH. Further studies are warranted to evaluate the prognostic value of multi-chamber CT-derived 3D strain in AS.

1 Introduction

Aortic stenosis (AS) is the most common valvular heart disease (VHD) in developed countries, which is now predominantly treated with transcatheter aortic valve implantation (TAVI) (1). Improved detection of structural and functional cardiopulmonary abnormalities associated with AS progression can enhance risk stratification and determine the optimal timing of intervention (2). Myocardial strain imaging integrates both volumetric and geometric cardiac properties and is increasingly recognised as a tool for a more precise assessment of extra-valvular remodelling. Global longitudinal strain has emerged as a sensitive biomarker for detecting early deformation abnormalities in AS, preceding overt reduction in ejection fraction (EF) (3). Moreover, impaired longitudinal strain of the left ventricle (LV), left atrium (LA), and right ventricle (RV) are associated with increased mortality in patients with AS undergoing TAVI (4).

Owing to its widespread availability and lack of ionising radiation, two-dimensional (2D) speckle-tracking transthoracic echocardiography (TTE) is the most common technique for performing a routine strain analysis. However, TTE variability can be substantial due to operator experience, vendor-specific settings, suboptimal image quality, apical foreshortening, and angle of incidence effect. More recently, strain analysis has become feasible with the use of 2D reconstructed planes from four-dimensional (4D) retrospective electrocardiography (ECG)-gated cardiac computed tomography angiography (CT) (5). To overcome the inherent limitations of 2D strain analysis in anatomically complex myocardial chambers, whole-chamber CT-derived three-dimensional (3D) strain can provide a truly geometry-independent characterisation of tissue biomechanics (6–8). Strain measurement on CT benefits from high isotopic spatial resolution and a wide field of view, allowing improved endocardial border visualisation and reproducible measurements. Given that CT is the gold standard for TAVI planning, extraction of additional strain parameters could enhance its value for risk stratification and outcome prediction without additional imaging, contrast administration, or radiation exposure.

Data comparing multi-chamber CT-derived 3D strain with conventional 2D TTE remain scarce. Furthermore, there is limited evidence on strain alterations in AS patients with pulmonary hypertension—a clinically important subgroup at increased risk of adverse outcomes following TAVI (9). Against this background, the primary objective of the present study is to determine the correlation, agreement, and reproducibility of 3D CT- vs. 2D TTE-derived strain for the LV, LA, and RV in patients with severe AS awaiting TAVI. The secondary objective is to compare the associations of 3D CT- and 2D TTE-derived strain with PH in AS.

2 Materials and methods2.1 Study design and population

This was a prospective cohort study of adult patients (>18 years) with severe symptomatic AS undergoing TAVI evaluation. All participants underwent clinically indicated TTE and CT on the same day. Severe AS was defined as an aortic valve area (AVA) < 1 cm2 (or an indexed AVA ≤0.6 cm2/m2) and/or a mean aortic valve gradient ≥40 mmHg and/or a peak aortic jet velocity ≥4 m/s, according to current guidelines (10). PH was defined according to TTE criteria, using peak tricuspid regurgitation jet velocity (TR Vmax) and additional parameters suggestive of PH, in line with guideline recommendations (11). Patients were divided into two groups for comparison: high probability of PH (H-PH) and low probability of PH (L-PH). The exclusion criteria were as follows: at least moderate VHD other than AS, previous valve surgery, implanted permanent pacemaker, significant lung disease, left ventricular EF (LV-EF) < 50%, significant coronary artery disease, prior myocardial infarction, insufficient TR to estimate pulmonary artery systolic pressure, and poor image quality. Baseline clinical characteristics, serum biomarkers, and standard TTE parameters were obtained at the time of recruitment. The study protocol complied with the declaration of Helsinki and was approved by the National Ethics Review Board (Integrated Research Application System Reference: 319698). The final study population consisted of 93 patients recruited between June 2023 and November 2024. All patients provided written informed consent.

