Variability and homogeneity of cardiovascular magnetic resonance myocardial T2-mapping in volunteers compared to patients with edema

We scanned 73 healthy volunteers (13 female, 20–70 years, mean 35 ± 13 years, median 30 years, BMI 23 ± 3 kg/m2) without any cardiovascular disease, no symptoms of inflammation and a normal electrocardiogram. All participants were seen by a cardiologist. We discouraged alcohol intake one day before the scan to avoid inflammatory reaction [20]. Ten volunteers were scanned twice (time delay 469 ± 219 days, median 381 days) to assess interstudy variability.

We investigated a group of consecutive patients (n = 28; 8 females, age 55 ± 17 years, range 20–81 years) with acute myocardial damage. Edema was defined as a regional area of hyperintense signal on T2-weighted fast spin echo images corresponding to evidence of focal myocardial damage like wall motion abnormality or late gadolinium enhancement (LGE). The group comprises 20 patients with acute myocardial infarction (imaging 3 ± 1 days after admission), 5 patients with acute myocarditis (four male, median age 22 years, positive troponin and typical LGE lesions in all of them, CMR on 1 ± 1 day after admission), 2 postmenopausal female patients with Takotsubo cardiomyopathy (typical presentation and history, no scar, but transient apical ballooning) and 1 patient with cardiac sarcoidosis (acute admission with positive troponin, focal edema corresponding to typical LGE lesion).

Using a 1.5 T scanner (Magnetom Avanto, Siemens Erlangen, Germany, software version B17) with a 12-channel chest coil we obtained steady state free precession (SSFP) cine loops (repetition time 2.8 ms, echo time 1.2 ms, slice thickness 8 mm, flip angle 80 degrees, in-plane resolution 1.8 mm/pixel) during breathhold in three long and at least one midventricular short axis matching the slice position for mapping. T2-weighted STIR images (repetition time = 2 RR-intervals, echo time 58 ms, slice thickness 8 mm, in-plane resolution 1.3 × 1.3 mm/pixel, imaging in mid-diastole) were obtained in the same short axis as cine and additionally in long axis, if a focal abnormality was seen. In patients we additionally acquired fast low angle shot (FLASH) inversion recovery gradient echo LGE images in short and long axes (slice thickness 8 mm, in-plane resolution 1.8 × 1.4 mm/pixel) after 0.2 mmol/kg Gadolinium-DTPA.

For T2-mapping we applied a prototype sequence based on either FLASH or SSFP gradient echo [16] in one midventricular short axis (SAX) and four-chamber-view (4CV) matching the orientation of cine and STIR images. Slice thickness was 8 mm for all images. The pixel-wise map was based on three single shot images with preceding T2 preparation pulses employing T2 evolution times of 0, 24 and 55 ms (Figure 1). Read-out trajectory was centric for FLASH and linear for SSFP. A non-rigid motion correction was applied to reduce in-plane motion artefacts [17]. Pixel-wise curve fitting was done automatically as part of the inline imaging processing. Spatial resolution was 2.7 × 2.1 mm/pixel. Scan time was 12 heartbeats. The maximum acquisition window was 150 ms per single-shot image. Mid-diastole was chosen for image acquisition. In a subgroup of 24 volunteers additional SSFP-based maps with higher spatial resolution of 2.2 × 1.8 mm/pixel were acquired during the same scan. In patients image orientation was adjusted to the focal abnormality, e.g. in Takotsubo cardiomyopathy mapping was done in long axis. In 14/28 patients mapping data were acquired with the SSFP-based sequence only.

We scanned a spherical phantom of 20 cm diameter filled with manganese chloride doped distilled water. The RR-interval was simulated to be 1000 ms. A 2D multi contrast spin echo sequence (repetition time = 2000 ms, 32 echo times equally spaced from 7 ms to 224 ms) in conjunction with a mono-exponential three-parameter-fit served as T2 reference measurement [21]. We additionally obtained measurements using the FLASH and SSFP sequences with identical parameters as used for imaging volunteers and patients.

For signal analysis we used Osirix (version 3.9.1 http://​www.​osirix-viewer.​com) and QMASS (version 7, Medis, Leiden, The Netherlands) in all subjects. The endocardial and epicardial contours were manually drawn on the last corresponding T2-weighted raw image with the echo time of 55 ms. The myocardium was then segmented (manually in Osirix, automatically in QMass) into 6 segments according to the AHA segmentation scheme [22]. Contours were copied to the map, corrected when necessary and global and segmental T2-values were recorded. Two independent observers analysed all the volunteer data. Both do have considerable experience in CMR image analysis (> 15 years and > 3 years, respectively) and put much effort in avoiding inclusion of blood or fat while drawing regions of interest. Interstudy reproducibility was measured for SAX and 4CV. The coefficient of variation (CoV) was calculated as the ratio of the standard deviation of the interscan difference divided by the mean of the measurement. Anteroseptal enddiastolic myocardial wall thickness was measured on short axis and four-chamber-view cine frames and compared to T2-times in the same segment. To assess residual diastolic wall motion M-mode-like myocardial signal intensity projections over time were generated from 2D SSFP short axis and four-chamber-view cine images in selected patients using an in-house developed implementation in Matlab 7.1 (The Mathworks, Natick, MA).

The same investigator analyzed all SSFP volunteer maps twice. In patients areas of focal abnormality matching hyperintense signal in STIR and LGE were selected and compared to remote myocardium. We excluded hypointense infarction cores indicating microvascular obstruction [17].

