Cardiac Magnetic Resonance Datasets Research Articles

Evaluating semi-supervision methods for medical image segmentation: applications in cardiac magnetic resonance imaging.

Neural networks have potential to automate medical image segmentation but require expensive labeling efforts. While methods have been proposed to reduce the labeling burden, most have not been thoroughly evaluated on large, clinical datasets or clinical tasks. We propose a method to train segmentation networks with limited labeled data and focus on thorough network evaluation. We propose a semi-supervised method that leverages data augmentation, consistency regularization, and pseudolabeling and train four cardiac magnetic resonance (MR) segmentation networks. We evaluate the models on multiinstitutional, multiscanner, multidisease cardiac MR datasets using five cardiac functional biomarkers, which are compared to an expert's measurements using Lin's concordance correlation coefficient (CCC), the within-subject coefficient of variation (CV), and the Dice coefficient. The semi-supervised networks achieve strong agreement using Lin's CCC ( ), CV similar to an expert, and strong generalization performance. We compare the error modes of the semi-supervised networks against fully supervised networks. We evaluate semi-supervised model performance as a function of labeled training data and with different types of model supervision, showing that a model trained with 100 labeled image slices can achieve a Dice coefficient within 1.10% of a network trained with 16,000+ labeled image slices. We evaluate semi-supervision for medical image segmentation using heterogeneous datasets and clinical metrics. As methods for training models with little labeled data become more common, knowledge about how they perform on clinical tasks, how they fail, and how they perform with different amounts of labeled data is useful to model developers and users.

Reproducibility of Systolic Strain in Mice Using Cardiac Magnetic Resonance Feature Tracking of Black-Blood Cine Images.

Mouse models are widely utilized to enhance our understanding of cardiac disease. The goal of this study is to investigate the reproducibility of strain parameters that were measured in mice using cardiac magnetic resonance (CMR) feature-tracking (CMR42, Canada). We retrospectively analyzed black-blood CMR datasets from thirteen C57BL/6 B6.SJL-CD45.1 mice (N=10 female, N=3 male) that were imaged previously. The circumferential, longitudinal, and radial (Ecc, Ell, and Err, respectively) parameters of strain were measured in the mid-ventricular region of the left ventricle. Intraobserver and interobserver reproducibility were assessed for both the end-systolic (ES) and peak strain. The ES strain had larger intraclass correlation coefficient (ICC) values when compared to peak strain, for both the intraobserver and interobserver reproducibility studies. Specifically, the intraobserver study showed excellent reproducibility for all three ES strain parameters, namely, Ecc (ICC 0.95, 95% CI 0.83-0.98), Ell (ICC 0.90, 95% CI 0.59-0.97), and Err (ICC 0.92, 95% CI 0.73-0.97). This was also the case for the interobserver study, namely, Ecc (ICC 0.92, 95% CI 0.60-0.98), Ell (ICC 0.76, 95% CI 0.33-0.93), and Err (ICC 0.93, 95% CI 0.68-0.98). Additionally, the coefficient of variation values were all <10%. The results of this preliminary study showed excellent reproducibility for all ES strain parameters, with good to excellent reproducibility for the peak strain parameters. Moreover, all ES strain parameters had larger ICC values than the peak strain. In general, these results imply that feature-tracking with CMR42 software and black-blood cine images can be reliably used to assess strain patterns in mice.

Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation: The M&Ms Challenge.

The emergence of deep learning has considerably advanced the state-of-the-art in cardiac magnetic resonance (CMR) segmentation. Many techniques have been proposed over the last few years, bringing the accuracy of automated segmentation close to human performance. However, these models have been all too often trained and validated using cardiac imaging samples from single clinical centres or homogeneous imaging protocols. This has prevented the development and validation of models that are generalizable across different clinical centres, imaging conditions or scanner vendors. To promote further research and scientific benchmarking in the field of generalizable deep learning for cardiac segmentation, this paper presents the results of the Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation (M&Ms) Challenge, which was recently organized as part of the MICCAI 2020 Conference. A total of 14 teams submitted different solutions to the problem, combining various baseline models, data augmentation strategies, and domain adaptation techniques. The obtained results indicate the importance of intensity-driven data augmentation, as well as the need for further research to improve generalizability towards unseen scanner vendors or new imaging protocols. Furthermore, we present a new resource of 375 heterogeneous CMR datasets acquired by using four different scanner vendors in six hospitals and three different countries (Spain, Canada and Germany), which we provide as open-access for the community to enable future research in the field.

