Left ventricular mass estimation by real-time 3D echocardiography favourably competes with CMR in congenital left ventricular disease

Assessment of left ventricular mass (LVM) is important in the evaluation of patients with congenital heart disease (CHD) and cardiac magnetic resonance imaging (CMR) is the gold standard. Recent software allows LVM calculation by real-time 3-dimensional echocardiography (RT3DE). We investigated the impact of different software analysis tools on LVM determination by CMR or RT3DE in a cohort of patients with heterogeneous left ventricular (LV) disease. 37 subjects (17 patients, mean age 18.7 y; 20 controls, mean age 13.2 y) underwent CMR and RT3DE. CMR LVM and RT3DE calculations were done using two different LV-analysis software packages for each modality: CMR i) customized software “CMR HDZ”, CMR ii) “CMR ISP”; RT3DE i) “Toshiba”, RT3DE ii) “Tomtec”, 4D LV-Analysis Version 3.1 (built 3.1.0.258661). Intra- and interobserver variabilities were calculated. Only RT3DE-derived LVM showed significant software-dependent differences. RT3DE-derived LVM (both softwares) was significantly higher than CMR-derived LVM (both softwares). The two different methods and four evaluation software packages for LVM assessment were well correlated with each other. Intra- and interobserver variability of LVM as assessed by each single modality or software was low. Despite software dependency and overestimation of RT3DE-assessed LVM by 5 to 10%, RT3DE still competes with the gold standard, CMR, even in patients with various forms of LV disease. The use of optimized software, especially for RT3DE, should improve the accuracy of LVM assessment, overcoming LVM overestimation.


Results
Due to poor image quality or incomplete volume capture in RT3DE studies, 2 patients (female, 11 years [y], double outlet right ventricle with pulmonary stenosis; female, 2 months, double aortic branch) and 1 control had to be excluded from evaluation, resulting in 17 patients (mean age ± standard deviation (SD) = 18.7 ± 16.2 y; range 0.1 to 72.9 y, 7 female) and 20 healthy individuals (13.2 ± 3.4 y; range 7.1 to 19.4 y, 9 female) included into the study. For detailed participant characteristics and diagnoses see Table 1.
All intra-class coefficients (ICC) comparing the two different methods and evaluation software packages for LVM assessment were >0.97 (Table 2).
Bland Altman statistics of the intermodality comparison showed a minimum mean overestimation of LVM and smallest intermodality limits of agreement (LOA) as assessed by RT3DE ("Toshiba") compared to "CMR HDZ" (Table 3   graphic assessment risk inaccuracy in measurement due to oblique M-Mode sampling, foreshortening of apical views, and the fact that 2D echocardiography results are estimated based on given formulas from geometry [11][12][13] . Both the training background of the examiner and the exact type or version of hardware and software used dramatically influence 2D echocardiography LVM determinations 14 20 , who included a high number of patients with left-sided heart obstructive lesions and/or small LV in addition to (corrected) obstructed (±volume load) left heart lesions, we investigated a more heterogeneous cohort with structurally normal hearts as well as with LV obstruction, right-sided obstructive and/or volume-loaded lesions, or "special" cases with a single LV or with a morphologically left subpulmonary ventricle in a biventricular circulation, thus reflecting the wide clinical spectrum of heart disease 2,20 .   www.nature.com/scientificreports www.nature.com/scientificreports/ obtained using different methods and assessment software tools. However, in contrast to CMR, where results of LVM were independent of the software tool used for analysis, in RT3DE the software used had a minor impact on LVM results. Thus, to date, the use of uniform software and the validation and optimization of assessment techniques are essential. We speculate that to use optimized hardware systems might yield an additional benefit, minimizing differences between RT3DE-and CMR-derived estimates of LVM 2 .
So far, the impact of assessment software has been examined with respect to ventricle volume measurement in children with and without CHD but only rarely with respect to ventricular mass (comparing manual with automated algorithms) 12,24,25 . With respect to efficacy, we explicitly opted for a semiautomated RT3DE analysis method shown to be timesaving compared with M-Mode and 2D echocardiography or conventional CMR evaluation software 2,20 .
Trabeculated myocardium. For the sake of homogeneity, we deviated from the previously published protocols of CMR assessment by assigning the regions of trabeculated myocardium to the LV cavity rather than to the LV wall as done by the echocardiographic software, thus rendering the results of the comparison of methods better comparable 5,26,27 . Even if trabeculated myocardium had not been strictly excluded from LVM calculation, for clinical purposes the results still would have been acceptably comparable: Laser et al. found that CMR-derived LVM increased by 5.6% if trabeculated myocardium was included to LVM calculation 2 .     26 . The reason to assess left ventricular muscle mass exclusively was that currently no such quantification tool is available for RT3DE with comparable accuracy in right ventricular muscular mass determination as has been proven for RV volumetry and ejection fraction 28 . Moreover, usually the RV wall is rather thin and the echo window is suboptimal, rendering the obtained echocardiography data difficult to process.