2.2 TAVI CT acquisition

CT image acquisition was performed over the entire cardiac cycle using a dual-source scanner (SOMATOM Force, Siemens Healthcare) and a dedicated research protocol (12). Scan parameters were as follows: reference tube voltage 100–120 kV and reference tube-current-time product 125–300 mAs, according to body mass index (BMI), gantry rotation time 250 ms, slice collimation 128 × 0.6 mm, and pitch value 0.2–0.45, depending on heart rate. A triphasic contrast bolus consisted of 50 mL of contrast (5 mL/s), 100 mL of contrast-saline mix (1:1) (5 mL/s), and 50 mL of saline flush (3 mL/s). Data acquisition was initiated when the threshold of 110 Hounsfield units was reached in the proximal descending aorta, followed by a 5 s delay. Spiral acquisition was performed during an inspiratory breath hold in a cranio-caudal direction with no tube current modulation. Images were reconstructed with a slice thickness of 1 mm in 5% increments to obtain a total of 20 reconstructions per cardiac cycle (0%–100% of RR interval) for feature tracking.

2.3 Measurement of 3D CT-derived myocardial strain and volume

Multi-phase TAVI CT was used to evaluate 3D myocardial strain according to previously validated methodology (Supplementary Methodology in the Supplementary Material) (6, 7). Patient-specific models of myocardial chambers were semi-automatically constructed from CT images at end-diastole using CemrgApp, with manual corrections to endocardial borders if necessary (13, 14). Anatomical landmarks were selected on end-diastolic images of the LV, LA, and RV to define longitudinal axes. Cardiac model deformation was automatically tracked throughout the cardiac cycle using 3D feature tracking (Figure 1). Longitudinal strain was defined as the percentage change in length at peak systole along the longitudinal axes and tangential to the endocardial surface, with ventricular end-diastole used as the zero reference. CT-derived peak longitudinal strain was measured for: (1) global LV (LV-GLSCT); (2) global LA (LA-LSCT); (3) global RV (RV-GLSCT); and (4) RV free-wall (RV-FWLSCT). 3D end-diastolic volume (EDV), end-systolic volume (ESV), and EF were calculated from myocardial chamber segmentations at end-diastole (0%) and peak systole.

Illustration of multi-chamber CT-derived strain and TTE-derived volume and strain assessment. Longitudinal strain was measured using ventricular end-diastole as the reference point on both CT and TTE. For TTE, LV strain was averaged from apical four-, three-, and two-chamber views, whereas RV strain was assessed from the RV-focused apical four-chamber view. A full-volume dataset for the LV and RV was acquired from apical four-chamber and RV-focused views to quantify EF, whereas LA volume was measured at ventricular end-systole using the biplane method. CT, computed tomography; EF, ejection fraction; LA, left atrium; LV, left ventricle; RV, right ventricle; TTE, transthoracic echocardiography.

2.4 TTE acquisition and analysis

All TTE images were acquired using an EPIQ CVx v7.0 system (Philips Healthcare, Netherlands) and X5–1c transducer. A 2D strain analysis was performed semi-automatically using TomTec Arena (2D Cardiac Performance Analysis, Tomtec Imaging Systems GmbH, Unterschleißheim, Germany) according to current recommendations, with the reference point placed at end-diastole (15). LV strain was measured from apical four-, three-, and two-chamber views; RV strain from the RV-focused apical four-chamber view; and LA strain from apical four- and two-chamber views (Figure 1). Care was taken to avoid apical foreshortening and achieve the highest frame rate possible (minimum ≥60 frames/s). Endocardial borders were reviewed at end-diastole and end-systole and manually corrected if necessary. In sinus rhythm, strain was measured during a single cardiac cycle selected from three-beat cine loops with similar RR intervals. In atrial fibrillation (AF), strain was averaged over three separate cardiac cycles. TTE-derived peak longitudinal strain was measured in absolute values, with ventricular end-diastole as the zero reference point, for: (1) global LV (LV-GLSTTE); (2) global LA (LA-LSTTE); (3) global RV (RV-GLSTTE); and (4) RV free-wall (RV-FWLSTTE).

