Motivation: Integrative mathematical and statistical models of cardiac anatomy and physiology can play a vital role in understanding cardiac disease phenotype and planning therapeutic strategies. However, the accuracy and predictive power of such models is dependent upon the breadth and depth of noninvasive imaging datasets. The Cardiac Atlas Project (CAP) has established a large-scale database of cardiac imaging examinations and associated clinical data in order to develop a shareable, web-accessible, structural and functional atlas of the normal and pathological heart for clinical, research and educational purposes. A goal of CAP is to facilitate collaborative statistical analysis of regional heart shape and wall motion and characterize cardiac function among and within population groups.Results: Three main open-source software components were developed: (i) a database with web-interface; (ii) a modeling client for 3D + time visualization and parametric description of shape and motion; and (iii) open data formats for semantic characterization of models and annotations. The database was implemented using a three-tier architecture utilizing MySQL, JBoss and Dcm4chee, in compliance with the DICOM standard to provide compatibility with existing clinical networks and devices. Parts of Dcm4chee were extended to access image specific attributes as search parameters. To date, approximately 3000 de-identified cardiac imaging examinations are available in the database. All software components developed by the CAP are open source and are freely available under the Mozilla Public License Version 1.1 (http://www.mozilla.org/MPL/MPL-1.1.txt).Availability: http://www.cardiacatlas.orgContact: a.young@auckland.ac.nzSupplementary information: Supplementary data are available at Bioinformatics online.

BackgroundHigh reproducibility of LV mass and volume measurement from cine cardiovascular magnetic resonance (CMR) has been shown within single centers. However, the extent to which contours may vary from center to center, due to different training protocols, is unknown. We aimed to quantify sources of variation between many centers, and provide a multi-center consensus ground truth dataset for benchmarking automated processing tools and facilitating training for new readers in CMR analysis.MethodsSeven independent expert readers, representing seven experienced CMR core laboratories, analyzed fifteen cine CMR data sets in accordance with their standard operating protocols and SCMR guidelines. Consensus contours were generated for each image according to a statistical optimization scheme that maximized contour placement agreement between readers.ResultsReader-consensus agreement was better than inter-reader agreement (end-diastolic volume 14.7 ml vs 15.2–28.4 ml; end-systolic volume 13.2 ml vs 14.0–21.5 ml; LV mass 17.5 g vs 20.2–34.5 g; ejection fraction 4.2 % vs 4.6–7.5 %). Compared with consensus contours, readers were very consistent (small variability across cases within each reader), but bias varied between readers due to differences in contouring protocols at each center. Although larger contour differences were found at the apex and base, the main effect on volume was due to small but consistent differences in the position of the contours in all regions of the LV.ConclusionsA multi-center consensus dataset was established for the purposes of benchmarking and training. Achieving consensus on contour drawing protocol between centers before analysis, or bias correction after analysis, is required when collating multi-center results.

A collaborative framework was initiated to establish a community resource of ground truth segmentations from cardiac MRI. Multi-site, multi-vendor cardiac MRI datasets comprising 95 patients (73 men, 22 women; mean age 62.73 ± 11.24 years) with coronary artery disease and prior myocardial infarction, were randomly selected from data made available by the Cardiac Atlas Project (Fonseca et al., 2011). Three semi- and two fully-automated raters segmented the left ventricular myocardium from short-axis cardiac MR images as part of a challenge introduced at the STACOM 2011 MICCAI workshop (Suinesiaputra et al., 2012). Consensus myocardium images were generated based on the Expectation-Maximization principle implemented by the STAPLE algorithm (Warfield et al., 2004). The mean sensitivity, specificity, positive predictive and negative predictive values ranged between 0.63-0.85, 0.60-0.98, 0.56-0.94 and 0.83-0.92, respectively, against the STAPLE consensus. Spatial and temporal agreement varied in different amounts for each rater. STAPLE produced high quality consensus images if the region of interest was limited to the area of discrepancy between raters. To maintain the quality of the consensus, an objective measure based on the candidate automated rater performance distribution is proposed. The consensus segmentation based on a combination of manual and automated raters were more consistent than any particular rater, even those with manual input. The consensus is expected to improve with the addition of new automated contributions. This resource is open for future contributions, and is available as a test bed for the evaluation of new segmentation algorithms, through the Cardiac Atlas Project (www.cardiacatlas.org).

Machine learning (ML) is making a dramatic impact on cardiovascular magnetic resonance (CMR) in many ways. This review seeks to highlight the major areas in CMR where ML, and deep learning in particular, can assist clinicians and engineers in improving imaging efficiency, quality, image analysis and interpretation, as well as patient evaluation. We discuss recent developments in the field of ML relevant to CMR in the areas of image acquisition & reconstruction, image analysis, diagnostic evaluation and derivation of prognostic information. To date, the main impact of ML in CMR has been to significantly reduce the time required for image segmentation and analysis. Accurate and reproducible fully automated quantification of left and right ventricular mass and volume is now available in commercial products. Active research areas include reduction of image acquisition and reconstruction time, improving spatial and temporal resolution, and analysis of perfusion and myocardial mapping. Although large cohort studies are providing valuable data sets for ML training, care must be taken in extending applications to specific patient groups. Since ML algorithms can fail in unpredictable ways, it is important to mitigate this by open source publication of computational processes and datasets. Furthermore, controlled trials are needed to evaluate methods across multiple centers and patient groups.

BackgroundAlthough left ventricular cardiac geometric indices such as size and sphericity characterize adverse remodeling and have prognostic value in symptomatic patients, little is known of shape distributions in subclinical populations. We sought to quantify shape variation across a large number of asymptomatic volunteers, and examine differences among sub-cohorts.MethodsAn atlas was constructed comprising 1,991 cardiovascular magnetic resonance (CMR) cases contributed from the Multi-Ethnic Study of Atherosclerosis baseline examination. A mathematical model describing regional wall motion and shape was used to establish a coordinate map registered to the cardiac anatomy. The model was automatically customized to left ventricular contours and anatomical landmarks, corrected for breath-hold mis-registration between image slices. Mathematical techniques were used to characterize global shape distributions, after removal of translations, rotations, and scale due to height. Differences were quantified among ethnicity, sex, smoking, hypertension and diabetes sub-cohorts.ResultsThe atlas construction process yielded accurate representations of global shape (errors between manual and automatic surface points in 244 validation cases were less than the image pixel size). After correction for height, the dominant shape component was associated with heart size, explaining 32% of the total shape variance at end-diastole and 29% at end-systole. After size, the second dominant shape component was sphericity at end-diastole (13%), and concentricity at end-systole (10%). The resulting shape components distinguished differences due to ethnicity and risk factors with greater statistical power than traditional mass and volume indices.ConclusionsWe have quantified the dominant components of global shape variation in the adult asymptomatic population. The data and results are available at cardiacatlas.org. Shape distributions were principally explained by size, sphericity and concentricity, which are known correlates of adverse outcomes. Atlas-based global shape analysis provides a powerful method for quantifying left ventricular shape differences in asymptomatic populations.Trial registrationClinicalTrials.gov NCT00005487