Clinical implications. Once trialled and validated in a larger cohort of patients with various forms of LV
disease, RT3DE LVM assessment could replace CMR examination in the strict follow-up of LVM with LV outflow tract obstruction or hypertrophic cardiomyopathy, and LVM overestimation could be overcome if used in combination with uniform and optimized software. For the LV that is atypical in shape, RT3DE LVM assessment must be systematically re-evaluated with respect to different hard-and software assessment tools. To this end the use of a model-based approach of RT3DE has been shown to be favorable in terms of time saving and clinical applicability 5 .
A clinical advantage of an easy-to-perform LVM assessment is improvement in prediction of events related to cardiovascular disease: According to the CARDIA study the Framingham risk score underestimates risks in young adults, while the risk score and LVM in combination are helpful independent predictors of cardiovascular disease 29 . Limitations. Our data are rather heterogeneous due to the small number of cases with different diagnoses, and of different age and gender; this reflects restricted access to study facilities and small numbers of patients with the same matching cardiac disease.
However, our study allows a more general assessment of methods. At the same time, our study should be regarded as a first step towards the examination of larger cohorts with more patients with the same cardiac disease. Thus, for the establishment of RT3DE in clinical routine, further intra-and interobserver studies with larger sample sizes and the establishment of reference values for different groups of congenital heart disease are mandatory. Additionally, we quantified the whole LVM but not the septum separately, and our study lacks datasets of patients with global or localized/asymmetric cardiac hypertrophy. Thus, in clinical practice, our findings and especially the transfer of our findings to patients with disease manifest as global hypertrophy, isolated septum hypertrophy, or asymmetric hypertrophy cannot be definitely judged. Feasibility is limited as excellent image quality with excellent reflections of the pericardial border and complete data sets without dropouts are mandatory. In our study, however, also due to our investigating co-operating older www.nature.com/scientificreports www.nature.com/scientificreports/ patients, with an exclusion rate < 10% image quality was sufficiently good. Additionally, in hypertrophic disease with greater echogenicity, myocardial borders might be defined more easily.

Conclusion and Outlook
Despite software dependency and overestimation of RT3DE-assessed LVM by 5 to 10 percent, RT3DE is still competitive with the gold standard, CMR, in patients with various forms of LV disease. The use of optimized software, especially for RT3DE, should improve the accuracy of LVM assessment and thereby overcome LVM overestimation.

Methods
Within a period of 2 weeks we examined 40 subjects (19 patients with different types of CHD and 21 healthy probands; Table 1), none of them requiring general anesthesia for examination. Each subject first underwent echocardiography and within the following hour CMR.

