NMR and MRI physics · Intermediate

T2* dephasing of isochromats

Written by SignalsLab

Early edition: these lessons are being reviewed.

Content

Summary

After an RF pulse tips magnetization into the transverse plane, the spins within a voxel precess at slightly different frequencies and fan out, so the net signal decays faster than true T2 would predict. This decay is characterized by T2*, which combines irreversible T2 relaxation with reversible dephasing from static field inhomogeneity.

Intuition

Picture a voxel as a crowd of small spin groups called isochromats, each a tiny vector in the transverse plane. Right after the pulse they all point the same way and add up to a large signal. But each experiences a slightly different local field because of magnet imperfections, tissue susceptibility differences, and air-tissue boundaries. So each rotates at its own rate. Fast ones race ahead, slow ones lag, and the vectors spread like runners after a starting gun. The vector sum shrinks even though each individual vector keeps its length. Part of this loss is reversible, because a 180-degree pulse can refocus static offsets (a spin echo). The part caused by random molecular interactions fluctuating in time is irreversible and is the true T2. T2* is always shorter than or equal to T2.

Key equations

1T2∗=1T2+γΔB0
(1)

Observed decay rate is the true T2 rate plus a rate from field inhomogeneity spread (ΔB0 as the field spread across the voxel).

Δωi=γΔBi
(2)

Each isochromat's frequency offset is proportional to its local field offset.

Mxy(t)=M01N∑i=1NeiΔωite−t/T2
(3)

Net transverse magnetization is the vector sum of isochromats with phases accumulating over time, times intrinsic T2 decay.

S(t)=S0e−t/T2∗
(4)

Free induction decay or gradient-echo signal envelope for a Lorentzian frequency spread.

Derivation

  1. Start with N isochromats all at phase zero immediately after a 90-degree pulse, each with magnitude M0.
  2. In the rotating frame at the Larmor frequency, isochromat i has offset Δω_i = γΔB_i and phase φ_i(t) = Δω_i t.
  3. The net signal is the average of e^{iφ_i}, multiplied by the irreversible factor e^{-t/T2}.
  4. Assume offsets follow a Lorentzian distribution with half-width Δω_h. The average of e^{iΔω t} is then e^{-Δω_h t}.
  5. So the envelope is e^{-t(1/T2 + Δω_h)}.
  6. Define 1/T2* = 1/T2 + Δω_h, with Δω_h = γΔB0 playing the role of the reversible inhomogeneity rate (often written 1/T2').

Worked example

Problem

A tissue has T2 = 80 ms. Field inhomogeneity contributes a reversible rate 1/T2' = 15 s^-1. Find T2* and the fraction of signal remaining at TE = 20 ms in a gradient echo.

  1. 1/T2 = 1/0.080 s = 12.5 s^-1.
  2. 1/T2* = 12.5 + 15 = 27.5 s^-1.
  3. T2* = 1/27.5 = 0.0364 s = 36.4 ms.
  4. Signal fraction = exp(-TE/T2*) = exp(-0.020 × 27.5) = exp(-0.55).
  5. exp(-0.55) ≈ 0.577.

Answer T2* ≈ 36 ms; about 58% of the initial signal remains at TE = 20 ms (a spin echo would retain exp(-0.25) ≈ 78%).

Common mistakes

  • Thinking T2* is a tissue constant like T2; it depends on shimming, voxel size, and field strength.
  • Believing dephasing means the individual spins lose magnetization; their magnitudes remain and only phase coherence is lost.
  • Assuming a 180-degree pulse recovers all signal; it removes only static offsets, not true T2 loss.
  • Assuming T2* can be longer than T2; it is always shorter or equal.
  • Adding times instead of rates when combining T2 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(116 lines)
import sys
import numpy as np
from manim import *
from blocks import *

# Simulation: T2* dephasing of isochromats with Gaussian frequency distribution.
# Each isochromat precesses at its own frequency.
# Net signal is the vector sum.
# T2* decay envelope is exp(-(sigma*t)^2 / 2).

