Content
Summary
MRI locates signal in space by adding small, linear magnetic field gradients to the main field , so that the Larmor frequency and phase of spins depend on position. A slice-select gradient with a shaped RF pulse excites one slab, a phase-encode gradient imprints a position-dependent phase along one in-plane axis, and a frequency-encode (readout) gradient makes frequency depend on position along the other axis during signal acquisition.
Intuition
Without gradients every proton in the magnet precesses at the same frequency, so the received signal tells us only the total amount of tissue. A gradient makes the field slightly stronger on one side of the body than the other, so spins precess at slightly different speeds depending on where they are. Think of runners on a circular track with different speeds: frequency tells you which lane a runner is in, and how far ahead of the others they have run (phase) tells you how long a gradient was on. Slice selection uses frequency to choose which lane gets excited at all. Phase encoding uses a short gradient pulse that leaves spins in a staircase of phase angles along y. Frequency encoding runs a gradient during readout so that x is coded in frequency. Repeating with many phase steps gives enough equations to separate positions; the 2D Fourier transform of the collected k-space data recovers the image.
Key equations
Larmor frequency varies linearly with position when a gradient G is applied.
Slice thickness is set by RF bandwidth and slice-gradient strength; the RF center frequency sets slice position.
Phase accumulated along y after a phase-encode gradient of duration T_pe.
Spatial frequency coordinates in k-space for readout and phase encoding.
The signal is the Fourier transform of the spin density (with k in cycles per unit length, i.e. k = γGt/2π).
Derivation
- Start with the Larmor relation ω0 = γB0, where γ/2π is about 42.58 MHz/T for hydrogen.
- Add a linear gradient along : , so .
- An RF pulse with bandwidth Δω only matches spins in a range of z: Δz = Δω/(γGz). Steeper gradient or narrower bandwidth gives a thinner slice.
- In the rotating frame, a spin at y during a gradient Gy applied for time T gains phase φ = γ Gy y T. Changing Gy amplitude each repetition gives a different phase ramp across y.
- During readout a gradient makes . The rotating-frame phase is , so the signal is a sum of frequencies, one per .
- Define kx=γGx t/2π and ky=γGy T/2π. Total signal is S(kx,ky)=∫∫ρ(x,y)exp[-i2π(kx x+ky y)]dxdy.
- Acquire one k-space line per phase-encode step, then apply the inverse 2D FFT to obtain ρ(x,y).
Worked example
A slice-select gradient of 10 mT/m is used with an RF pulse of bandwidth 1.0 kHz. (a) Find the slice thickness. (b) With a readout gradient of 5 mT/m, find the frequency difference between two points 10 cm apart in x. Use γ/2π = 42.58 MHz/T.
- Convert gradient to frequency per distance: 42.58 MHz/T × 0.010 T/m = 425.8 kHz/m = 0.4258 kHz/mm.
- Slice thickness = bandwidth / (frequency per distance) = 1.0 kHz / 0.4258 kHz/mm ≈ 2.35 mm.
- Readout: 42.58 MHz/T × 0.005 T/m = 212.9 kHz/m.
- Over 0.10 m: 212.9 kHz/m × 0.10 m = 21.29 kHz.
Answer Slice thickness ≈ 2.3 mm; frequency difference ≈ 21.3 kHz across 10 cm.
Common mistakes
- Thinking gradients change everywhere; they add a small position-dependent field with zero at isocenter.
- Confusing phase encoding with frequency encoding: phase encode is applied briefly before readout and stepped between repetitions, while frequency encode is on during acquisition.
- Believing slice thickness depends only on the RF pulse; it depends on both bandwidth and gradient strength.
- Forgetting that gradients are axes of the logical (slice, phase, frequency) frame, not fixed x, y, z; the assignment can be swapped.
- Mixing γ and γ/2π, giving a 2π error in frequencies.