Clinical validation of three cardiovascular magnetic resonance techniques to measure strain and torsion in patients with suspected coronary artery disease

125 patients, all participants in the Doppler-CIP study [11] at Linköping University Hospital, were prospectively enrolled from November 2010 until March 2012. Nine patients were excluded from the analysis due to inadequate image quality—one in DENSE, three in tagging, one in both DENSE and tagging and four in FT, resulting in data from 116 patients available for analysis, 89 (77%) males with a mean age of 67 years (range 49–85). Inclusion criteria were: (1) a history of typical angina or high risk of CAD (> 15% risk of developing cardiovascular events according to the European risk SCORE at the time of assessment [12]), or a positive stress test with > 2 mm ST-depression, (2) known CAD, defined as prior myocardial infarction (> 3 months), or CAD on invasive coronary angiogram. Exclusion criteria were; unwillingness to participate, an acute coronary syndrome during the preceding three months, more than moderate valvular disease, pacemaker implantation, claustrophobia, an estimated glomerular filtration rate < 60 ml/min/1.73 m² or permanent atrial fibrillation. The Doppler-CIP study used a wide inclusion criterion. Therefore, the patient cohort was divided into a group without imaging signs of cardiac disease, having normal LV mass (LVM), blood pressure, LVEF, and wall motion and no signs of late gadolinium enhancement (LGE), and a group with imaging signs of cardiac disease. In addition, a group of 10 healthy subjects was recruited. They declared a history free from cardiovascular disease, did not take cardioactive drugs and had a normal transthoracic echocardiographic exam. These healthy subjects underwent a limited investigation including cine for volumes, FT strain and a specific acquisition for DENSE strain.

We acquired CMR data using a 1.5 T CMR system (Achieva Nova Dual, Philips Healthcare, Best the Netherlands) with a protocol including cine bSSFP, tagged images, DENSE, and LGE, all acquired during breath holding. The cine bSSFP images were acquired in the LAx views (2-, 3-, and 4-chamber views) as well as in a stack of SAx views covering the entire LV. Typical parameters were: slice thickness 8 mm, TR/TE 3.6/1.81, flip angle 60°, field of view 350 × 350 mm, matrix 288 × 288 and breath holding time 16 s. The cine images were retrospectively reconstructed into 30 cardiac phases, corresponding to a temporal resolution in the range of 24–41 ms (mean 37 ms).

Tagged images were acquired using complementary spatial modulation of magnetization (CSPAMM), with three SAx planes (basal, mid, and apical) planned in end systole (ES). The typical acquisition parameters for tagging were: inter-tag distance 8 mm, slice thickness 6 mm, TR/TE 4.6/2.1 ms, flip angle 15°, field of view 320 × 320 mm, matrix 256 × 256, 21 heart phases, and breath holding time 16 s. The mean temporal resolution was 43 ms.

DENSE images were acquired as previously described [13] in the same three SAx positions as the tagged images. Briefly, the DENSE sequence is built on three balanced multipoint encodings and threefold SPAMM (3-SPAMM) with fat suppression achieved by using a water-selective first radiofrequency pulse in the DENSE 1-1 SPAMM preparation. A six-shot spiral acquisition was used, with spiral readout interleave duration of 8 ms and TR/TE 11.2/1.27 ms, respectively and three interleaves were acquired each cardiac cycle. A through-slice dephasing of 0.25 Hz was used, and an in-plane displacement encoding strength of 0.30 Hz/pixel. We acquired three time frames: 45 ms prior to aortic valve closure, at valve closure, and 45 ms after the valve closed—where the time of valve closure was determined by bSSFP in the apical LAx view. The acquisition duration was 18 heartbeats, and the spatial resolution was 1.36 mm × 1.36 mm with a slice thickness of 6.0 mm.

LGE data were acquired at the same slice positions as the cine images (three LAx views and a stack of SAx slices) beginning on average 14 min (11–23 min) after the administration of 0.2 mmol/kg bodyweight gadopentetate dimeglumine (Bayer Healthcare, Berlin, Germany). An inversion recovery 3D spoiled gradient echo sequence with TR/TE 4.4/1.3 ms, respectively, was used. Slice thickness and slice gap in the SAx direction were 10 mm and -5 mm (5 mm overlap), respectively.