Real-Time 3-dimensional echocardiography (RT3DE).
RT3DE was performed by one experienced sonographer (KTL) using Artida (Artida 4D-System, Toshiba Medical Systems, Neuss, Germany, transducer PST25-TX). The examination consisted of a standard 3D protocol, including left-sided position of the patient, apical view with the apex next to the transducer including the entire LV portion in the 2D planning view using 4 subvolumes at 90° to 110°, and 4-beat acquisition during one end-expiratory breath-hold 2 . All RT3DE data of LVM were analyzed using two different software packages: i) "Toshiba" (Wall Motion tracking system, Toshiba Medical Systems, Neuss, Germany) and ii) "Tomtec" (Image Arena platform [Version 4.6, built 4.6.2.12; TomTec, Unterschleissheim, Germany], 4D LV-Analysis Version 3.1 [built 3.1.0.258661]). After a first automated contour-tracking process, endocardial and epicardial borders were manually re-drawn. Myocardial mass was calculated as myocardial volume between the endocardial and epicardial border multiplied by myocardial density Cardiac magnetic resonance imaging (CMR). Assessment of LVM was done using a multi-transmit 3.0 Tesla magnetic resonance imaging system (Achieva 3.0 T TX, Philips Healthcare, Best, The Netherlands). We used a published standard protocol 5,26,27 : Maximum gradient performance was 80 mT/m, slew rate 200 T/m/s. For signal detection a 32-element phased-array receive-only surface coil was used. Typically, 14-23 axial slices with no slice gap were collected applying a multi-slice multi-phase vector electrocardiogram-triggered steady-state free precession gradient-echo sequence. Pulse repetition time/echo time/excitation angle were 2.7 ms/1.35 ms/40°, www.nature.com/scientificreports www.nature.com/scientificreports/ slice thickness of 5-6 mm, matrix size of 160 × 240 encompassing a field-of-view of 384 mm. This setup results in a spatial in-plane resolution of 1.6 × 1.6 mm². By default, 25 cardiac phases were acquired under breath-holding periods of below 12 seconds duration.
Image data were transferred to external computer workstations for assessment of LVM. End-diastolic images were used to define LVM in each case by applying 2 different software packages: i) a dedicated customized CMR-Tool labelled "CMR HDZ" (HDZ MR-Tools; R2015, HDZ, Bad Oeynhausen, Germany); ii) CMR Analysis tool labelled "CMR ISP" (IntelliSpace Portal, Release v7.0.1; Philips Healthcare, Best, The Netherlands). Both software systems used threshold-based semiautomated detection of endocardial/vascular borders. Papillary muscles were assigned as an exemption to the blood pool to be comparable with the echocardiographic analysis (Figs 6, 7).
Reproducibility. Data were evaluated by two experienced observers. Analysis of RT3DE was performed blinded to CMR data, and vice versa. To define reproducibility, intra-and interobserver variability was determined in 10 randomly selected data pairs with results blinded. For intraobserver reliability testing, >2 weeks after the first analysis a blinded second analysis was done by the same observer.
Informed consent. Informed consent for study participation was obtained from each participant or from his parent and/or legal guardian in case of participants under the age of 18 y. Statistical analysis. Statistical analysis was done using the Statistical Software Package SPSS 20.0 (SPSS, Chicago, IL). The Shapiro-Wilk test was used to test LVM data on normal distribution, and Levene statistics were used to evaluate variance homogeneity. Comparison of LVM as assessed by the 2 different software analysis systems for echocardiography and CMR each and between the different modalities was performed using paired Student's t-test for normally distributed data; otherwise the Wilcoxon test was used. Agreement between CMR and echocardiographic LVM and the different software packages, as well as intra-and interobserver variability, was calculated using the ICC and by Bland-Altman-analysis 30 . A p-value of < 0.05 was considered sufficient for clinical significance.
Graphics. Graphics were created using Microsoft Office Professional Plus 2010 (Excel and Power Point) and using Metapost, a free interpreter software by John D. Hobby, version 1.211, distributed within the TEX framework via the public domain Comprehensive TEX Archive Network.
Ethical approval and informed consent. The study protocol was approved by the local ethics committee (ethics committee of the Medical faculty of the Ruhr University Bochum, section Bad Oeynhausen, registration number AZ 9/2011), the methods were carried out in accordance with the relevant guidelines and regulations, and the subjects gave written informed consent.