Additional diagnostic value of cardiac magnetic resonance feature tracking in patients with biopsy-proven arrhythmogenic cardiomyopathy

BackgroundWe aim to evaluate the value of Cardiac magnetic resonance (CMR) feature tracking (CMR-FT) in addition to Task Force Criteria(TFC) in patients with (arrhythmogenic cardiomyopathy) AC biopsy-proved. MethodsThirty-five patients with AC histologically proven who performed CMR with late gadolinium enhancement (LGE) acquisition were enrolled. The study population was divided in Group1 (negative CMR TFC and LV ejection fraction≥55%) and Group2 (positive CMR TFC and/or LVEF<55%) and compared to an age and gender-matched control group. CMR datasets of all patients were analyzed to calculate LV indexed end-diastolic (LVEDi) and end-systolic (LVESi) volumes and RV indexed end-diastolic (RVEDi) and end-systolic (RVESi) volumes, both LV ejection fraction (LVEF) and RV ejection fraction (RVEF). Moreover, LV and RV global longitudinal (GLS), circumferential (GCS) and radial (GRS) strain were measured. ResultsThe AC patients showed both higher LVEDi (p:0.002) and RVEDi (p:0.017) and lower LVEF (p: 0.016) as compared to control patients. Moreover, AC patients showed impaired LV-GLS (p < 0.001), LV-GRS (p < 0.001), LV-GCS (p < 0.001) and RV-GRS (p:0.026) as compared to control subjects. Group1 patients showed a significant reduction of LV-GRS (p < 0.05) and LV-GCS p < 0.01) as compared to control subjects. At univariate analysis LV-GCS was the most discriminatory parameter between Group1 vs heathy subjects with an optimal cut-off of −15.8 (Sensitivity: 74%; Specificity: 10%). ConclusionsIn patients with AC biopsy-proven, CMR-FT could improve the diagnostic yield in the subset of patients who results negative for imaging TFC criteria resulting as useful gatekeeper for indication of myocardial biopsy in case of equivocal clinical and imaging presentation.

Eosinophilic heart disease: diagnostic and prognostic assessment by cardiac magnetic resonance.

Eosinophilic heart disease (EHD) is a rare cardiac condition with a wide spectrum of phenotypes. The diagnostic and prognostic value of cardiac magnetic resonance (CMR) in EHD remains unknown. This was a retrospective analysis of 250 patients with eosinophilia referred for a CMR scan (period 2000-2020). CMR data sets and clinical/laboratory data were collected. Patients were followed up for a mean of 24 months (range 1-224) for the composite endpoint of death, acute coronary syndrome, hospitalization for acute heart failure, malignant ventricular arrhythmias, or the need for implantable cardiac defibrillator/pacemaker. The main objectives were to explore the diagnostic value of CMR in EHD; relationships between cardiac function, late gadolinium enhancement (LGE), and EHD phenotypes; and the prognostic value of fibrosis and oedema by CMR. The prevalence of findings compatible with EHD was 39% (patients with cardiac symptoms: 57% vs. screening: 20%, P < 0.001). EHD phenotypes included subendocardial LGE (n = 58), mid-wall/subepicardial LGE (n = 26), pericarditis (n = 5) or dilated cardiomyopathy (n = 8). Myocardial oedema was present in 10% of patients. Intracardiac thrombi (7%) were associated with EHD phenotype (χ2=47.3, P = 1.3×10-8). LGE extent correlated with LVEDVi (rho = 0.268, P = 5.3×10-5) and LVEF (rho=-0.415, P = 8.6×10-11). A CMR scan positive for EHD [hazard ratio (HR) = 5.61, 95% confidence interval (CI): 1.82-17.89, P = 0.0026] or a subendocardial LGE pattern (HR = 5.13, 95% CI: 1.29-20.38, P = 0.020) were independently associated with the composite clinical endpoint. The diagnostic yield of CMR screening in patients with persistent eosinophilia, even if asymptomatic, is high. The extent of subendocardial fibrosis correlates with LV remodelling and independently predicts clinical outcomes in patients with eosinophilia.