FT was performed on the SAx cine bSSFP stack and was limited to Lagrangian analysis, where the comparison was focused on Lagrangian circumferential strain. The LV was divided into 16 segments, excluding the apical cap from the 17-segment model of the American Heart Association [14]. The observer was blinded from the patient history and other image sequences when analyzing each image type.

A single observer (#1) extracted LV volumes, LVEF, and LVM from the SAx cine images using Segment v 1.9 R2966. Wall motion was determined from all cine views (three LAx and one stack of SAx) using the following qualitative scores: 1 = normokinesis; 2 = hypokinesis; 3 = akinesis; 4 = dyskinesis.

A single observer (#2) performed the FT analysis using the 2D-CPA MR software (v 1.2, TomTec Imaging System, Unterschleissheim, Germany). The endocardium and the epicardium were delineated manually in diastole and the software tracked the displacement of 48 points along the endocardium. Lagrangian strain was measured for both the endocardial and the epicardial layers, the mean of which provided transmural strain [15, 16]. The rotation of the apical and basal slice were computed in the software, then divided manually by the slice distance to obtain torsion.

The tagged images were segmented at a later time by observer #2 using Segment (v 2.2 R7056, Medviso AB, Lund, Sweden). Sixty points along the epicardial circumference and 30 points along the endocardium of each slice were semi-automatically tracked to compute Lagrangian strain from the deformation of the line connecting the points [17]. Torsion was obtained as the difference in rotation between two slices, as with FT.

Two observers (#3 and 4) segmented the myocardium for analysis of the DENSE data in the circumferential direction. The endocardium and the epicardium were delineated manually, and the slices were semi-automatically divided into segments, with reference to the attachment of the right ventricular (RV) anterior wall to the septum. We analyzed transmural Lagrangian strain for 16 segments using a DENSE analysis framework developed in-house using MATLAB (R2010b, Mathworks, Natick, Massachusetts, USA). Images with abnormal phase wrapping were automatically excluded. Reproducibility has been previously reported [13].

One observer (#1) determined positive LGE on both a segmental level expressed as mean scar thickness, “transmurality”, and on a global level as a percentage of LVM (Segment v 1.9 R2966, Medviso AB, Lund, Sweden). The software suggests the delineation of scar using a weighted method that was manually corrected if needed. The weighted method involves the following steps: 1. The mean signal intensity and standard deviation are calculated in five sectors in each section. The midmural half of the sector with the lowest mean signal intensity is considered remote myocardium. 2. A section-specific threshold level is calculated by adding the mean of the remote sector and a fixed number of standard deviations from the mean signal intensity in the remote region. 3. A fast-level set algorithm is applied, and the speed term is set to a value calculated by subtracting the section-specific threshold level from the signal intensity [18].

In this study, LVEF was considered reduced when below 57% and the normal reference span for LVM index (LVMI) was 45–81 g/m² for males and 37–77 g/m² for females, as recommended when papillary muscles are excluded [19]. The presence of hypertension was determined when systolic and diastolic blood pressure exceeded 140/90 mmHg respectively, according to prevailing recommendations in 2010 [20].

The evaluation included two key components: (1) a global analysis of torsion and strain amplitude [Global circumferential strain (GCS), global radial strain (GRS)] compared to LVEF, LV volumes, LVMI, and LGE, (2) a segmental analysis of strain to detect segments with LGE transmurality > 50% using area under the curve (AUC) in a receiver operating characteristics (ROC) curve analysis.

The statistical analyses were performed using SPSS (Statistical Package for the Social Sciences, International Business Machines, Inc., Armonk, New York, USA). Analysis of skewness and kurtosis showed normal distribution, permitting Pearson correlation, Student’s t-test and Linear regression to be used. We calculated the sensitivity at 80% specificity from the ROC curve analysis. The level of significance was set to either **0.01 or *0.05.

