Author + information
- Received April 18, 2012
- Revision received June 11, 2012
- Accepted June 14, 2012
- Published online December 1, 2012.
- Ruud B. van Heeswijk, PhD⁎,†,‡,⁎ (, )
- Hélène Feliciano, MSc⁎,†,‡,
- Cédric Bongard, BSc§,
- Gabriele Bonanno, MSc⁎,†,‡,
- Simone Coppo, MSc⁎,†,‡,
- Nathalie Lauriers, MSc§,‖,
- Didier Locca, MD§,‖,
- Juerg Schwitter, MD§,‖ and
- Matthias Stuber, PhD⁎,†,‡
- ↵⁎Reprint requests and correspondence:
Dr. Ruud B. van Heeswijk, University Hospital of Lausanne (CHUV), Center for BioMedical Imaging (CIBM), Rue de Bugnon 46, Lausanne, VD 1011, Switzerland
Objectives This study sought to establish an accurate and reproducible T2-mapping cardiac magnetic resonance (CMR) methodology at 3 T and to evaluate it in healthy volunteers and patients with myocardial infarct.
Background Myocardial edema affects the T2 relaxation time on CMR. Therefore, T2-mapping has been established to characterize edema at 1.5 T. A 3 T implementation designed for longitudinal studies and aimed at guiding and monitoring therapy remains to be implemented, thoroughly characterized, and evaluated in vivo.
Methods A free-breathing navigator-gated radial CMR pulse sequence with an adiabatic T2 preparation module and an empirical fitting equation for T2 quantification was optimized using numerical simulations and was validated at 3 T in a phantom study. Its reproducibility for myocardial T2 quantification was then ascertained in healthy volunteers and improved using an external reference phantom with known T2. In a small cohort of patients with established myocardial infarction, the local T2 value and extent of the edematous region were determined and compared with conventional T2-weighted CMR and x-ray coronary angiography, where available.
Results The numerical simulations and phantom study demonstrated that the empirical fitting equation is significantly more accurate for T2 quantification than that for the more conventional exponential decay. The volunteer study consistently demonstrated a reproducibility error as low as 2 ± 1% using the external reference phantom and an average myocardial T2 of 38.5 ± 4.5 ms. Intraobserver and interobserver variability in the volunteers were –0.04 ± 0.89 ms (p = 0.86) and –0.23 ± 0.91 ms (p = 0.87), respectively. In the infarction patients, the T2 in edema was 62.4 ± 9.2 ms and was consistent with the x-ray angiographic findings. Simultaneously, the extent of the edematous region by T2-mapping correlated well with that from the T2-weighted images (r = 0.91).
Conclusions The new, well-characterized 3 T methodology enables robust and accurate cardiac T2-mapping at 3 T with high spatial resolution, while the addition of a reference phantom improves reproducibility. This technique may be well suited for longitudinal studies in patients with suspected or established heart disease.
The T2 relaxation time is a physiological tissue property that can be exploited with cardiac magnetic resonance (CMR) to generate contrast between healthy and diseased tissues. This contrast is mainly caused by the dependency of the T2 value on the relative amount of free water (1). Edema is part of the tissue response to acute injury and affects this free water content. Therefore, T2 changes have been reported in edematous regions in patients with infarction (2), hemorrhage (3), graft rejection (4), or myocarditis (5). In recent years, qualitative T2-weighted CMR has therefore gained considerable interest. However, the traditional dark-blood T2-weighted fast spin echo (FSE) pulse sequence that is used for this purpose is limited because of its motion sensitivity and subsequent risk for misinterpretation of the images. Simultaneously, a quantitative characterization of the tissue is not easily possible and image interpretation remains subjective. Therefore, a more objective, quantitative, and motion-insensitive technique is required. In response to this strong need, initial T2-prepared variants of balanced steady-state free precession sequences have been proposed for quantitative T2-mapping at 1.5 T (6). Using such methods, the successful differentiation between edematous and healthy tissue after myocardial infarction has been demonstrated (7), and an improved performance relative to conventional FSE imaging was reported in both patients with edema after myocardial infarction (8) and acute inflammatory cardiomyopathies (9).
