MRI Lab III: Sequences

PD Dr. rer. nat. Carl Ganter

Klinikum rechts der Isar

Technische Universität München
Institut für diagnostische und interventionelle Radiologie

cganter@[Link]
[Link]
Brief Outline
In the last lecture we considered an equidistant train of instantaneous RF pulses, separated by
periods with constant gradient moments:

We showed that the configuration model (CM) provides a decomposition of the local magnetization
density for the time points t±
j immediately before or after any RF pulse:

X
x, t± i n [ p x + ω(x) τ ] (n)
t±



m j =: e m j (1)
n

MRI Lab III: Sequences 2 / 27

Brief Outline
Simply by inserting the CM (1) into the Bloch equations, we derived recurrence relations between the
configuration vectors m(n) t±


j at subsequent time points.

For the actual calculations in this lecture, we supplement explicit expressions for the rotations matrices
R (α, φ), which can be decomposed as successive rotations around the z - and x-axis:

R (α, φ) := Rz (φ) Rx (α) Rz (− φ) (2)

with
c2α/2 s2α/2 − i sα ei φ
   

Rx (α) :=  s2α/2 c2α/2 i sα  Rz (φ) :=  e− i φ  (3)

− i sα /2 i sα /2 cα 1

We used the abbreviations cα := cos α and sα := sin α.

MRI Lab III: Sequences 3 / 27

Brief Outline
To stay within the scope of our model, we also neglected the finite duration of signal recording events
and treated them as RF pulses with zero flip angle αj = 0:

In this lecture, we will see that this representation is well suited to analyze actual MRI sequences.

In particular, we will investigate the basic properties of spin echo (SE) and gradient echo (GRE)
sequences.

MRI Lab III: Sequences 4 / 27

Spin Echo (SE) Sequences
Since their structure is already covered by our model, we start with the former:

For image generation, we need to repeat this block several times.

The repetition time TR is given by the time between subsequent excitation pulses.

The spoiler gradient “Sp” eliminates residual transverse magnetization irreversibly.

MRI Lab III: Sequences 5 / 27

Spin Echo (SE) Sequences
We wrote α and β , instead of π/2 and π , respectively, since the latter values cannot be realized,
reliably. (B1+ variations, imperfect slice profiles)

In the previous lecture, we convinced ourselves that the reconstructed voxel signal is proportional to
the zero-order transverse configuration component at the echo time t2 = 2τ = TE.

(0)
v ∝ m+ (2τ ) = SE1 = i e− i φ e− 2R2 τ sα s2β/2 · m(0) 0−


z (4)

(0)
t−


To derive Eq. (4), we exploited that the initial magnetization was in the longitudinal state mz 0 ,
such that there is only one possible pathway to the first spin echo SE1 :

i e− i φ sα (0) e− R2 τ (−1) s2β/2 (−1) e − R2 τ (0)

m(0) t− t+ t− t+





z 0 −→ m− 0 −→ m− 1 −→ m+ 1 −→ m+ (t2 ) (5)

and made use of the recurrence relations from the last lecture.

MRI Lab III: Sequences 6 / 27

Spin Echo (SE) Sequences
(0)
Note that since the reconstructed signal v was fully determined by the zero-order configuration m+ ,
it does not depend on the local resonance frequency ω (x).

Specifically, v does not depend on bulk off-resonance or susceptibility effects.

This property actually characterizes spin echoes and holds for any signal, obtained at time points t±
j
in presence of crusher gradients.

In multi-SE sequences, with more than one refocusing pulse, more than one magnetization pathway
can contribute to later echoes.

For two refocusing pulses, there are two magnetization pathways, which add up to the final signal.

It is illustrative, to depict them diagrammatically.

MRI Lab III: Sequences 7 / 27

Stimulated Echoes (STE)

Stimulated echoes like STE2 (red line) store the magnetization partly in the longitudinal component
and require at least three RF pulses.

MRI Lab III: Sequences 8 / 27

CPMG Condition
(0)
As for the first spin echo SE1 , we can calculate m+ (4τ ) = SE2 + STE2 .

We obtain the following results for the two addends:

SE2 = − i ei φ e− 4R2 τ sα s4β/2 · m(0) 0−

z (6)
− i φ − 2R2 τ − 2R1 τ −
sα s2β m(0)


STE2 = i e e e · z 0 (7)

(0)
The total signal magnitude m+ (4τ ) apparently depends on the phase offset φ.