Assessment of Global Longitudinal and Circumferential Strain Using Computed Tomography Feature Tracking: Intra-Individual Comparison with CMR Feature Tracking and Myocardial Tagging in Patients with Severe Aortic Stenosis.

In this study, we used a single commercially available software solution to assess global longitudinal (GLS) and global circumferential strain (GCS) using cardiac computed tomography (CT) and cardiac magnetic resonance (CMR) feature tracking (FT). We compared agreement and reproducibility between these two methods and the reference standard, CMR tagging (TAG). Twenty-seven patients with severe aortic stenosis underwent CMR and cardiac CT examinations. FT analysis was performed using Medis suite version 3.0 (Leiden, The Netherlands) software. Segment (Medviso) software was used for GCS assessment from tagged images. There was a trend towards the underestimation of GLS by CT-FT when compared to CMR-FT (19.4 ± 5.04 vs. 22.40 ± 5.69, respectively; p = 0.065). GCS values between TAG, CT-FT, and CMR-FT were similar (p = 0.233). CMR-FT and CT-FT correlated closely for GLS (r = 0.686, p < 0.001) and GCS (r = 0.707, p < 0.001), while both of these methods correlated moderately with TAG for GCS (r = 0.479, p < 0.001 for CMR-FT vs. TAG; r = 0.548 for CT-FT vs. TAG). Intraobserver and interobserver agreement was excellent in all techniques. Our findings show that, in elderly patients with severe aortic stenosis (AS), the FT algorithm performs equally well in CMR and cardiac CT datasets for the assessment of GLS and GCS, both in terms of reproducibility and agreement with the gold standard, TAG.

Deep convolutional neural networks for automatic segmentation of left ventricle cavity from cardiac magnetic resonance images

This work conducts a feasibility study of deep learning approaches for automatic segmentation of left ventricle (LV) cavity from cardiac magnetic resonance (CMR) images. Automatic LV cavity segmentation is a challenging task, partially due to the small size of the object as compared to the large CMR image background, especially at the apex. To cater for small object segmentation, the authors present a localisation‐segmentation framework, to first locate the object in the large full image, then segment the object within the small cropped region of interest. The localisation is performed by a deep regression model based on convolutional neural networks, while the segmentation is done by the deep neural networks based on U‐Net architecture. They also employ the Dice loss function for the training process of the segmentation models, to investigate its effects on the segmentation performance. The deep learning models are trained and evaluated by using public endocardium‐annotated CMR datasets from York University and MICCAI 2009 LV Challenge websites. The average dice metric values of the authors’ proposed framework are 0.91 and 0.93, respectively, on these two databases. These results are promising as compared to the best results achieved by the current state‐of‐art, which shows the potentials of deep learning approaches for this particular application.

Blood Pool Segmentation Results in Superior Virtual Cardiac Models than Myocardial Segmentation for 3D Printing.

The method of cardiac magnetic resonance (CMR) three-dimensional (3D) image acquisition and post-processing which should be used to create optimal virtual models for 3D printing has not been studied systematically. Patients (n=19) who had undergone CMR including both 3D balanced steady-state free precession (bSSFP) imaging and contrast-enhanced magnetic resonance angiography (MRA) were retrospectively identified. Post-processing for the creation of virtual 3D models involved using both myocardial (MS) and blood pool (BP) segmentation, resulting in four groups: Group 1-bSSFP/MS, Group 2-bSSFP/BP, Group 3-MRA/MS and Group 4-MRA/BP. The models created were assessed by two raters for overall quality (1-poor; 2-good; 3-excellent) and ability to identify predefined vessels (1-5: superior vena cava, inferior vena cava, main pulmonary artery, ascending aorta and at least one pulmonary vein). A total of 76 virtual models were created from 19 patient CMR datasets. The mean overall quality scores for Raters 1/2 were 1.63±0.50/1.26±0.45 for Group 1, 2.12±0.50/2.26±0.73 for Group 2, 1.74±0.56/1.53±0.61 for Group 3 and 2.26±0.65/2.68±0.48 for Group 4. The numbers of identified vessels for Raters 1/2 were 4.11±1.32/4.05±1.31 for Group 1, 4.90±0.46/4.95±0.23 for Group 2, 4.32±1.00/4.47±0.84 for Group 3 and 4.74±0.56/4.63±0.49 for Group 4. Models created using BP segmentation (Groups 2 and 4) received significantly higher ratings than those created using MS for both overall quality and number of vessels visualized (p<0.05), regardless of the acquisition technique. There were no significant differences between Groups 1 and 3. The ratings for Raters 1 and 2 had good correlation for overall quality (ICC=0.63) and excellent correlation for the total number of vessels visualized (ICC=0.77). The intra-rater reliability was good for Rater A (ICC=0.65). Three models were successfully printed on desktop 3D printers with good quality and accurate representation of the virtual 3D models. We recommend using BP segmentation with either MRA or bSSFP source datasets to create virtual 3D models for 3D printing. Desktop 3D printers can offer good quality printed models with accurate representation of anatomic detail.