The availability of a quantitative, accurate, and highly reproducible T2-mapping methodology at 3 T would be of great importance for the use in longitudinal studies aimed at monitoring and guiding therapy, because a T2 value measured within a specific target area could act as its own control measurement. However, to our knowledge both the accuracy and reproducibility of T2-mapping have not been ascertained. For these reasons, we have developed and tested a free-breathing T2-mapping technique at 3 T that incorporates radial gradient echo (GRE) image acquisition and adiabatic T2 preparation (T2prep-GRE). Bloch equation simulations were performed to optimize both sequence parameters and the analysis procedure. The resultant magnetic resonance methodology was then validated in vitro. Quantitative results were compared with those of a gold-standard spin-echo T2-mapping sequence to determine the accuracy of the T2 measurements. The reproducibility of the technique was then investigated in healthy volunteers, both in separate scanning sessions and with and without a T2 reference phantom positioned in the field of view. Using this setup, the hypothesis was tested that the use of a reference phantom improves reproducibility of the T2-mapping. Finally, the thus-optimized and characterized methodology was applied to test the ability to discern healthy from diseased myocardium in patients with established subacute myocardial infarction.
The goal of these simulations was to maximize the amount of signal per unit time while establishing optimal fit parameters to increase the accuracy of the T2 measurement. Therefore, a numerical simulation of the Bloch equations (10) was performed using Matlab (The Mathworks, Natick, Massachusetts). Simulation parameters included myocardial relaxation times T2 = 45 ms and T1 = 1,470 ms (11) at 3 T; a segmented k-space radial GRE acquisition with a repetition time (TR) of 7.6 ms and an echo time (TE) of 2.8 ms; a navigator delay of 40 ms; and incremental T2prep durations (TET2prep) of 0, 30, and 60 ms for T2 fitting. The average transverse magnetization (Mxy) of radial readouts during 27 heartbeats was then considered representative for the resultant Mxy for a given T2prep duration. To determine the fitting equation that leads to highest accuracy, the magnetization Mxy for 3 TET2prep values was fitted with both a standard exponential decay and an empirical equation:where M0 refers to the longitudinal magnetization at TET2prep = 0 and δ is an empirical offset that accounts for T1 relaxation. Independent variables that were used to study the quality and robustness of the fit and to maximize Mxy included heart rate, radiofrequency excitation angle (α), the number of acquired radial profiles in k-space per heartbeat and the number of RR intervals between acquisition trains. After having selected the parameter set that led to a maximum Mxy, the range of stability of the T2 fitting algorithm was determined for both the standard and empirical equation in a T2 range from 1 to 250 ms, which sufficiently covers physiological T2 values expected at 3 T.
Implementation and imaging sequence
T2prep-GRE was implemented on 2 3 T magnetic resonance scanners (Magnetom Trio and Verio, Siemens Healthcare, Erlangen, Germany) with a 32-channel chest coil (Invivo, Gainesville, Florida) and with sequence parameters as described. Because T2 preparation at high magnetic field strength is susceptible to B1 inhomogeneity (12), an adiabatic T2prep (13) with user-specified TET2prep preceded the imaging part of the sequence that had a temporal resolution of 97 ms and a spatial resolution of 1.25 × 1.25 × 5 mm3. For respiratory motion suppression during free breathing, a lung-liver respiratory navigator (14) was used. For each T2 map, the imaging sequence was repeated with 3 incremental TET2prep (0, 30, and 60 ms). After acquisition of the 3 source images, affine coregistration (15) was applied to increase the accuracy of the T2-mapping before the final pixel-by-pixel computation of the T2 maps was performed using a custom-written Matlab analysis tool in which the optimized Eq. (1) was incorporated.