Full-volume 3D datasets of the LV and RV were acquired during a breath hold from apical four-chamber and RV-focused views, respectively (16). Images were optimised to include the entire ventricular cavity within the highest possible frame rate scan volume. Volume measurements were performed semi-automatically using TomTec Arena (Dynamic HeartModel and 3D Auto RV, Tomtec Imaging Systems GmbH, Unterschleißheim, Germany). Ventricular endocardial borders were identified at end-diastole and end-systole and manually adjusted if necessary to calculate 3D EDV, ESV, and EF. In sinus rhythm, datasets were acquired using a minimum of four-beat full-volume mode. In AF, measurements in single-beat high-volume rate mode were averaged over three different datasets. LA volume was measured at end-systole using the biplane method (Figure 1).

2.5 Statistical analysis

Variables were assessed for normality using Q–Q plots and the Shapiro–Wilk test. Continuous variables were reported as mean ± standard deviation or median (interquartile range) and categorical variables as frequencies (%). Between-group differences were assessed using Student's t-test or the Mann–Whitney U test for continuous variables and the chi-squared or Fisher's exact test for categorical variables, as appropriate. The receiver operating characteristic (ROC) curve analysis assessed the ability of strain measurements to discriminate H-PH, and optimum strain cut-offs were obtained using the Youden index. Areas under the curve (AUCs) with corresponding confidence intervals (CIs) were compared with DeLong's test. Correlation and agreement between CT and TTE measurements were assessed using the Pearson correlation coefficient and Bland–Altman analysis, respectively. Intra- and inter-observer reproducibility of strain measures was assessed using the intra-class correlation coefficient (ICC). Statistical significance was defined as p < 0.05. All analyses were performed using R 4.4.0 (R Foundation for Statistical Computing, Vienna, Austria).

3 Results3.1 Study population

Table 1 describes the clinical characteristics of 93 study patients, comprising 43 (46.2%) with H-PH and 50 (53.8%) with L-PH. Patients with H-PH had a higher prevalence of AF, elevated N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels, more frequent New York Heart Association functional class ≥ III, shorter six-minute walk test (6MWT) distance, lower quality of life on the Kansas City Cardiomyopathy Questionnaire (KCCQ), and greater frailty (all p < 0.05). On TTE, patients with H-PH had a smaller AVA, lower LV-EF, reduced RV function, larger atria, higher LV filling pressures (E/e') and, as expected, greater TR Vmax (all p < 0.05). The total dose-length product (DLP) was significantly higher in H-PH than in L-PH [933 [774, 1,230] mGycm vs. 728 [606, 1,028] mGycm, p = 0.019]. There was no significant inter-group difference in heart rate at the time of CT acquisition.