It is maximized for ei φ = − e− i φ , i.e. a perpendicular orientation of excitation and refocusing:

φ = ± π/2

(0)
This finding, which is known as CPMG condition, also holds for later echoes m+ (2nτ ) .
(CPMG = Carr-Purcell-Meiboom-Gill)

MRI Lab III: Sequences 9 / 27

Stimulated Echoes (STE)
Stimulated echoes vanish for β = nπ , since the rotation matrices (3) do not effect transitions between
transverse and longitudinal components then.

Task: Let us gain some intuition about stimulated echoes and the relevance of the CPMG condition
with help of the simulator.

Start mri_lab, execute the script cpmg and try out different parameters:

• Deviation from perfect refocusing

• CPMG condition on/off

• Ratio T2 /T1

What do you observe?

MRI Lab III: Sequences 10 / 27

Excitation by Refocusing Pulses
Imperfect refocusing pulses with β ̸= π also partly act as excitation pulses.

These pulses will therefore generate fresh transverse magnetization from available longitudinal
(0)
magnetization mz t−


2j+1 , which is always regenerated by repolarization.

(0)
Question: Can we observe this newly generated transverse magnetization in the signals m+ (t2j )?

MRI Lab III: Sequences 11 / 27

Excitation by Refocusing Pulses
Answer: Simple diagrammatic analysis shows that the magnetization pathways, leading from
(0) (0)
mz t−


2j+1 to m+ (t2k ) with k > j must contain an odd number of periods, in which the
magnetization is in the transverse state.

The crusher gradients will therefore always dephase these pathways and any magnetization, excited
by the refocusing pulses, is not observable in the reconstructed signal.

As a second consequence, the multi-SE signal will always decay to zero for a long train of refocusing
pulses.

Single-shot multi-SE therefore generates a transient signal, for which the steady-state is zero.

For multi-shot multi-SE, a finite steady-state builds up for the initial longitudinal magnetization
(0)
mz (0− ), which mainly depends on T1 and TR.

MRI Lab III: Sequences 12 / 27

Gradient Echo Sequences (GRE)
Gradient echo (GRE) sequences differ from (multi-)SE sequences in that every time point tj is
occupied by an RF pulse and therefore not available for signal recording.

We therefore have to insert the recording events into the precession periods somehow:

Variants of steady-state free precession (SSFP) sequences differ in how this is done.

MRI Lab III: Sequences 13 / 27

SSFP Variants

MRI Lab III: Sequences 14 / 27

Transient Phase and Steady-State
Let us first consider a few commonalities of SSFP sequences, though.

Any periodic sequence of the form

mj = A mj−1 + b (8)

will generate a steady-state

lim mj = [ I − A ]− 1 b (9)
j →∞

The approach to this final state is called transient phase.

Its detailed form and duration (maximally a few T1 times) depends on the sequence.

Both, the transient phase and the steady-state can be used for image generation in MRI.

In the following, we will limit our attention to the (more common) latter case.

MRI Lab III: Sequences 15 / 27

Steady-State of SSFP
After switching to the FM frame, steady-state imaging with SSFP can be accomplished with less
restrictive conditions than strict periodicity (8).

The combination of constant flip angles αj ≡ α with a phase cycling condition

φj − φj−1 = j · ϕ + φ (10)

allows steady-state imaging in the following situations:

balanced p=0 ϕ=0

coherent p ̸= 0 ϕ = 0

RF spoiled p ̸= 0 ϕ ̸= 0

MRI Lab III: Sequences 16 / 27

 ϕ=0 ϕ ̸= 0

MRI Lab III: Sequences 17 / 27

Susceptibility Effects
As mentioned, the time points tj , where the spin echoes occur, are occupied by RF pulses.

In consequence, SSFP sequences depend on field inhomogeneity.

For example, in the FID sequence

the accelerated signal decay due to susceptibility effects has the expected form:

∗ (0)
v ∝ e− TE/T2 · m+ (11)

Analogous relations can be derived for the other SSFP variants.

MRI Lab III: Sequences 18 / 27

Sensitivity to T2∗

This example shows the signal loss in a

T2∗ -weighted GRE sequence due to field
inhomogneity, caused by an intracerebral
hemorrhage.