Seventeen phantoms with different T2 values that consisted of varying concentrations of NiCl2 and agar (16) together with sodium azide as a antimicrobial preservative were constructed and T2 maps were generated with the T2prep-GRE sequence to assess the performance of the pulse sequence and to validate the results of the simulations. A spin-echo sequence with 8 to 11 incremental TE (TE = 4 to 500 ms, TR = 5 s) was used to define the gold standard T2, whereas an inversion recovery spin-echo sequence with 8 to 11 inversion times (TI = 14 to 3,000 ms, TR = 7 s) was used to determine the gold standard T1. To characterize the accuracy and precision of the T2prep-GRE–derived T2 values using Equation 1, a linear correlation with the spin-echo gold standard T2 values was performed. To ascertain whether the phantom T2 values are subject to change as a function of time, the T2 values of the phantoms were determined monthly up to 6 months after their construction.
Permission from the Institutional Review Board was obtained for all volunteer and patient scans, and written informed consent was obtained from all participants prior to the procedure. To characterize the performance of the T2prep-GRE T2-mapping methodology for longitudinal studies, 10 volunteers (6 men, age: 27 ± 4 years) underwent 2 separate scanning sessions with an identical protocol. Between the sessions, the volunteers were extracted from the scanner room. To obtain an external reference standard T2 value in each measurement, a phantom with known T1 and T2 values (see Phantom studies section) similar to those of the healthy myocardium (11) was positioned in the field of view. After shimming of the heart based on a local gradient-echo field map (17), T2 maps were obtained in a short-axis view.
To test the hypothesis that the external reference phantom leads to an improved reproducibility of the T2-mapping protocol, the 2 scanning sessions were compared as follows. The entire left ventricular myocardium in the image and a homogeneous and central area of the phantom were manually segmented by 2 experienced observers (R.B.v.H., C.B.), and their average T2 was directly (without the use of the external reference phantom) calculated (T2myo,dir and T2phan,dir). Using the “true,” known T2 value of the phantom T2phan,true as determined with the spin-echo sequence described earlier, a corrected myocardial T2 value T2myo,corr was calculated:The percentage difference between T2myo,dir and T2myo,corr for both scanning sessions as well as the intraobserver and interobserver variability were then calculated.
As a next step, the optimized and validated methodology described earlier was used in 11 patients (9 men, age: 50 ± 13 years) in the unique setting of subacute phase after percutaneous coronary revascularization of an acute ST-segment elevation myocardial infarction (STEMI). Short-axis T2 maps at a mid-ventricular level were acquired in all patients together with qualitative breath-hold black-blood T2-weighted FSE images (TR/TE = 2,540/70 ms, echo train length = 17).
After calculating the T2 maps in these patients, the myocardium and the reference phantom were manually segmented. The average T2 values and standard deviations were subsequently determined in both regions of interest and were compared with the values obtained in healthy volunteers. The tissue with elevated T2 values was considered as being the infarcted region. Simultaneously, a more objective and automated selection of the region of elevated T2 was selected by only including pixel T2 values that were 3 SD above the average T2 value of the healthy myocardium. In both the T2 maps and the T2-weighted FSE images, the center of the segmented left ventricle was selected by the user and the radial extent of the infarction was manually determined as the edge of the continuous spread of the automatically detected elevated T2 values, after which a linear regression of the spread in the 2 image types was performed. The automatically selected regions of infarct on the T2 maps were then also related to the location of the luminal narrowing by x-ray angiography, where available.
All statistical tests were paired or unpaired (as applicable) 2-tailed Student's t tests, where p < 0.05 was considered statistically significant. Correlations between continuous variables were calculated with the Pearson correlation coefficient r. Coefficients of determination R2 were calculated for all linear regressions through the origin. Intraobserver and interobserver variability were calculated by Bland-Altman analysis (18).
Numerical simulations of the Bloch equations for the pulse sequence resulted in maximum signal per unit time for a radiofrequency excitation angle of 20° and 21 radial k-space lines per heartbeat. The empirical fitting equation led to the most accurate T2 determination if the offset delta was set to 0.06 (Fig. 1A). In contrast, when the conventional exponential curve fitting procedure was applied to the simulated magnetization, the T2 value was always overestimated by ∼12% (Fig. 1B). These findings were consistent over a broad range of simulated T2 values (Fig. 1C). Further numerical simulations suggested a minor heart rate dependency of the T2 measurements relative to 60 beats/min with a 2.2% underestimation at 90 beats/min and a 1.5% overestimation at 40 beats/min.