CharacteristicsAll patients (n = 93)L-PH (n = 50)H-PH (n = 43)p-valueAge (years)82.4 ± 5.681.5 ± 5.083.5 ± 6.10.100Gender (female)30 (32%)19 (38%)11 (26%)0.201BMI (kg/m2)27.2 ± 5.926.7 ± 5.727.9 ± 6.20.332Hypertension74 (80%)41 (82%)33 (77%)0.531Hypercholesteremia58 (62%)33 (66%)25 (58%)0.435Diabetes mellitus34 (37%)22 (44%)12 (28%)0.108Current/ex-smoker32 (34%)20 (40%)12 (28%)0.221AF/flutter24 (26%)2 (4.0%)22 (51%)<0.001*NYHA ≥ III44 (47%)12 (24%)32 (74%)<0.001*6MWT (m)191 (122, 281)239 (174, 304)136 (96, 202)<0.001*CFS ≥443 (46%)15 (30%)28 (65%)<0.001*Katz <622 (24%)6 (12%)16 (37%)0.004*KCCQ40 (27, 59)49 (31, 70)35 (23, 40)0.002*eGFR (mL/min/1.73m2)63 ± 1967 ± 1758 ± 210.019*NT-proBNP (ng/L)1,107 (520, 2,331)619 (429, 1,290)1,978 (1,087, 4,040)<0.001*AVA (cm2)0.65 ± 0.160.68 ± 0.170.61 ± 0.130.023*Simpson's LV-EF (%)62.3 ± 6.166.6 ± 4.557.2 ± 2.7<0.001*LV mass (g)226 (190, 279)217 (165, 265)243 (199, 305)0.018*LA area (cm2)27 (22, 32)24 (19, 27)30 (27, 36)<0.001*E/e'17 (13, 20)15 (11, 18)18 (15, 22)0.002*PASP (mmHg)36 (32, 64)32 (26, 36)64 (61, 75)<0.001*PV acceleration time (ms)103 ± 22120 ± 984 ± 16<0.001*TR Vmax (m/s)2.80 (2.38, 3.38)2.44 (2.23, 2.78)3.39 (3.28, 3.78)<0.001*TAPSE (mm/mmHg)20.4 ± 5.423.7 ± 3.616.6 ± 4.6<0.001*RV FAC (%)40 ± 1046 ± 632 ± 9<0.001*RV basal diameter (mm)4.52 ± 0.744.06 ± 0.665.06 ± 0.37<0.001*RA area (cm2)20 (16, 25)16 (14, 18)25 (22, 33)<0.001*IVC diameter (mm)1.83 ± 0.721.19 ± 0.282.55 ± 0.22<0.001*Heart rate (bpm)68.6 ± 6.8968.1 ± 5.9768.2 ± 7.660.324Total DLP (mGycm)856.9 (640, 1,137)728 (606, 1,028)933 (774, 1,230)0.019*

Baseline characteristics.

AF, atrial fibrillation; AVA, aortic valve area; BMI, body mass index; CFS, clinical frailty score; DLP, dose-length product; eGFR, estimated glomerular filtration rate; ESV, end-systolic volume; FAC, fractional area change; H-PH, high probability of pulmonary hypertension; IQR, interquartile range; IVC, inferior vena cava; KCCQ, Kansas City cardiomyopathy questionnaire; LA, left atrium; L-PH, low probability of pulmonary hypertension; LV, left ventricle; LV-EF, left ventricular ejection fraction; NT-proBNP, N-terminal pro B-type natriuretic peptide; NYHA, New York Heart Association functional class; PASP, pulmonary artery systolic pressure; PV, pulmonary valve; RA, right atrium; TAPSE, tricuspid annular plane systolic excursion; RV, right ventricle; SD, standard deviation; TR, tricuspid regurgitation; 6MWT, six-minute walk test.

Data are reported as mean ± SD, median (IQR), or frequency (%). The p-value refers to the comparison between L-PH and H-PH groups.

3.2 Agreement of CT and TTE

Table 2 presents paired comparisons of longitudinal strain, myocardial volumes, and EF measurements obtained using CT and TTE. CT-derived measures of LV-GLS, LA-LS, RV-GLS, and RV-FWLS were significantly lower in magnitude than the corresponding TTE values (all p < 0.001). Strong inter-modality correlations were observed for LV-GLS (r = 0.837), RV-GLS (r = 0.853), and RV-FWLS (r = 0.780), whereas the correlation for LA-LS was moderate (r = 0.677) (all p < 0.001) (Figure 2). Among all strain parameters, RV-FWLS showed the greatest under-estimation bias on CT compared with TTE [mean bias: −4.78%; 95% limit of agreement (LoA): 2.26 and −11.8%].

