T1-mapping in the heart: accuracy and precision

Abstract

The longitudinal relaxation time constant (T1) of the myocardium is altered in various disease states due to increased water content or other changes to the local molecular environment. Changes in both native T1 and T1 following administration of gadolinium (Gd) based contrast agents are considered important biomarkers and multiple methods have been suggested for quantifying myocardial T1 in vivo. Characterization of the native T1 of myocardial tissue may be used to detect and assess various cardiomyopathies while measurement of T1 with extracellular Gd based contrast agents provides additional information about the extracellular volume (ECV) fraction. The latter is particularly valuable for more diffuse diseases that are more challenging to detect using conventional late gadolinium enhancement (LGE). Both T1 and ECV measures have been shown to have important prognostic significance. T1-mapping has the potential to detect and quantify diffuse fibrosis at an early stage provided that the measurements have adequate reproducibility. Inversion recovery methods such as MOLLI have excellent precision and are highly reproducible when using tightly controlled protocols. The MOLLI method is widely available and is relatively mature. The accuracy of inversion recovery techniques is affected significantly by magnetization transfer (MT). Despite this, the estimate of apparent T1 using inversion recovery is a sensitive measure, which has been demonstrated to be a useful tool in characterizing tissue and discriminating disease. Saturation recovery methods have the potential to provide a more accurate measurement of T1 that is less sensitive to MT as well as other factors. Saturation recovery techniques are, however, noisier and somewhat more artifact prone and have not demonstrated the same level of reproducibility at this point in time.This review article focuses on the technical aspects of key T1-mapping methods and imaging protocols and describes their limitations including the factors that influence their accuracy, precision, and reproducibility.

Figures

Figure 1

MOdified Look-Locker Inversion Recovery (MOLLI) scheme for T1-mapping in the heart [32]. The original protocol employed 3 inversions with 3, 3, and 5 images acquired in the beats following inversions, and 3 heart beat recovery periods between inversions, referred to here as 3(3)3(3)5. All images are acquired at the same delay from the R-wave trigger for mid-diastolic imaging. Curve fitting is performed on a pixel-wise basis using the actual measured inversion times.

Figure 2

The apparent inversion recovery (T1*) is influenced by the SSFP readout. The effective inversion recovery is fit using a 3-parameter model, and the T1 is estimated using the so-called Look-Locker correction.

Figure 3

SAturation recovery Single Shot Acquisition (SASHA) scheme for T1-mapping in the heart [44]. A single image is acquired without saturation and used as the fully recovered measurement followed by a series of saturation recovery images at different saturation recovery times (TSi). All images are acquired at the same delay from the R-wave trigger for mid-diastolic imaging. Curve fitting is performed on a pixel-wise basis.

Figure 4

Illustration of accuracy versus precision. Accuracy refers to systematic errors, which create a bias, whereas, precision relates to the random component due to noise (http://www.jcmr-online.com/content/15/1/56/figure/F1).

Figure 5

The estimate of T1 using SSFP based MOLLI 5s(3s)3s is sensitive to T2 with increased underestimation error at lower values of T2 which results in a T1 map which has a small degree of T2 weighting. For native myocardium, an increase of 100% in T2 from 45 to 90 ms results in an increase in the apparent T1 of approximately 4%.

Figure 6

Simulated off-resonance response of MOLLI for 5s(3s)3s protocol with TR = 2.8 ms using FA = 35 (left) and 25 (right) at various T1’s for myocardial T2 = 45 ms. Using a lower flip angle (FA) trades SNR (precision) for improved accuracy and reduced off-resonance sensitivity. For T1 = 1000 ms, sensitivity to off-resonance over ±100 Hz is 40 ms and 25 ms for FA = 35° and 20°, respectively.

Figure 7

T1-maps acquired at different center frequencies using MOLLI 5s(3s)3s at 3 T. Despite the use of a 2nd order shim in a local volume around the heart, off-resonance variation across the heart, as seen in the field map (left), leads to an artifactual local variation in the apparent T1 as indicated by arrows [42]. (adapted from http://www.jcmr-online.com/content/15/1/63/figure/F5 and http://www.jcmr-online.com/content/15/1/63/figure/F6).

Figure 8

Simulated off-resonance response of SASHA using a 2-parameter fit, BIR4-90 saturation pulse, TR = 2.8 ms, FA = 70° at various T1’s for myocardial T2 = 45 ms. Using a 3-parameter fit has virtually no off-resonance sensitivity (< 10 ms error across ±100 Hz).