Application of Virtual Three-Dimensional Models for Simultaneous Visualization of Intracardiac Anatomic Relationships in Double Outlet Right Ventricle.

Our goal was to construct three-dimensional (3D) virtual models to allow simultaneous visualization of the ventricles, ventricular septal defect (VSD) and great arteries in patients with complex intracardiac anatomy to aid in surgical planning. We also sought to correlate measurements from the source cardiac magnetic resonance (CMR) image dataset and the 3D model. Complicated ventriculo-arterial relationships in patients with complex conotruncal malformations make preoperative assessment of possible repair pathways difficult. Patients were chosen with double outlet right ventricle for the complexity of intracardiac anatomy and potential for better delineation of anatomic spatial relationships. Virtual 3D models were generated from CMR 3D datasets. Measurements were made on the source CMR as well as the 3D model for the following structures: aortic diameter in orthogonal planes, VSD diameter in orthogonal planes and long axis of right ventricle. A total of six patients were identified for inclusion. The path from the ventricles to each respective outflow tract and the location of the VSD with respect to each great vessel was visualized clearly in all patients. Measurements on the virtual model showed excellent correlation with the source CMR when all measurements were included by Pearson coefficient, r=0.99 as well as for each individual structure. Construction of virtual 3D models in patients with complex conotruncal defects from 3D CMR datasets allows for simultaneous visualization of anatomic relationships relevant for surgical repair. The availability of these models may allow for a more informed preoperative evaluation in these patients.

Fast automatic myocardial segmentation in 4D cine CMR datasets.

A novel automatic 3D+time left ventricle (LV) segmentation framework is proposed for cardiac magnetic resonance (CMR) datasets. The proposed framework consists of three conceptual blocks to delineate both endo and epicardial contours throughout the cardiac cycle: (1) an automatic 2D mid-ventricular initialization and segmentation; (2) an automatic stack initialization followed by a 3D segmentation at the end-diastolic phase; and (3) a tracking procedure. Hereto, we propose to adapt the recent B-spline Explicit Active Surfaces (BEAS) framework to the properties of CMR images by integrating dedicated energy terms. Moreover, we extend the coupled BEAS formalism towards its application in 3D MR data by adapting it to a cylindrical space suited to deal with the topology of the image data. Furthermore, a fast stack initialization method is presented for efficient initialization and to enforce consistent cylindrical topology. Finally, we make use of an anatomically constrained optical flow method for temporal tracking of the LV surface. The proposed framework has been validated on 45 CMR datasets taken from the 2009 MICCAI LV segmentation challenge. Results show the robustness, efficiency and competitiveness of the proposed method both in terms of accuracy and computational load.

Nearly automated motion artifacts correction between multi breath-hold short-axis and long-axis cine CMR images

Background: We aimed at developing and testing a nearly automated method for breath-hold artifacts compensation in short-axis (SA) cardiac magnetic resonance (CMR) images. The purpose was the reduction of potential misalignment between standard cine SA and two- and four-chamber long-axis (LA) CMR images to allow 3D reconstruction for segmentation or modeling purposes.Methods: The 3D position of each SA image was optimized on the basis of the pixel intensities at the intersections with the two- and four-chamber LA images. The algorithm accuracy was first tested on a dedicated virtual phantom dataset, derived from a high resolution computed tomography frame where known misalignments were applied. The method was then applied to SA and LA CMR end-diastolic and end-systolic frames datasets obtained in 20 consecutive patients. Assessment of the results was performed by two independent observers by visual comparison and by quantifying the residual distances between LA and SA left ventricle endocardial contours before and after correction.Results: Errors on the simulation dataset were quantified as residual distance from the ground truth position of SA planes and values were found of the order of the pixel resolution. On CMR datasets, a perceived improvement was reported in about 70% of the slices in need for correction and median residual error between manual SA and LA contours was reduced from 2.4mm to 1.8mm.Discussion: Results found on virtual and clinical datasets proved feasibility and usefulness of the method as a necessary pre-processing step for volumetric analysis of CMR data in clinical setting.

