Comparison of different cardiovascular magnetic resonance sequences for native myocardial T1 mapping at 3T

T1 mapping, which refers to the pixelwise quantification of the myocardial longitudinal relaxation time T1, is a rapidly emerging Cardiovascular Magnetic Resonance (CMR) technique which has demonstrated a strong potential to characterize abnormalities related to acute and chronic myocardial injury [20]. Global T1 measurements can be determined either by analyzing the intensity of a group of pixels in a parametric T1 map, but also by deriving an exponential T1 recovery curve based on ROIs from a number of steady-state free precession (SSFP) images with varied inversion times. [20]. Among the applied sequences, inversion-recovery techniques based on Look-Locker protocols (MOLLI, ShMOLLI) and saturation-recovery sequences (SASHA) have been used most frequently. Yet, more than 10 years after the first clinical study on myocardial T1 mapping [18, 19] and numerous reports on its utility for identifying different processes such as edema [8, 10] and diffuse fibrosis [2], the specific clinical utility of currently used protocols is not well understood. Specifically, there is a paucity of comparative data between these approaches beyond phantom experiments [13], especially in a suitable reference middle-aged population with low to moderate cardiovascular risk.

The aim of our study was to compare the ability of three major T1 mapping techniques, ShMOLLI, MOLLI 5(3)3 and SASHA, to determine the native relaxation T1 times in phantoms and in subjects from a population-wide cohort.

Recorded conditions at inclusion of the subjects in the database were: absent (62.5 %), negative ischemic evaluation (22.5 %), rhythm disorder (10 %) or other (5 %). Demographic characteristics, clinical and CMR results are presented in Table 1. We analyzed the different added-risk groups: normal volunteers (without any risk factors); low added-risk group (presence of 1–2 risk factors, without known hypertension); and hypertensives (with or without at most two other risk factors on top of hypertension). The hypertensive group, mostly constituted of moderate added-risk patients, had higher prevalence of dyslipidemia (p < 0.01), while the low added-risk group had a higher prevalence of a family history of premature CV disease (p < 0.01). The latter had also higher prevalence of risk factors (p < 0.001), but in 78 % of the cases only one risk factor was present. No other differences were found between groups, namely in routine MR parameters or T1 values within sequences.

Our results indicate that, despite excellent inter-observer variability and good in-vitro correlation of various T1 mapping sequences, results vary in vivo between saturation-recovery sequences and inversion-recovery approaches. Furthermore, saturation-recovery sequences may provide less homogenous parametric maps (Fig. 5).

Native T1 of tissue in vivo can be estimated from signal intensities in images with varying inversion or saturation times through calculating an exponential T1 recovery curve [20]. The original Look-Locker technique was modified to a single-breath-hold technique with steady-state free precession (SSFP) readouts, known as Modified Look-Locker Inversion recovery (MOLLI) sequences [19]. This approach was further modified by Piechnik et al. with a Shortened adaptation with conditional curve fitting/ShMOLLI [23], allowing shorter breath-holds and less heart rate susceptibility. Modified MOLLI versions with fewer off-resonance artifacts were then necessary [14]. However, all these techniques still require a correction factor to account for underestimation of T1 due to magnetization attenuation [9]. Furthermore, T2 affects the results and, while T1 and T2 changes typically are correlated, the T1 acquired by MOLLI techniques may not represent a “true” T1. Chow et al. added a protocol based on images with varying saturation times, called SAturation recovery Single-sHot Acquisition (SASHA) sequence [5], which showed good accuracy [13]. No correction factor is necessary and there is no significant impact of tissue T2, but the curve fitting assumes a fully saturated tissue environment, which may not be present. Hybrid sequences are also available, that integrate some of the advantages of saturation and inversion-recovery [31].

All these techniques have different issues and while a consensus group is working on standardizing terminology and techniques [20], problems remain regarding the standardized use of techniques and the lack of accepted reference values. While some diseases such as cardiac amyloidosis or acute myocardial infarction show a strong increase in native T1, more subtle differences are obtained in conditions with less myocardial damage such as valvular heart disease with LV hypertrophy [27]. Thus, even small differences between calculated results limit the clinical utility of T1 mapping, if sequences and evaluation procedures are not standardized.

Reported reference values for the different techniques were acquired in studies with variable sample sizes and for different field strengths [7, 24, 30]. MOLLI sequences result in reference values (mean ± 2*SD) of 900 and 1050 ms at 1.5 T, while at 3 T myocardial T1 was found to be between 1100 and 1250 ms [21, 22]. Our results, obtained in a cross-sectional, middle-aged population at a low to moderate risk, are similar. Furthermore, we did not find a significant correlation with the cardiovascular risk profile, which is consistent with two recent studies, one by Dabir et al. [7], where a low-risk population had similar T1 values as normal volunteers, and one by Sado et al. with a reported overlap of T1 times of normal subjects and patients with arterial hypertension [27].