Figure 9

Influence of heart rate on estimate of myocardial T1 for various MOLLI protocols with T2 = 45 ms, and flip angle = 35°. The original MOLLI protocol (top left) had a significant sensitivity to heart rate which may be reduced by increasing the time between inversions as in 5s(3s)3s protocol (top right), or by discarding samples for longer T1 as done in a ShMOLLI conditional reconstruction, approximated by 5(0) sampling for longer T1-values (bottom left). For lower values of T1 associated with Gd contrast, it is possible to improve accuracy using a 4s(1s)3s(1s)2s sampling scheme without incurring significant heart rate dependence (bottom right).

Figure 10

Sensitivity of myocardial T1 estimate using MOLLI 5s(3s)3s to excitation flip angle for various T1 values which has increasing T1 underestimation for increasing flip angle (top graph) and in-vivo examples for native myocardium at 1.5 T showing SNR maps, T1-maps, and standard deviation (SD) maps for flip angles of 20°-35° showing trade-off of SNR and precision.

Figure 11

Imperfect inversion combined with the influence of SSFP readout alters the apparent T1* of the myocardium.

Figure 12

Phase sensitive inversion recovery (PSIR) fitting uses a 3-parameter model, whereas magnitude IR fitting using a multi-fitting approach estimates 3-parameters plus the zero-crossing. The multi-fitting magnitude IR fitting approach is prone to errors in estimating the zero-crossing in situations where the zero-crossing is close to the measured inversion times leading to a significant loss of precision for specific values of T1 and RR for a given protocol.

Figure 13

Mixing of non-inverted blood with inverted blood may alter the apparent inversion or saturation recovery.

Figure 14

In-vivo inversion recovery in blood illustrating in-flow of non-inverted blood. Fit to RV (red) and LV (blue) blood pool measurements for inversion times up to 5 seconds, and fit to LV blood (green) for measurements up to 1 second and last measurement at 10 seconds. Fits for measurements up to 5 sec (blue and red) underestimate the blood T1 due to the mixture of inverted and non-inverted spins, which flow in from outside the inversion volume.

Figure 15

Example of T1-maps in 2 subjects (a) (left) subject with heart rate of 58 bpm acquired using a MOLLI protocol with 256x144 matrix and (b) (right) subject with heart rate of approx. 90 bpm using a 192 × 120 matrix. Although the interpolated maps are of good quality, the subject with higher heart rate and thinner wall has only about 3.5 pixels across the septum leading to a degree of partial volume error in ROI measurements.

Figure 16

Magnetization transfer (MT) significantly affects inversion recovery leading to an underestimation of native myocardial T1 using the MOLLI method. Saturation recovery using higher SSFP readout flip angle causes an underestimation of SASHA using a 2-parameter fit. The 3-parameter fit SASHA is not influenced significantly by MT.

Figure 17

The effect of magnetization transfer (MT) on the inversion recovery for native myocardial tissue using MOLLI (top) and on saturation recovery using SASHA (bottom). MT changes the shape of the inversion recovery causing a shorter apparent T1*. MT has insignificant effect on the saturation recovery using SASHA with a 3-parameter fit.

Figure 18

Comparison of precision of various reported T1-mapping protocols using Monte-Carlo estimate of SD (n = 65536). The heart rate was 60 bpm, and the SNR for MOLLI methods was 25, and for SASHA was 43 to account for the increased flip angle using the saturation recovery protocol.

Figure 19

Example In-vivo T1-maps and corresponding pixel-wise SD maps acquired using MOLLI 5s(3s)3s, ShMOLLI, and SASHA protocols using 2- and 3-parameter fitting. Variation in SD across the heart is apparent due to variation in SNR from surface coil sensitivity roll-off. MOLLI has the best precision but underestimates T1 due to the approximate nature of the Look-Locker correction and due to magnetization transfer (MT). Note that SASHA with 2-parameter fitting has a small T1-underestimation; 3-parameter fitting is more accurate but has significant loss of precision.

Figure 20

Example native T1 and SD maps using MOLLI 5(3s)3 for a subject with HCM exhibiting focal native T1 abnormalities in the septal region corresponding to T1 elevation of 84 ms relative to the lateral wall representing an elevation of 2.3 SD on a pixel-wise basis (septal SD = 36 ms). (adapted from http://www.jcmr-online.com/content/15/1/56/figure/F9).