Impact of image plane selection on measures of left ventricular non-compaction

Background Cardiac magnetic resonance (CMR) is frequently used to assess the extent of trabeculations in cases of left ventricular non-compaction. Most routine CMR studies do not acquire full 3D isotropic cine datasets and consequently suboptimal acquisition of prescribed image views could lead to clinically important inaccuracies in calculating non-compacted to compacted myocardial ratios (NC/C). This study compares measurement of the NC/C ratio between optimal and off-axis views derived from 3D cardiac computer tomography (CT) datasets and compares these with those derived from optimal CMR datasets. Methods We studied 10 participants who underwent CMR and contrast CT imaging (retrospective gating) as part of the Evaluation of Integrated Cardiac Imaging for the Detection, Characterization and Monitoring of Ischemic Heart Disease study protocol (EVINCI study). NC/C ratios were determined for CMR and for CT images in mid and apical anterior and inferior segments based on 2 chamber views and short axis views in diastole. NC/C ratios in the same segments were compared in “off-axis” 2-chamber view and short axis views in optimal position and tilted by 30 and 45 degrees for CT. Results Comparing measurements from optimal 2-chamber views for CMR and CT, there was no difference for NC/ C ratio (CMR 1.1 ± 0.5 vs. CT 0.8 ± 0.3, p = 0.2). The NC/C ratio was higher in MRI short axis view compared with CT short axis (1.0 ± 0.4 vs. 0.6 ± 0.4, p = 0.001). The NC/C ratio was overestimated from off-axis 2 chamber CT views compared to the optimal 2 chamber view: NC/C 1.0 ± 0.4 vs. 0.8 ± 0.3, p = 0.001. There were no differences between NC/C ratio in optimally oriented short axis and angulated planes: 0.6 ± 0.4 vs. 0.6 ± 0.3 (30 degrees) vs. 0.6 ± 0.3 (45 degrees), p = 0.9.

Velocity Vector Imaging in the Measurement of Left Ventricular Myocardial Mechanics on Cardiac Magnetic Resonance Imaging: Correlations with Echocardiographically Derived Strain Values

Regional and global function can be measured by echocardiography using speckle-tracking, a technique that has previously been validated against crystal sonomicrometry. However, the application of Velocity Vector Imaging (VVI) to images obtained from cardiac magnetic resonance (CMR) imaging has never been validated against those values derived from VVI applied to two-dimensional echocardiographic images in the same patient group. The aim of this study was to validate for the first time the application of VVI to retrospectively acquired CMR data sets for the assessment of left ventricular strain and rotation, using echocardiographic strain assessment by VVI as the reference technique. Cine steady-state free precession CMR data sets and two-dimensional echocardiographic images obtained on the same day in 36 adult patients with hypertrophic cardiomyopathy were analyzed retrospectively using VVI to quantify global longitudinal and circumferential strain and rotation parameters. The absolute differences in longitudinal strain between the two imaging modalities were -1.1 ± 3.3% (endocardial) and -2.2 ± 3.6% (full thickness). The absolute differences in circumferential strain were -4.7 ± 5.3% (endocardial) and -3.4 ± 3.8% (full thickness). CMR consistently resulted in higher strain values than echocardiography. The absolute differences in twist were -0.2 ± 5.6% (endocardial) and 0.1 ± 5.8% (full thickness). The application of VVI to CMR data sets allows a feasible and reproducible method for strain analysis in HCM, demonstrating excellent agreement with two-dimensional echocardiography-derived values. Given the superior image quality obtained with CMR in a significant proportion of patients, this technique provides a method for strain assessment without the need for dedicated CMR acquisition and analytic techniques.

