Variation in cardiovascular magnetic resonance myocardial contouring: Insights from an international survey

Level of Evidence: 5 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019;50:1336–1338.

more readers used detailed (56%) than smoothed (44%) endocardial contouring. Papillary muscles and trabeculations were included in the LV blood volume in 46%. The majority of respondents (77%) used normal reference ranges derived from methods that included papillary muscles and trabeculations as LV mass, but only 32% of observers drew them as such-a major inconsistency. Mitral valve plane tracking software was used by only 26%, while 6% did not report LV mass at all. For the right ventricle, most centers (64%) used detailed endocardial contouring, with a smaller proportion using smoothed contours (27%) or merely visual assessment (11%). Most centers (66%) did not run a specific training program on volumetric analysis. This survey has revealed wide variation between international centers in the contouring methods used to quantitate cardiac volumes, mass, and function. According to SCMR recommendations, the choice of reference ranges and clinical reporting technique should match, although these data highlight discordance. This may relate to Hudsmith et al 5 7 Alfakih et al 6 5 In-house reference range/Other 9

Left Ventricular Myocardial Contours
Detailed 56 Smoothed 44 Data are percentages taken from a total of 65 respondents from 55 international cardiac MRI centers. a Not mutually exclusive. b Includes QMass Medis Medical, GE Healthcare (Suiteheart), CIM (University of Auckland).
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.   4 The panels demonstrate contouring of the ventricles using identical SSFP cine short axis images, from base to apex at end-diastole and end-systole, below which are tables of the respective LV parameters. For detailed contouring, those papillary muscles and trabeculations continuous with the LV endocardial border were included in mass and excluded from the blood pool. Note not only the differences in values, but also the different classification of LV hypertrophy based on the two techniques.
time constraints, access to software with "thresholding" capability (ie, intensity-based segmentation), availability of vendor-specific reference ranges without information detailing papillary muscle inclusion/exclusion, or failure to read and understand the reference methods. It could also reflect the lack of a representative example of the contouring method in the reference article; most, [3][4][5][6] but not all, 7,8 published reference ranges include exemplar figures, which should be obligatory for any future reference technique publication. We have performed volumetric and mass analyses in 20 consecutive patients (57 AE 16 years; male 70%) with hypertrophic cardiomyopathy (HCM), using the two most common postprocessing methodologies based on our survey data (Table 2). 4,7 HCM was defined by the presence of LV wall thickness ≥ 15 mm unexplained by loading conditions. Nine patients (45%) with a supranormal LV ejection fraction by detailed contours were reclassified with a normal ejection fraction by smoothed contours. Four patients (20%) with LV hypertrophy according to detailed contours had a normal indexed LV mass by smoothed contours. These data suggest using smoothed contours is untenable in HCM and its phenocopies-the presence of large papillary muscles and extensive trabeculations leads to inaccuracy, often missing hyperdynamic function. 9,10 Additionally, when serial imaging is requested to characterize the clinical course of HCM, variable contouring practice between centers will result in reported differences in LV parameters that has the potential to mislead clinicians.
Inconsistent conclusions arising from analyses using these two distinct contouring techniques are not restricted to patients with overt pathology. In Fig. 1, we provide examples of myocardial contours in a healthy 52-year-old Caucasian female (without prior cardiovascular disease and with a normal 24-hour ambulatory blood pressure). Again, there are important differences in the values obtained for LV parameters dependent on the contouring method. A more striking finding, however, is the difference between the two reference datasets in the normal cutoff value for LV mass. This healthy control subject is classified abnormal with eccentric LV hypertrophy according to the UK Biobank dataset, 4 while LV mass is well within the normal range using methods described by Maceira et al. 7 The exclusion of non-Caucasians and subjects aged < 45 years further limits the applicability of the UK Biobank normal range dataset. 4 The authors' consensus opinion is that papillary muscles are myocardial tissue and to improve accuracy (closeness of a measured value to a true value) should routinely be excluded from blood volumes and included in LV mass. Most centers currently use software capable of producing contours using thresholding, which enables observers to perform detailed contouring that takes account of papillary muscles without sacrificing time.
The wide variation in cardiac MRI reporting practice emphasized by this survey reflects a global issue. Although machine learning holds promise (removing interobserver error, increasing standardization, and permitting reference range changes "on the fly" as models refine), its arrival into the clinical arena is not anticipated for some years. There is, therefore, a pressing need to formalize the choice of postprocessing methodology and specify a normal reference range that the MRI community should follow.