In our study, T1 as estimated by ShMOLLI and MOLLI was significantly lower than estimated by SASHA. While there are no data on reference values for T1 measurements based on SASHA at 3T, the observed difference is consistent with a previous review by Kellman and Hansen [13], who found a difference of about 200 ms between inversion and saturation-based methods, when determined at 1.5T. We observed a similar difference in the phantom experiments, which was amplified on the in-vivo measurements. In their review, Kellman and Hansen discuss a variety of factors that may affect T1 mapping accuracy, such as patient and tissue characteristics, protocol parameters, scanner and sequence adjustments, and fit models. In our sample the heart rate during SASHA acquisition was not different from the other sequences, being low (66.5 ± 12.1) and very close to the heart rate used in the phantom experiments, where there was not a wider difference in T1 estimation between sequences. Other tissue characteristics may further impact results relative to the phantom models, mostly in inversion-recovery techniques, leading to lower T1 values. In particular, magnetization transfer (MT) may be responsible for about 15 % of the assumed underestimation of T1 times with MOLLI, while this confounding effect is reduced with SASHA, a sequence that is less dependent on T2, at least when using a 3-parameter fit [25]. Magnetization transfer may therefore have been one of the major contributors to the inter-sequence differences. In our case, both scanner and sequence parameters were kept unchanged throughout all experiments. The SASHA fitting routine assumes perfect saturation (90° flip angle), but B1+ inhomogeneities and variations in the transmit gain could produce deviations from 90° and result in a T1 bias. Although using different sequences, Stikov et al. [29] reported similar findings in the brain, suggesting improper B1 correction as the reason for poor agreement between sequences in vivo (with variable flip angles), but not in phantoms. Importantly, a variable flip angle approach for SASHA has been reported recently as a means of reducing systematic errors [6].

The lower precision of SASHA, i.e. a higher variability of results, is a known disadvantage [26]. We also observed a marked heterogeneity of T1 compared with the other sequences. Different curve fit algorithms or sampling schemes may reduce this problem [15]. Our results did not seem to be related to artifacts in the SSFP readouts. For example, even though SASHA had fewer susceptibility artifacts in the SSFP images, the resulting T1 maps were of lower quality than maps based on MOLLI.

We used a ROI-based analysis for T1 measurements because the ROI-based approach allows for corrections of motion-correction imperfections or problems with cardiac phase registration. The pixel-map based approach is however a fair alternative, as shown in our study. The maps, as main overall advantage, enable a qualitative interpretation, priceless on clinical evaluation. That type of interpretation is quite difficult when using the SSFP readouts of the ROI-analysis, due to the aggressive windowing needed to define the myocardium on images with different T1s (shades of gray). We found the qualitative interpretation to be more frequently achieved with the inversion-recovery sequences, due to the referred homogeneity issues we encountered with SASHA’s maps. The maps further allow the analysis of myocardial segments. We confirmed the findings of other groups, with lower T1 at the anterior segments and higher values in the inferior/infero-lateral segments [30]. While a physiologic phenomenon cannot be completely excluded, it more likely represents an artifact.

Our study had several limitations. The sample is small and heterogeneous, and suffered further from missing heart rate data from the scanner. Even though we excluded those values from the analysis, the pixel-map based analysis from the Siemens inline generated maps, could have allowed for T1 estimation when we experienced this error (average T1 on 7 out of 8 cases: 1176 ± 18 ms). Nevertheless, the impact of cardiovascular risk on T1 has to be studied in larger samples, to confirm the significant differences of measured T1 between sequences. This is a field at constant development, and any analysis may risk to become outdated by the time of reporting. Marked improvements have, in fact, been implemented on some of these sequences, specifically a variable flip angle approach for SASHA that may have led to smaller observed differences between sequences and reduced homogeneity issues with that sequence. We only acquired a single myocardial slice and variations may be more or less pronounced in basal and apical slices. Yet, the observed differences and our conclusions on clinical utility of T1 mapping sequences remain valid.

While performing T1 mapping in phantoms and in a cross-sectional population with low to moderate cardiovascular risk, we observed a small interobserver variability of T1 measurements. The MOLLI sequence showed the smallest overall variability and SASHA the best accuracy. Most importantly, measured T1 varied between inversion-recovery vs. saturation-recovery sequences in vivo, much more than in phantom images. Thus, physicians applying or interpreting the technique should be aware that phantom validation does not directly translate to in vivo application, because physiological factors, most likely related to B1 inhomogeneities, the position of the heart in the coil reception field, T2 sensitivity, magnetization transfer effects and physiologic variations impact on T1 values (whether between different methods like IR vs SR, or within a single method), and have therefore to be further investigated for a better understanding of the clinical utility of different T1 mapping approaches.