An excellent agreement between the T2 maps generated with conventional FSE and the T2prep-GRE method that incorporates the empirical equation was found in the phantom study (Fig. 2). With a correlation r = 0.996 and a slope of 0.97, it was found that T2 computation using the proposed T2-mapping methodology is accurate and precise over a large range of T2. When comparing the T2 values of the phantoms that were measured 6 months apart, no significant change was observed (p = 0.83), and the maximum difference that was measured in a phantom over time was 1.5 ms.
The optimized protocol was successfully applied in all 10 healthy volunteers (Fig. 3). The directly calculated myocardial T2myo,dir varied 4 ± 2% on average between the 2 scanning sessions, while the corrected myocardial T2myo,corr obtained using the external reference phantom varied significantly less with 2 ± 1% (p = 0.005) between scanning sessions (Fig. 3C). Averaged over all volunteers, T2myo,dir was 41.2 ± 4.1 ms, whereas the average T2myo,corr was 38.5 ± 4.5 ms (p = 0.07).
The intraobserver mean difference for T2myo,corr was –0.04 ms (95% confidence interval [CI]: –1.2 to 0.6 ms, p = 0.86), while the interobserver mean difference was –0.4 ms (95% CI: –1.2 to 0.4 ms, p = 0.87) (Fig. 4).
The T2prep-GRE T2-mapping protocol was successfully applied and T2 maps generated in all 11 STEMI patients, whereas respiratory motion artifacts led to lower-quality T2-weighted FSE images in 3 of these cases, of which 1 was excluded from further analyses. On the T2 maps of the remaining cases, a clear demarcation of regions with elevated quantitative T2 values visually coregistered with the findings on T2-weighted CMR and x-ray coronary angiography as shown in Table 1 and the example in Figure 5. The average T2 over all patients in the healthy remote region was 41.5 ± 3.6 ms. This was statistically significantly higher than that in healthy volunteers (38.5 ± 4.5 ms; p = 0.04), although it should be noted that this average included 3 severe STEMI patients in which the T2 value of the healthy remote segment was measured higher than 50 ms. The average manually determined T2 value in the infarcted regions was 61.2 ± 10.1 ms, while the automatic method resulted in 62.4 ± 9.2 ms (p = 0.27). There was a good overall correlation between the manually and automatically determined T2 values (r = 0.91, slope = 1.01, R2 = 0.77). Furthermore, the circumferential location of the signal-enhanced regions by T2-weighted FSE imaging and increased T2 values by T2-mapping visually agreed very well and corresponded with the myocardial segment that was supplied by the vessel that had a stenosis on the corresponding x-ray angiograms (Fig. 6).
A linear regression of the radial spread in the T2-weighted images and T2 maps, as illustrated in Figure 7, demonstrated a slight increase of the radial spread in the images obtained with T2-mapping (r = 0.92, slope = 0.89, R2 = 0.80).
The presented T2prep-GRE methodology accurately and reproducibly enables T2-mapping at 3 T during free breathing. A series of incremental steps were essential and enabling for the translation from theory to the patient setting.
Both the empirical equation that was established using Bloch equation simulations and the standard exponential decay model led to an equally good fit. However, because the latter does not take T1 relaxation into account, a consistent ∼12% T2 overestimation was observed, while the use of the empirical equation resulted in a <1% T2 underestimation only.
The phantom experiments confirmed that the use of the optimized 3 T methodology resulted in accurate T2 measurement relative to conventional spin-echo measurements as the gold standard. However, for maximized performance of the technique and definition of δ, the T1 of the measured tissue has to be known. This raises concern as the T1 value of the myocardium may be subject to change. It has been reported (19) that the T1 of healthy and infarcted myocardium may differ by 18%. Such a change in T1 would result in a 2.8% underestimation of the T2 value according to our Bloch equation simulations, which seems acceptable given the standard deviation in T2 measurements of 6% to 10% in this study. If the phantoms are to be used in longitudinal studies, the T1 and T2 values need to be constant over time. To this end, the antimicrobial sodium azide was added, and it was confirmed that no significant changes in T2 were detected over the course of 6 months.
