NMR and MRI physics · Advanced

Balanced SSFP: steady-state magnetisation and banding

Written by SignalsLab

Early edition: these lessons are being reviewed.

Content

Summary

Balanced SSFP (bSSFP, TrueFISP/FIESTA/bFFE) uses a fast train of RF pulses with fully refocused gradients on all axes, so transverse magnetisation is preserved from one TR to the next. The steady-state signal depends on T1/T2 and on the off-resonance phase accrued per TR, producing characteristic dark bands where that phase equals odd multiples of π.

Intuition

Imagine spins being kicked every TR by a pulse of flip angle α, alternating in sign (0, 180 phase cycling) or equivalently viewed in a frame that handles the alternation. Between pulses, spins relax partly and precess by an angle θ=2πδfTR, where δf is the off-resonance. Because gradients are balanced, no extra dephasing is added, so the magnetisation left over from one cycle becomes the starting point for the next. After a few T1/T2 times the system settles to a fixed point. Signal is high and depends on the ratio T2/T1, giving bright fluid and blood. But the fixed point depends on θ: at on-resonance the signal is at a maximum, and at θ=π (δf=1/(2TR)) the transverse components cancel destructively, giving a null: a banding artefact. Shortening TR widens the bands apart, pushing them away from the region of interest.

Key equations

θ=2πΔfTR
(1)

Off-resonance phase accrued per repetition; banding repeats every 1/TR in Hz.

Smax≈M0sin⁡α(T1/T2+1)−cos⁡α(T1/T2−1)
(2)

On-resonance bSSFP steady-state transverse signal (TR << T1,T2).

αopt=arccos⁡(T1/T2−1T1/T2+1)
(3)

Flip angle that maximises signal; gives S=M02T2T1.

Δfnull=12TR
(4)

Off-resonance frequency at the centre of the first dark band (θ=π).

Derivation

  1. Per TR, magnetisation 𝐌=(Mx,My,Mz) undergoes: RF rotation by α about x, then free precession by θ about z with transverse decay E2=exp⁡(−TR/T2), and longitudinal recovery with E1=exp⁡(−TR/T1).
  2. Write the cycle as Mn+1=AMn+B, where A=Rz(θ)⋅diag(E2,E2,E1)⋅Rx(α) and B=(0,0,M0(1−E1)) after recovery.
  3. The steady state satisfies M_ss = (I - A)^(-1) B.
  4. Solving gives a transverse signal with numerator proportional to M0(1−E1)sin⁡(α) and a denominator that is a function of E1, E2, α and cos⁡(θ).
  5. For on-resonance (θ=0) and TR≪T1,T2, this simplifies to the Smax formula shown.
  6. Maximising over alpha gives the optimal flip angle; the signal then scales with T2/T1, independent of absolute T1 and T2 values.
  7. The denominator depends on cos⁡(θ); at θ=π the transverse contributions from successive pulses subtract, creating a signal minimum, which is the band.

Worked example

Problem

A bSSFP sequence uses TR = 4 ms. Where do the dark bands fall, and what is the optimal flip angle for tissue with T1 = 1200 ms and T2 = 40 ms?

  1. Band centres are at θ=π, so δf=1/(2⋅TR)=1/(0.008 s)=125 Hz.
  2. Bands repeat every 1/TR = 250 Hz, so at +/-125 Hz, +/-375 Hz, and so on.
  3. T1/T2 = 30. cos(alpha_opt) = (30-1)/(30+1) = 29/31 = 0.935.
  4. alpha_opt = arccos(0.935) ≈ 20.6 degrees.
  5. Peak signal is 0.5⋅M0⋅1/30≈0.091M0.

Answer Dark bands at about +/-125 Hz (repeating every 250 Hz); optimal flip angle about 21 degrees for this tissue. For fluids with T2/T1 near 1 the optimum is nearer 90 degrees.

Common mistakes

  • Believing bSSFP contrast is T1-weighted; it is mainly T2/T1 weighted.
  • Thinking banding is caused by T2* dephasing within a voxel; it arises from off-resonance phase accumulation across a TR.
  • Lengthening TR to reduce banding; that brings bands closer together, so you should shorten TR.
  • Forgetting that the 0/180 phase cycling shifts the band pattern by half a period, so the signal at on-resonance is a maximum.
  • Using the T2/T1 formula when TR is not short compared with T1 and T2.

Animation

Animation coming soon

The animated version of this lesson is being prepared for the web. In the Signals AI app, lessons are drawn step by step.

