Study population and design
We retrospectively studied 50 patients who underwent CMR imaging (including SENC and LGE) and transthoracic 2D echocardiography at the University of Chicago, Chicago, Illinois, USA over a one-year period. Patients under 18 years of age and those who underwent a cardiac intervention between the two imaging tests were excluded. No patients were excluded on the basis of image quality of either modality. The Institutional Review Board approved this retrospective study with a waiver of consent.
Figure 1 schematically depicts the study design. The above three techniques were used to measure strain: feature tracking (FT) and strain encoding (SENC) images were analyzed to obtain both global longitudinal strain (GLS) and global circumferential strain (GCS), while speckle tracking echocardiography (STE) was used to obtain GLS. These measurements were compared between them and also correlated with presence of LGE.
Echocardiographic imaging and analysis
Transthoracic echocardiographic imaging was performed using the iE33 system with the X5–1 probe (Philips Healthcare, Best, the Netherlands). Apical long-axis, LV-focused two-, three- and four-chamber (2Ch, 3Ch and 4Ch) views were acquired, after optimizing the sector size, gain, depth, compress and time-gain compensation. Frame rate was maximized (73 ± 20 fps) by increasing the depth and decreasing the sector width.
The images were stored digitally and measured offline according to the guidelines  by an experienced reader, blinded to clinical data and all prior strain measurements and LGE findings. End-diastole (ED) was identified as the frame coinciding with the peak of the QRS complex, whereas end-systole (ES) was identified as the frame with the smallest LV cavity. LV GLS analysis was performed on the three long-axis views using vendor independent speckle-tracking software (Echo Insight, Epsilon Imaging, Ann Arbor, Michigan, USA). This software is based on tracking ultrasound speckles frame-by-frame in order to quantify myocardial deformation. It has been previously validated by comparisons against other established techniques used to measure myocardial strain [3, 4, 7].
Specifically, for each view, strain analysis was performed by manually tracing at ED the region of interest along the endocardial border from the mitral valve annulus to the LV apex and back to the annulus (Fig. 2a). The software then automatically tracked the endocardial contours throughout the cardiac cycle. Manual adjustments were made to the contours as needed to optimize tracking. All views were segmented according to the American Heart Association (AHA) guidelines and segmental strain was calculated automatically throughout the cardiac cycle. GLS was calculated throughout the cardiac cycle, resulting in a time-strain curve for each view (Fig. 2). Peak GLS values were averaged for the three views, resulting in a unique GLS value for each patient.
Cine CMR imaging and feature tracking analysis
CMR images were acquired on a 1.5 T scanner using a 5-channel surface coil (Achieva, Philips Healthcare). Retrospectively gated cine images were acquired using a balanced steady-state free precession pulse sequence in the standard long-axis views (2Ch, 3Ch, 4Ch) and short-axis slices (6 mm thickness, 4 mm gap), covering the LV from base to apex. Scanning parameter were: TR = 2.9 ms, TE = 1.5 ms, flip angle 60°, temporal resolution 30-40 ms.
FT was performed offline by an experienced observer, blinded to all prior strain measurements and LGE findings, using vendor independent software (SuiteHEART, Neosoft, Pewaukee, Wisconson, USA). Similar to echocardiographic speckle tracking, the FT algorithm identifies image features in the myocardium that are consistently identifiable throughout the cardiac cycle, and tracks them frame-by-frame to quantify myocardial deformation. This is achieved by the following five steps: (1) deformation models are created based on b-splines using contours and images; (2) position images are created by calculating how much each pixel within the myocardium is displaced over the cardiac cycle; (3) strain tensor is calculated; (4) tensor image is transformed from Cartesian coordinates to the radial/cross-radial coordinates; and (5) velocity and strain rate tensors are calculated using the central difference.
The long-axis cine images were used to determine GLS and short-axis slices covering the entire heart to determine GCS. First, ED and ES were determined automatically in each view and manually corrected as needed. The tracing of the region of interest in the long-axis images was performed by an automated machine-learning based process, tracking epi- and endocardial contours from the mitral valve annulus to the apex and back to the annulus. These contours were then reviewed on every frame throughout the cardiac cycle and manually corrected as needed to optimize endocardial detection and tracking while taking care to exclude papillary muscles and endocardial trabeculae from the LV cavity (Fig. 3a). All views were segmented according to the AHA guidelines and segmental strain was calculated automatically throughout the cardiac cycle. If segments were inadequately tracked, tracing was repeated until optimal tracking was achieved. Peak systolic GLS and GCS were calculated as a mean value of all segments (Fig. 3).