MRI Lab III: Sequences 19 / 27

Rapid Imaging with SSFP
Unlike (multi-)SE sequences, every RF pulse acts as an excitation pulse.

The nonzero steady-state of SSFP sequences relies on a permanent resupply with fresh,
repolarized magnetization.

SSFP sequences are therefore well suited for rapid imaging, since acquisitions with very short
repetition times TR ≪ T1 become feasible at good signal-to-noise (SNR) levels.

Since repetition times of a few milliseconds are also considerably shorter than typical transverse
relaxation times (TR ≪ T2 ), transverse coherences contribute significantly to the reconstructed
signal.

For balanced and coherent SSFP sequences, the contrast therefore depends on both relaxation
times T1,2 (sometimes their ratio T2 /T1 ) and sequence parameters, like the flip angle α.

MRI Lab III: Sequences 20 / 27

Balanced SSFP
Since balanced SSFP sequences are characterized by the absence of crusher gradients (formally by
p = 0), there is no suppression of magnetization pathways.

The reconstructed signal is therefore proportional to the sum over all configuration vectors:

(n)
X
v ∝ e− i n[ ω TR + φ ] m+ (12)
n

Apparently, the signal depends very sensitively on the local off-resonance frequency ω (voxel average).

Indeed, for ω TR + φ ≈ k 2π , the signal almost drops to zero (banding artifact).

Due to the absence of crushers, balanced SSFP sequences are essentially insensitive to motion
and are therefore even used in cardiac imaging.

MRI Lab III: Sequences 21 / 27

Coherent SSFP
Different from the balanced case, coherent SSFP sequences employ crusher gradients.

(0)
Since only the zero-order configuration m+ contributes to the reconstructed voxel signal, the
sensitivity to off-resonance is removed with similar contrast but somewhat reduced SNR.

However, the main problem with coherent SSFP is its considerable sensitivity to motion.

In fact, motion along unbalanced (crusher) gradients can significantly disturb the phase coherence
of the magnetization pathways and thereby cause a signal loss.

Formally, constant motion along crusher gradients has an analogous effect as RF spoiling (see
below).

MRI Lab III: Sequences 22 / 27

Double Echo Steady-State (DESS)

High-resolution orthopedic images are one

possible application of coherent SSFP
sequences.

Here, with a combination of FID (n = 0) and

ECHO (n = − 1).

MRI Lab III: Sequences 23 / 27

Problem: T1-Weighting
For diagnostic purposes, the ability to generate rapid T1 -weighted images, is crucial.

With respect to SSFP, the ideal approach would be to spoil any residual transverse magnetization
prior to any RF pulse.

It is easy to show that the associated signal is given by the Ernst formula:

(0) 1 − E1
m+ ∝ sα meq (13)
1 − cα E 1

with E1 := e− TR/T1 .

The optimal signal decreases with TR/T1 and is obtained at smaller flip angles.

MRI Lab III: Sequences 24 / 27

Ernst Formula

MRI Lab III: Sequences 25 / 27

RF Spoiling
Crusher gradients do not act as spoilers, though, due to the spin echoes generated at time points t±
j .

In a strict sense, perfect elimination of transverse magnetization is indeed impossible in rapid SSFP.

Nonetheless, in good approximation this can be achieved with RF spoiling.

Here, a pseudorandom variation of the RF phase can be accomplished with a nonzero value of the
phase difference increment ϕ.

(0)
In this case, m+ approaches a steady-state in the frequency modulated (FM) frame.

For properly chosen values of ϕ, the resulting signal (in presence of crusher gradients) approximates
the Ernst formula (13) sufficiently well.

MRI Lab III: Sequences 26 / 27

Summary
The two basic families of MRI sequences, (multi-)SE and SSFP can be elegantly described with
methods from EPG / CM theory.

Despite apparent structural similarities, they showed fundamentally different properties.

Spin echo sequences are insensitive to field inhomogeneity, but cannot be accelerated to the same
degree as SSFP sequences.

The latter are therefore the primary choice for applications in the important fields of rapid and
quantitative MRI.

In this lecture, we discussed fundamental properties of (multi-)SE and SSFP sequences, most of
which do not depend on real-world complications. (like adding preparation pulses)

Specifically, we were not (much) limited by details of spatial encoding.

MRI Lab III: Sequences 27 / 27