N = 30  # Number of isochromats
T2_STAR = 1.5  # Characteristic decay time (arbitrary units)
TE_MAX = 6.0  # Max time to simulate
DT = 0.02  # Time step

# Generate frequencies from a Gaussian distribution centered at 0.
# For a Gaussian distribution of frequencies, the signal decays as exp(-(sigma*t)^2 / 2).
# We want this to match exp(-t/T2*), so sigma = sqrt(2)/T2*.
SIGMA = np.sqrt(2) / T2_STAR

# Fixed seed for reproducibility
rng = np.random.default_rng(42)
FREQS = rng.normal(0, SIGMA, N)

def simulate():
    """Return times, phases (steps x N), and net magnitude |sum(e^{i phi})|."""
    ts = np.arange(0, TE_MAX + DT / 2, DT)
    phases = np.zeros(N)  # Initial phases
    mags = []
    for t in ts:
        # Calculate net magnetization
        # M_xy = sum(exp(i * phi_k))
        # phi_k = omega_k * t
        phi_k = FREQS * t
        mags.append(abs(np.mean(np.exp(1j * phi_k))))
    return ts, np.array(mags)

TS, MAG = simulate()


class Main(Scene):
    def construct(self):
        TitleCard(self, "T2* Dephasing", "Gaussian distribution of isochromats")
        
        # Left: Circle with isochromat arrows
        center = LEFT * 3.5 + DOWN * 0.2
        R = 1.8
        ring = Circle(radius=R, color=C_DIM).move_to(center)
        
        # Right: Signal plot
        plot_axes = Axes(
            x_range=[0, TE_MAX, 1], 
            y_range=[0, 1.1, 0.2],
            x_length=6.0, 
            y_length=3.5,
            axis_config={"color": C_DIM, "include_tip": False}
        )
        plot_axes.move_to(RIGHT * 3.2 + DOWN * 0.2)
        xl = Text("time", font_size=24).next_to(plot_axes, DOWN, buff=0.15)
        yl = Text("|M|", font_size=24).next_to(plot_axes, UP, buff=0.1).shift(LEFT * 2)
        
        self.play(Create(ring), Create(plot_axes), FadeIn(xl), FadeIn(yl))
        
        # Sample frames for animation
        idx = list(range(0, len(TS), 4))
        # Color gradient for isochromats
        cols = color_gradient([C_BAD, C_ACCENT, C_GOOD, C_MAIN], N)
        
        def arrows(k):
            g = VGroup()
            for i in range(N):
                a = FREQS[i] * TS[k]
                v = R * np.array([np.cos(a), np.sin(a), 0])
                g.add(Arrow(center, center + v, buff=0, color=cols[i],
                            stroke_width=4, max_tip_length_to_length_ratio=0.2))
            return g
        
        cur = arrows(0)
        self.add(cur)
        
        # Plot trace
        pts = [plot_axes.c2p(TS[0], MAG[0])]
        trace = VMobject(color=C_ACCENT, stroke_width=4)
        self.add(trace)
        
        StepCaption(self, "Spins fan out due to field inhomogeneity")
        
        for k in idx[1:]:
            pts_k = [plot_axes.c2p(TS[j], MAG[j]) for j in range(0, k + 1, 1)]
            new_trace = VMobject(color=C_ACCENT, stroke_width=4).set_points_as_corners(pts_k)
            new_arrows = arrows(k)
            self.play(
                Transform(cur, new_arrows), 
                Transform(trace, new_trace),
                run_time=0.15, 
                rate_func=linear
            )
            if k % 20 == 0: # Update caption periodically
                StepCaption(self, "Net signal decays as spins dephase")
        
        # Final state
        final_mags = MAG[-1]
        StepCaption(self, f"Final signal: {final_mags:.2f} (T2* decay)")
        self.wait(1.5)


if __name__ == "__main__" and "--selftest" in sys.argv:
    # Check initial signal is 1
    assert abs(MAG[0] - 1.0) < 1e-9
    # Check signal decays over time
    assert MAG[-1] < 0.5, f"Signal should decay significantly, got {MAG[-1]}"
    # Check monotonic decrease in envelope (mostly, due to noise)
    # Just check that it's lower at the end than at 50%
    mid_idx = len(MAG) // 2
    assert MAG[mid_idx] > MAG[-1], "Signal should be higher at mid-point"
    print("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.