The methodology was further characterized in an in vivo healthy volunteer study where its effectiveness and reproducibility were evaluated. Adding the reference phantom allowed for the compensation of drift between scans. The interobserver and intraobserver variability of the corrected T2 values were similar to those reported in related studies (8,9) at 1.5 T. While the T2 values of healthy myocardium were consistent with those reported in the literature (20), the addition of a reference phantom significantly aided in the reduction of the difference in myocardial T2 values between 2 scanning sessions. Such T2 value differences may occur due to slight changes related to B0 and B1 inhomogeneity, the relative accuracy of the fitting procedure, coil placement, among others. Furthermore, and consistent with prior reports that established T2-mapping at lower field strength (6,7), only 3 points were used for the mono-exponential 2-parameter fit for the T2 determination in this study. Although more points may result in an improved accuracy and robustness of the procedure, this remains to be investigated and has to be carefully balanced versus an increase in scanning time.
In the small cohort of 11 STEMI patients, the quantitative T2 values of the edematous regions (defined on conventional T2-weighted imaging) showed an increase of approximately 50% relative to their healthy remote counterparts in all cases. This also enabled a robust automated detection of these regions that correlated well with the more subjectively selected user-specified regions of T2 enhancement. The T2 of the healthy remote segments in the patients was slightly but significantly higher than that found in healthy volunteers. However, the study was not age-matched and an age-dependent increase in T2 between the studied cohorts cannot be excluded.
The circumferential location of elevated signal on T2-weighted images and x-ray angiograms agreed very well, as did the comparison of the radial spread of the edematous region as determined through T2-mapping and T2-weighted imaging, which was expected because myocardial contrast in both modalities is based on the degree of edema. However, T2-weighted imaging only defines presence and extent of elevated T2, while T2-mapping is quantitative and may therefore provide a very important quantitative endpoint for many studies related to cardiovascular disease.
In the 3 severe STEMI cases, the finding that the measured T2 value of the reference phantom was unchanged relative to the gold standard measurements improved confidence that unusually high T2 values (∼50 ms) were indeed found in the unaffected, “healthy” remote myocardial tissue. While the use of an external reference phantom was originally designed to improve interscan reproducibility, this suggests that it may equally benefit the accuracy of a single study in patients where the overall T2 value of the entire myocardium is elevated. Example applications include studies in myocarditis, heart failure, or transplant patients.
The methodology presented in this study enables robust and accurate quantitative cardiac T2-mapping at 3 T, while the addition of a reference phantom improves reproducibility. Therefore, it may be well suited for longitudinal studies in patients with ischemic heart disease.
This work was supported by the Centre d'Imagerie BioMédicale (CIBM) of the University of Lausanne, the University of Geneva (UNIGE), the University Hospital of Geneva (HUG), the University Hospital of Lausanne (CHUV), the Federal Institute of Technology of Lausanne (EPFL), and the Leenaards and Jeantet Foundations, as well as the Emma Muschamp Foundation. The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- cardiac magnetic resonance
- fast spin echo
- gradient echo
- ST-segment elevation myocardial infarction
- T2 preparation module
- echo time
- T2prep duration
- inversion time
- repetition time
- Received April 18, 2012.
- Revision received June 11, 2012.
- Accepted June 14, 2012.
- American College of Cardiology Foundation
- Payne A.R.,
- Casey M.,
- McClure J.,
- et al.
- Eitel I.,
- Kubusch K.,
- Strohm O.,
- et al.
- Zagrosek A.,
- Abdel-Aty H.,
- Boye P.,
- et al.
- Verhaert D.,
- Thavendiranathan P.,
- Giri S.,
- et al.
- Thavendiranathan P.,
- Walls M.,
- Giri S.,
- et al.