Data are given as mean ± standard deviation unless indicated otherwise. We compared results from different sequences and orientations and repeated scans with paired student´s t-test and Pearson correlation coefficient r. P-values < 0.05 were considered significant. 95% confidence intervals (CI) were calculated. Bland-Altman-plots were obtained to analyse intra- and interobserver variability. Volunteer and patient data were compared using 95% tolerance intervals with 90% coverage. These tolerance intervals cover 90% of all observations of a normal distribution with 95% confidence when the mean and the standard deviation are known. Influencing factors were identified by sequence using the candidates age, gender, heart rate, body surface area, body mass index and the interaction terms in an analysis of variance with forward selection (criterion for selection p < 5%). We compared the data on healthy volunteers and patients using a mixed linear model with compound symmetry as working correlation matrix to account for the multiple measurements of up to four sequences by subject. Gender and age (35 years or younger, more than 35 years) were included as covariates. We performed a multivariate analysis including wall thickness, diastolic motion and a combination thereof to assess the impact on myocardial T2.

Evaluation of multi contrast spin echo data in the phantom yielded T2 = 62.0 ± 0.5 ms. The T2-mapping variant using FLASH readout resulted in T2 = 55.5 ± 1.0 ms, whereas SSFP readout revealed T2 = 70.9 ± 0.8 ms (Figure 2).

Four volunteers were excluded from analysis due to pathological findings (pleural effusions, tachycardic atrial fibrillation and left ventricular hypertrophy). The remaining 69 formed the healthy study group.

In all subjects maps could be generated in diagnostic quality. Mean heart rate was 73 ± 10 bpm. Global myocardial T2 measurements are summarized in Table 1. Global T2-values did not correlate with heart rate (r = 0.002), age (r = 0.30), body surface area (r = 0.44) or body mass index (r = 0.36). SSFP-based mapping resulted in higher values for T2 than FLASH-based mapping (SAX: 55 ± 5 vs. 52 ± 5 ms; 4CV: 59 ± 6 vs. 57 ± 6 ms; p < 0.0001 for both). This was true for global (Table 1) and segmental measurements (Figure 3). Global T2-mapping resulted in higher values in 4CV-orientation than in short axis (p < 0.0001 for both sequences; Table 1).

With both sequences anteroseptal segments had higher T2-values than inferior segments in SAX (59 ± 6 vs. 48 ± 5 ms for FLASH and 59 ± 7 vs. 52 ± 4 ms for SSFP; p < 0.0001 for both, Figure 3). In 4CV the apical septal segment had higher T2 than the basal lateral segment for both FLASH (61 ± 8 vs. 55 ± 6 ms; p < 0.0001) and SSFP (64 ± 9 vs. 55 ± 6 ms; p < 0.0001; Figure 3). In SAX the mean absolute difference between a single segment and the whole slice was 4 ± 1 ms and 3 ± 2 ms for FLASH and SSFP, respectively. In 4CV the mean absolute difference between a single segment and a global measurement was 5 ± 2 ms for both, FLASH and SSFP.

Intra- and inter-observer variability was low (Figure 4). The mean difference for T2 between repeated measurements in one observer was 1.07 ± 1.03 ms (r = 0.94). Mean difference for T2 between two observers was 1.6 ± 1.5 ms (r = 0.87). The CoV for repeated scans was 7.6% for SAX and 6.6% for 4CV. The results for myocardial T2 did not differ depending on the analysis software used.

Even among the normal volunteer group we found segmental and global T2-values equal to or higher than 70 ms (Figure 5) in 14 of 69 healthy subjects. On a segmental basis this occurred in 3/414 segments in FLASH and 7/414 segments in SSFP in short axis. In 4CV maps it occurred in 26/414 segments in FLASH and 29/414 in SSFP. In 13 out of 14 volunteers the apical septal segment was affected. In 4CV there were 4 volunteers each, who had global myocardial T2 ≥70 ms in FLASH or SSFP. There was no significant difference in age between those with and without T2 ≥70 ms (31 ± 12 vs. 38 ± 13 years; p = 0.08). Anteroseptal wall thickness as measured in enddiastolic cine frames (mean 3.5 mm, range 2–8 mm for 4CV) was inversely related to myocardial T2. The thinner the anteroseptal segment was, the higher was T2 (r = −0.6 for 4CV). The m-mode analysis revealed a mean residual diastolic motion of 2.3 mm (range 0.1 – 5.2 mm). More diastolic motion correlated with higher myocardial T2 (r = 0.37). In a multivariate analysis diastolic wall motion (p < 0.001) and the combination of wall thickness and motion (p < 0.003) significantly correlated with myocardial T2. Dichotomizing the volunteer group for age and gender reveals the larger range for myocardial T2 in younger and mainly female volunteers (Figure 6).

Mapping with improved spatial resolution did change results in individual subjects (Figure 7), but not for the whole group (p = 0.3) Even after improving spatial resolution 8 out of 24 volunteers had segmental myocardial T2 ≥70 ms on 4CV maps, while only one segment was affected in SAX. In 4CV this occurred in the apical septal segment in all but one case.

In all 28 patients with acute myocardial damage areas of increased T2 could be detected. Using SSFP-based mapping T2 of edema was 73 ± 9 ms on average (range 64–99 ms, 95% CI 70–76 ms), while T2 of remote myocardium was 51 ± 3 ms (range 44–57 ms, 95% CI 50–52 ms; Figure 8). T2 did not differ between ischemic (Figure 9) and nonischemic (Figure 10) edema.

The power calculation took into account the differentiation of patients and volunteers as a whole group. The study and its sample size were not designed to reveal subtle differences among e.g. women of different decades of age. There is no dedicated gold standard for true edema. We assumed edema in myocardial areas where it made sense clinically based on concomitant late gadolinium enhancement or obvious wall motion abnormalities in patients with evidence of acute myocardial damage in laboratory results and electrocardiogram.