ParameterCT (n = 93)TTE (n = 93)p-valueStrainLV-GLS (%)−15.6 ± 6.3−17.2 ± 3.60.029*LA-LS (%)11.0 ± 5.015.0 ± 9.0<0.001*RV-GLS (%)−16.8 ± 5.0−20.3 ± 4.1<0.001*RV-FWLS (%)−18.0 ± 5.7−22.8 ± 4.6<0.001*Volumes and EFLV-EDV (mL)128 ± 25116 ± 22<0.001*LV-ESV (mL)50 ± 1345 ± 100.013*LV-EF (%)59 (55, 67)60 (57, 64)0.933LA-ESV (mL)138 (112, 173)90 (67, 113)<0.001*RV-EDV (mL)156 (135, 196)144 (123, 177)0.012*RV-ESV (mL)64 (49, 107)63 (46, 95)0.438RV-EF (%)58 (45, 65)56 (43, 63)0.223

Comparison of longitudinal strain, myocardial volume, and ejection fraction measurements using CT and TTE.

CT, computed tomography; EDV, end-diastolic volume; EF, ejection fraction; ESV, end-systolic volume; IQR, interquartile range; LA, left atrium; LA-LS, left atrial longitudinal reservoir strain; LV, left ventricle; LV-GLS, left ventricular global longitudinal strain; RV, right ventricle; RV-FWLS, right ventricular free-wall longitudinal strain; RV-GLS, right ventricular global longitudinal strain; SD, standard deviation; TTE, transthoracic echocardiography.

Data are reported as mean ± SD or median (IQR). The p-value refers to the comparison between TTE and CT measurements.

Pearson correlation coefficients (i) and Bland–Altman (ii) analyses for LV-GLS (A), LA-LS (B), RV-GLS (C), and RV-FWLS (D) between 3D CT and 2D TTE. Inter-modality correlation coefficients (r) are reported alongside their corresponding p-values. The mean inter-modality difference (bias) and LoA are shown in the dashed lines. CI, confidence interval; CT, computed tomography; LA-LS, left atrial longitudinal reservoir strain; LoA, limit of agreement; LV-GLS, left ventricular global longitudinal strain; RV-GLS, right ventricular global longitudinal strain; RV-FWLS, right ventricular free-wall longitudinal strain; TTE, transthoracic echocardiography.

With respect to myocardial volumes, TTE significantly underestimated LV-EDV, LV-ESV, LA-ESV, and RV-EDV compared with CT (all p < 0.05), while no significant differences were observed in LV-EF and RV-EF (p > 0.05). Strong inter-modality correlations were observed for LV-EF, LV-EDV, and LV-ESV (r≥0.920, p < 0.001), for RV-EF, RV-EDV, and RV-ESV (r≥0.950, p < 0.001 for all), and for LA-ESV (r = 0.855, p <0.001) (Supplementary Figures 1–S3). Among volumetric parameters, LA-ESV showed the greatest under-estimation bias on TTE compared with CT (mean bias: −50.7 mL; 95% LoA: 6.6 and −108.0 mL) (Supplementary Figure 3).

3.3 Association of myocardial strain with H-PH in AS

Supplementary Tables 1, 2 present detailed comparisons of myocardial strain, volume, and EF measurements obtained using TTE and CT in AS patients with L-PH and H-PH. All longitudinal strain parameters on TTE were significantly reduced in H-PH compared with L-PH: LV-GLSTTE (−19.6 ± 2.4% vs. −14.5 ± 2.5%, p < 0.001), LA-LSTTE (20 ± 8% vs. 10 ± 8, p < 0.001), RV-GLSTTE (−22.5 ± 3.2% vs. −17.7 ± 3.6%, p < 0.001), and RF-FWLSTTE (−25.8 ± 3.1% vs. −19.3 ± 3.5%, p < 0.001) (Figure 3). Similarly, all CT-derived longitudinal strain values were significantly lower in H-PH than in L-PH: LV-GLSCT (−18.9 ± 5.6% vs. −11.5 ± 4.5%, <0.001), LA-LSCT (13.6 ± 3.8% vs. 7.7 ± 4.7%, p < 0.001), RV-GLSCT (−19.4 ± 4.2% vs. −13.7 ± 4.0%, p < 0.001), and RV-FWLSCT (−21.1 ± 5.1% vs. −14.4 ± 4.1%, p < 0.001) (Figure 4).