Patients were separated into those with signs of myocardial disease (n = 97, 1552 segments) and those free from manifestations of myocardial disease (n = 19, 304 segments) determined from LVMI, LVEF, the presence of abnormal wall motion, positive LGE or a history of myocardial infarction (Table 1). The ten healthy subjects were on average 13 years younger than the patients, had similar LVMI but larger LV volumes. As expected, patients with myocardial disease had larger LV end-systolic volume than the reference group (p = 0.037) and their LVEF trended lower (p = 0.014). Thirty-four patients (84 segments) had LGE in excess of 50% wall thickness in at least one segment, and 37 patients had an LVEF below 57%. Patients with signs of myocardial disease had lower DENSE circumferential strain amplitude and higher standard deviation than those without myocardial disease manifestations. There is a significant difference in strain for segments in the group with no sign of myocardial disease compared to segments with > 50% LGE extent for all three techniques. The overlap between the segment groups was greater with FT than with the other techniques (Fig. 1). Figure 2 shows an example of a patient with a large infarct scar, with circumferential strain assessed by DENSE, FT and tagging in the corresponding segments.

GCS was for patients with no signs of myocardial disease (n = 19) − 16.1 ± 3.7% (DENSE), − 18.3 ± 1.6% (FT) and − 18.6 ± 2.8% (tagging). In the same group of patients torsion was 0.37 ± 0.09°/mm (DENSE), 0.12 ± 0.26°/mm (FT) and 0.31 ± 0.08°/mm (tagging). GCS, GRS and torsion results were correlated with other global parameters such as LVEF, blood pressure, LV volumes, LGE and LVMI, see Table 2. The correlations between GCS and GRS vs LVEF were r = 0.64 or r = 0.71 shown in Fig. 3. The correlation between GCS, GRS, torsion, and the global percentage of scar was low, but higher for GCS (Fig. 4). The best-fit linear regression between strain and scar size was for tagging GCS vs LGE r = 0.52 while DENSE GCS vs LGE was r = 0.46.

Segmental circumferential strain assessed by DENSE, FT, and tagging showed a low but significant positive correlation with segmental scar transmurality. Based on all 1856 segments from 116 patients, the correlation was r = 0.41 for DENSE (− 15.4 ± 5.2%), r = 0.15 for FT (− 16.8 ± 10.3%), and r = 0.37 for tagging (− 16.6 ± 6.5%). Segmental radial strain displayed low negative correlations with scar transmurality, r = − 0.10 (DENSE), r = − 0.07 (FT) and r = − 0.16 (tagging). When limited to segments with a transmurality > 50% per segment (n = 84, 34 patients), the correlation for radial strain was statistically significant for DENSE (r = 0.45, p < 0.01) and tagging (r = 0.31, p < 0.01) but not for FT (Fig. 5). ROC analysis of segmental circumferential strain for the detection of scar transmurality > 50% based on all segments in Fig. 6, (left), showed the largest area under the curve (AUC) for strain amplitude from DENSE, AUC 0.87 (FT 0.66, p < 0.001 and tagging 0.83, p = 0.03).When the specificity was set to 80%, the sensitivity to detect scar transmurality > 50% was 82%, 35%, and 71% for DENSE, FT and tagging circumferential strain, respectively. In Fig. 6 (right), use of GCS for the detection of any segment with LGE > 50% in an individual patient shows AUC for DENSE 0.78, for FT 0.69 and for tagging 0.69 without significant difference between the methods.

There were slight topographic differences when segmental circumferential strain was correlated with LGE transmurality in the basal, mid and apical LV slices. In the apex (464 segments), DENSE vs LGE correlated at r = 0.59, FT at r = 0.22 and tagging at r = 0.51; in midventricular segments (696 segments) the correlation with LGE was for DENSE r = 0.36, FT r = 0.15 and tagging r = 0.36. In the basal 696 segments the correlations with LGE were for DENSE r = 0.31, FT r = 0.08 and tagging r = 0.27. Thus, segmental circumferential strain from DENSE and tagging correlated best with segmental scar transmurality in the apical part of the LV.