Right ventricular quantification in clinical practice: two-dimensional vs. three-dimensional echocardiography compared with cardiac magnetic resonance imaging

To establish the additional value of three-dimensional echocardiography (3D echo) for assessment of right ventricular (RV) size and function in patients with congenital heart disease (CHD) in everyday clinical practice, the accuracy and reproducibility of 3D echo were compared with conventional two-dimensional echocardiography (2D echo) and cardiac magnetic resonance (CMR) imaging as reference. Patients with CHD and primarily affected right ventricles (n = 62), patients with CHD and primarily affected left ventricles (LV group, n = 27), and healthy controls (n = 31) were studied. 2D echo-, 3D echo- and CMR data sets were obtained. Moderate correlations were found between RV dimensions by 2D echo and CMR-derived RV end-diastolic volumes (r = 0.32-0.77). The correlations between RV volumes obtained by 3D echo and CMR imaging were better (r = 0.71-0.97) than the 2D echo-derived correlations (P < 0.001). Only the 2D echo-derived RV inlet diameter correlated better in healthy controls than in the RV group. Receiver operating characteristic curve analysis revealed that 3D echo-derived end-diastolic volume best identified RV dysfunction (sensitivity 95% and specificity 100%). The 3D echo-derived measurements were as reproducible as the 2D echo-derived measurements (n = 37, coefficients of variation ranging from 5 to 19%), with tricuspid annular plane systolic excursion being the most reproducible measurement (coefficient of variation of 6%). 3D echo improved quantitative RV size and function assessment compared with 2D echo in patients as well as in healthy controls. Everyday clinical use of 3D echo for RV assessment can be reality with the currently available software and provides incremental benefit in assessment of the right ventricle.

Multi–detector row cardiac computed tomography accurately quantifies right and left ventricular size and function compared with cardiac magnetic resonance

Cardiac magnetic resonance (CMR) accurately quantifies right ventricular (RV) and left ventricular (LV) volumes and function. Limited availability of CMR and increasing use of MR-incompatible cardiovascular devices underscore the potential utility of cardiac computed tomography (CT) for ventricular quantification. This study quantified biventricular size and systolic function with multi-detector row CT compared with CMR imaging. Twenty-six subjects prospectively underwent CT and CMR examinations on a 16-detector CT and 1.5 T MR scanner, respectively; claustrophobia in one and nongated CT imaging in another precluded complete imaging in 2 subjects. Contiguous multiphase short-axis images were generated from axial CT data, and steady-state free precession cine MR produced contiguous short-axis cines. Semiautomated software generated ventricular borders to calculate volume, mass, and ejection fraction (EF) from both sets of images. Blinded observers completed quantification and wall motion analyses of 23 CMR and CT data sets independently. All measures of LV size and function by cardiac CT correlated well with CMR over a wide range of LV function (LVEF 30%-72% by CMR), including end-diastolic volume (r = 0.97), end-systolic volume (r = 0.97), EF (r = 0.97), and mass (r = 0.95). Of 24 cases, 6 had inadequate contrast opacification of the RV precluding RV segmentation. In the remaining 18 CMR-CT data pairs, RVEF showed moderate agreement (r = 0.86), and RV volumes correlated well (r = 0.97 and 0.94 for RV end-diastolic volume and RV end-systolic volume, respectively). Ten percent of LV segments visualized by CT were inadequate for wall motion assessment due to motion artifact or inadequate contrast between myocardium and endocardium. For segments adequately visualized by both techniques, the mean kappa statistic was 0.88 (range 0.78-1.0), consistent with good agreement. Cardiac CT accurately quantifies LV size and function; RV quantification with cardiac CT requires optimized contrast opacification of the RV.

A new virtual reality approach for planning of cardiac interventions

A novel approach to three-dimensional (3D) visualization of high quality, respiratory compensated cardiac magnetic resonance (MR) data is presented with the purpose of assisting the cardiovascular surgeon and the invasive cardiologist in the pre-operative planning. Developments included: (1) optimization of 3D, MR scan protocols; (2) dedicated segmentation software; (3) optimization of model generation algorithms; (4) interactive, virtual reality visualization. The approach is based on a tool for interactive, real-time visualization of 3D cardiac MR datasets in the form of 3D heart models displayed on virtual reality equipment. This allows the cardiac surgeon and the cardiologist to examine the model as if they were actually holding it in their hands. To secure relevant examination of all details related to cardiac morphology, the model can be re-scaled and the viewpoint can be set to any point inside the heart. Finally, the original, raw MR images can be examined on line as textures in cut-planes through the heart models.