Comparison of 2D TTE-derived longitudinal strain parameters between patients with H-PH and L-PH. Box-and-whisker plots: the box length represents the IQR; the horizontal box line represents the median; the whiskers denote the maximum and minimum values excluding outliers (shown as dots); *** denotes p < 0.001. H-PH, high probability of pulmonary hypertension; IQR, interquartile range; LA-LS, left atrial longitudinal reservoir strain; L-PH, low probability of pulmonary hypertension; LV-GLS, left ventricular global longitudinal strain; RV-GLS, right ventricular global longitudinal strain; RV-FWLS, right ventricular free-wall longitudinal strain; TTE, transthoracic echocardiography; 2D, two-dimensional.

Comparison of 3D CT-derived longitudinal strain parameters between patients with H-PH and L-PH. Box-and-whisker plots: the box length represents the IQR; the horizontal box line represents the median; the whiskers denote the maximum and minimum values excluding outliers (shown as dots); *** denotes p < 0.001. CT, computed tomography; H-PH, high probability of pulmonary hypertension; IQR, interquartile range; LA-LS, left atrial longitudinal reservoir strain; L-PH, low probability of pulmonary hypertension; LV-GLS, left ventricular global longitudinal strain; RV-GLS, right ventricular global longitudinal strain; RV-FWLS, right ventricular free-wall longitudinal strain; 3D, three-dimensional.

ROC analyses were performed to assess the discriminatory ability of TTE- and CT-derived longitudinal strain to detect H-PH in severe AS (Supplementary Table 3). In a head-to-head inter-modality comparison, it was found that LV-GLSTTE was superior to LV-GLSCT for detecting H-PH [AUC: 0.94 [95% CI 0.89–0.99] vs. 0.85 [95% CI 0.78–0.93], p = 0.013], whereas differences for LA-LS, RV-GLS, and RV-FWLS were non-significant (all p > 0.05) (Figure 5). The optimal longitudinal strain cut-off values for identifying H-PH were consistently lower in magnitude on CT than on TTE: LV-GLS (−16.4% vs. −17.9%), LA-LS (10.2% vs. 14.4%), RV-GLS (−15.3% vs. −20.1%), and RV-FWLS (−18.0% vs. −21.2%) (Table 3). In sub-analyses of volume-based measures, the associations of LV-EF, RV-EF, and LA-ESV with H-PH did not differ significantly between CT and TTE (all p < 0.05) (Supplementary Table 4).

Receiver operating characteristic analyses for detecting H-PH using LV-GLS (A), LA-LS (B), RV-GLS (C), and RV-FWLS (D) from 2D TTE (blue) and 3D CT (red). Receiver operating characteristic curve analyses are reported as the AUC with the corresponding 95% CIs and p-values. AUC, area under the curve; CI, confidence interval; CT, computed tomography; H-PH, high probability of pulmonary hypertension; LA-LS, left atrial longitudinal reservoir strain; LV-GLS, left ventricular global longitudinal strain; RV-GLS, right ventricular global longitudinal strain; RV-FWLS, right ventricular free-wall longitudinal strain; TTE, transthoracic echocardiography; 2D, two-dimensional; 3D, three-dimensional.

Strain parameterCut-off (%)Specificity (95% CI)Sensitivity (95% CI)2D TTELV-GLS−17.90.840 (0.740–0.940)0.954 (0.884–1.000)LA-LS14.40.820 (0.700–0.920)0.861 (0.767–0.954)RV-GLS−20.10.800 (0.680–0.900)0.814 (0.698–0.930)RV-FWLS−21.20.732 (0.620–0.831)0.977 (0.930–1.000)3D CTLV-GLS