Gold Standard T1 Mapping

Joëlle Barral, Erik Gudmundson, Maryam Etezadi-Amoli, Nikola Stikov

Introduction

The purpose of this page is to provide some guidance for T1 mapping and a methodology to evaluate new techniques.

Almost everybody in MR does T1 mapping at one point or another. There is a huge body of literature about different T1 mapping methods. However, the published T1 values of specific tissues in healthy subjects vary well-beyond the expected inter-subject variability. We have two objectives: first, provide the tools to perform reliable T1 mapping and avoid common pitfalls; second, provide a gold standard method and a framework to evaluate new methods. This is a work in progress and your feedback will be greatly appreciated: jbarral AT mrsrl DOT stanford DOT edu.


Motivation

You may want to do T1 mapping to:



Example of 14 gold standard T1 mapping
scans of chicken muscle performed overnight.
Temperature was monitored with a Luxtron® probe.

Pitfalls and Confounds

T1 should not depend on the method, i.e.: T1 will depend on the subject, i.e.: The temperature dependency is not a concern for healthy volunteers, where it can be assumed to be steady and around 37°C. However, it is a concern if you are imaging the skin. It is also a concern if you are imaging a phantom, especially if it is a liquid one. In such a case, you need to monitor temperature. You can either monitor ambient temperature, or use a fiber optic probe, or do both. If temperature changes within a T1 mapping scan, you are trying to estimate a moving target.

Methods

Many methods exist. They often aim at gaining speed. The best method depends on your application. As a rule of thumb, the more time you have, the better the SNR and the more reliable your T1 estimation.

Gold Standard

We define the gold standard as the single slice 2D spin-echo inversion recovery (SE-IR) sequence.

A typical protocol is: FOV 10 cm, matrix size 512x128, slice thickness 2 mm, TR 2550 ms, TE 10 ms, BW 32 kHz, Frequency direction Left/Right, 4 inversion times TIs: 50, 400, 1100 and 2500 ms.


Pulse profile and zoom in the passband. A Silver-Hoult adiabatic pulse of length 8.64 ms was simulated, with a prescribed slice thickness of 2 mm.
T1 and T2 values of muscle at 1.5 T were used [1].

Fitting Procedure

Don't make assumptions on the signal equation! For example, the flip angle depends on T1, since the inversion pulse is typically 8 ms long. You can use Bloch simulations if you need to be convinced. For a SE-IR sequence with different inversion times TIn, the correct model is S(TIn) = exp(i φ) (ra + rb exp(-TIn/T1)), with φ, ra, and rb real parameters. The model can be generalized into a+b exp(-TIn/T1), with a and b complex parameters.

Our fitting procedure uses a non-linear least squares method and a grid search, since we consider that it is enough to know T1 with 1 ms precision (you can use a finer grid if you like), and we know that T1 can be found within a given range, say 1 ms to 5000 ms. The main advantage of our procedure over a standard Levenberg-Marquardt algorithm is its speed.

The chopping procedure during Prescan can give an additional 180° phase on the whole image for certain TIs. If you have to fix the phase, you can do it manually once you have all the data.


Fitting Code

The Matlab code is provided in Download. It accepts DICOM images where magnitude, phase, real part and imaginary part have been saved.
The results shown here were obtained from data acquired with the gold standard method. We scanned a saline phantom at 3 T using a wrist coil. Results obtained for brain imaging at 1.5 T and for skin imaging at 1.5 T, 3 T and 7 T can be found in [2].
Four images taken at TI = [50, 300, 1000, 2000] (magnitude is shown).
Global T1 map (a montage is displayed
if several slices were imaged).
Check of the fit at a specific voxel.


T1 histogram of the chosen ROI.


Simulations

Monte-Carlo simulations can be used:

Simulations Code

The Matlab code is provided in Download. The results shown here were obtained simulating the gold standard method.


Results obtained with 10,000 simulations and the following parameters:
Noise Standard Deviation 0.03, Flip Angle 172, TR 2550, TI = [50, 400, 1100, 2500].


T1 Values of Specific Tissues

[1, 3] are useful references.

Download

The Matlab code can be downloaded here.

References

[1] Musculoskeletal MRI at 3.0 T: Relaxation Times and Image Contrast, Gold GE et al, AJR 183:343-351 (2004)
[2] A Robust Methodology for In Vivo T1 Mapping, Barral JK, Gudmundson E et al, MRM 64:1057-1067 (2010)
[3] T1, T2 Relaxation and Magnetization Transfer in Tissue at 3T. Stanisz, GJ et al, MRM 54:507-512 (2005)
[4] "Quantitative MRI of the brain" by P. Tofts provides a good introduction to T1 mapping (cf. website and chapter online)

Joëlle Barral (jbarral AT mrsrl DOT stanford DOT edu)
Last Updated June 2, 2010