CMR strain encoding and analysis
SENC images were acquired in three long-axis (2Ch, 3Ch, 4Ch) and three short-axis views (basal, mid, apical) with the following settings: TR = 13 ms; TE = 0.7 ms; FA = 30°; 256x256mm2; slice thickness = 10 mm; 24 ms SENC magnetization preparation prior to continuous acquisition of 40 ms (3 spiral interleaves) per temporal frame over 1 R-R cycle.
GLS was quantified using the three short-axis slices and GCS using the long-axis slices by the same observer three weeks after the analysis of the FT images to prevent bias, using vendor independent software (Myostrain 5.0, Myocardial Solutions, Morrisville, North Carolina, USA). Unlike STE and FT, SENC measures strain in the direction perpendicular to the imaging plane: circumferential from the long-axis and longitudinal from the short-axis images. Radial strain is not usually assessed using SENC. This is achieved by using specialized pulse sequences designed to measure strain and generate color-encoded strain maps superimposed on a static anatomic image of the heart .
ED and end-systole (ES) were selected manually in all slices according to the size of the myocardial cavity and the color-coding of the images, representing the state of contraction (blue = contracting- yellow = relaxing). Epi- and endocardial contours were drawn manually at ES, again using the mitral valve annulus and apex as landmarks in the long-axis views and excluding the papillary muscles and trabeculae from the LV cavity (Fig. 4). GLS and GCS were automatically calculated for each view and then averaged. No segments were excluded from analysis.
Late gadolinium enhancement
LGE images were acquired in the same long-axis planes as the cine images and also in the short-axis slices covering the entire LV, 5–10 min after the infusion of gadolinium contrast (OmniscanTM or MultiHance TM, 0.05–0.1 mmol/kg, injected at 4 ml/s, IV), using a T1-weighted gradient echo pulse sequence with a phase sensitive inversion recovery reconstruction (TR = 4.5 ms; TE = 2.2 ms; TI = 250–300 ms, flip angle 30°, flip angle 5°, voxel size = 2x2x10 mm, SENSE factor = 1–2, no gaps). An inversion time scout sequence was used to select an inversion time between 200 and 300 ms for optimal nulling of normal myocardium. The presence of LGE was qualitatively evaluated by a clinical expert (Level II or III certified ) blinded to all strain results, but with access to patients’ clinical data, to identify areas of hyper-enhancement in the myocardium consistent with either post-infarct scar or cardiac involvement in infiltrative disease .
In a subset of 10 randomly selected patients, measurements were repeated by the same observer, two weeks after the first analysis (to prevent recall bias) and by a second independent observer for every modality and technique, all blinded to prior measurements and LGE findings.
All values were assessed for normality using the Shapiro-Wilk test. Normally distributed data is expressed as mean ± SD, non-normally distributed data using median and interquartile range (IQR). Linear regression and Bland-Altman analyses were used to determine inter-technique agreement between STE, FT and SENC for GLS and between FT and SENC for GCS. Intra- and inter-observer variability was expressed in terms of intraclass-correlations (ICC) and coefficients of variation (CoV). Receiver operating characteristics (ROC) - curves were generated to establish the relationship of each strain parameter (STE-GLS, FT-GLS, FT-GCS, SENC-GLS, SENC-GCS) to the presence of LGE and the area under curve (AUC) was calculated. A Mann-Whitney test was conducted to determine if strain values measured using each technique, differed significantly between the patient groups with and without LGE. Binary logistic regression analyses were performed to determine the associations between strain measurements and the presence of LGE, which was expressed in terms of odds ratios (OR). A p-value of ≤0.05 was considered significant in two-tailed tests. Variables that were significantly associated with LGE presence were checked for collinearity by Spearman rank correlations and entered into separate multivariate logistic regression models for each technique to avoid overfitting and identify strain parameters that were independently associated with LGE. Statistical analyses were conducted using SPSS (Version 25.0, Statistical Package for the Social Sciences (SSPS), International Business Machines, Inc., Armonk, New York, USA).