Code

scene.py(101 lines)
import sys
import numpy as np
from manim import *
from blocks import *
from mri_blocks import (PulseSequenceDiagram, BlochSphere2D, SignalPlot,
                        rf_pulse, free_precession)

# ---------------- physics (seconds, Hz) ----------------
TR = 0.005          # 5 ms
TE = TR / 2         # echo at mid-TR
T1 = 0.3
T2 = 0.1
ALPHA = np.deg2rad(60)


def bssfp_trace(df, n_tr, alpha=ALPHA, tr=TR, t1=T1, t2=T2):
    """Run bSSFP with alternating (+a, -a) pulses; record M at TE of every TR."""
    M = np.array([0.0, 0.0, 1.0])
    states = []
    for k in range(n_tr):
        phase = 0.0 if k % 2 == 0 else np.pi          # alternating RF phase
        M = rf_pulse(M, alpha, phase)
        M = free_precession(M, tr / 2, t1, t2, df)     # to TE
        states.append(M.copy())
        M = free_precession(M, tr / 2, t1, t2, df)     # to end of TR
    return np.array(states)


def bssfp_profile(dfs, n_tr=600):
    """Steady-state |Mxy| at TE versus off-resonance (Hz)."""
    return np.array([np.hypot(*bssfp_trace(d, n_tr)[-1][:2]) for d in dfs])


DFS = np.linspace(-200, 200, 61)
PROFILE = bssfp_profile(DFS)
TRANSIENT = bssfp_trace(0.0, 10)


class Main(Scene):
    def construct(self):
        TitleCard(self, "Balanced SSFP", "steady state and banding")

        # --- timing diagram (ms) ---
        pd = PulseSequenceDiagram(rows=("RF", "Gx", "ADC"), t_max=10.0, width=11, height=2.5)
        pd.move_to(UP * 1.65)
        parts = VGroup(
            pd.add_rf(0.0, 0.4, kind="hard", label="+α"),
            pd.add_rf(5.0, 5.4, kind="hard", label="-α"),
        )
        for t0 in (0.0, 5.0):
            parts.add(pd.add_gradient("Gx", t0 + 0.5, t0 + 1.5, amp=-0.6),
                      pd.add_gradient("Gx", t0 + 1.5, t0 + 3.5, amp=0.6),
                      pd.add_gradient("Gx", t0 + 3.5, t0 + 4.5, amp=-0.6),
                      pd.add_adc(t0 + 1.7, t0 + 3.3))
        parts.add(pd.add_marker(2.5, "TE"), pd.add_marker(5.0, "TR"))
        self.play(Create(pd), run_time=1.0)
        self.play(FadeIn(parts), run_time=1.0)
        StepCaption(self, "Every gradient sums to zero within each TR")

        # --- Bloch sphere: on-resonance approach to steady state ---
        sphere = BlochSphere2D(radius=1.0).move_to([-4.2, -1.55, 0])
        arrow = sphere.vector([0, 0, 1])
        cursor = pd.time_cursor()
        counter = Text("TR 0", font_size=24, color=C_ACCENT).next_to(sphere, RIGHT, buff=0.3)
        self.play(FadeIn(sphere), FadeIn(arrow), FadeIn(counter))
        self.add(cursor)
        StepCaption(self, "Alternating ±α: M settles at each echo (TE)")
        for k, M in enumerate(TRANSIENT):
            start = 0.0 if k % 2 == 0 else 5.0
            pd.tracker.set_value(start)
            new_c = Text(f"TR {k + 1}", font_size=24, color=C_ACCENT).move_to(counter)
            self.play(pd.cursor_to(start + 2.5), Transform(arrow, sphere.vector(M)),
                      Transform(counter, new_c), run_time=0.7)

        # --- steady-state profile vs off-resonance ---
        sp = SignalPlot(x_range=(-200, 200, 100), y_range=(0, 0.5, 0.25),
                        x_label="Δf (Hz)", y_label="|Mxy|", width=5.2, height=2.1)
        sp.move_to([2.6, -1.5, 0])
        self.play(FadeIn(sp), run_time=0.6)
        StepCaption(self, "Steady-state signal versus off-resonance")
        self.play(Create(sp.plot(DFS, PROFILE)), run_time=2.0)
        i_null = [int(np.argmin(np.where(DFS < 0, PROFILE, 9))),
                  int(np.argmin(np.where(DFS > 0, PROFILE, 9)))]
        dots = VGroup(*[sp.dot_at(DFS[i], PROFILE[i], color=C_BAD) for i in i_null])
        self.play(FadeIn(dots))
        StepCaption(self, "Bands where off-resonance phase per TR is ±π: Δf = ±1/(2TR)")
        self.wait(1.5)


if __name__ == "__main__" and "--selftest" in sys.argv:
    E1, E2 = np.exp(-TR / T1), np.exp(-TR / T2)
    a = ALPHA
    analytic = (1 - E1) * np.sin(a) / (1 - (E1 - E2) * np.cos(a) - E1 * E2) * np.sqrt(E2)
    on_res = PROFILE[np.argmin(np.abs(DFS))]
    assert abs(on_res - analytic) / analytic < 0.02, (on_res, analytic)
    pos = DFS > 0
    df_null = DFS[pos][np.argmin(PROFILE[pos])]
    assert abs(df_null - 1 / (2 * TR)) <= 7, df_null
    assert PROFILE.min() < 0.3 * PROFILE.max()
    assert np.allclose(PROFILE, PROFILE[::-1], atol=1e-3)  # symmetric profile
    print("selftest ok")

The Manim scene code that draws this lesson's animation in the Signals AI app (Python, Manim Community).

ccBYNC© 2026 SignalsLab. Lesson text shared as is under CC BY-NC 4.0.