Magnetic Resonance Imaging: RF Pulse Excitation, K-Space Sampling, and Superconducting Magnets

Magnetic Resonance Imaging: RF Pulse Excitation, K-Space Sampling, and Superconducting Magnets

Magnetic Resonance Imaging (MRI) non-invasively images human soft tissue with sub-millimeter anatomical detail without ionizing radiation. Signal acquisition leverages the nuclear magnetic resonance of hydrogen nuclei ($^1H$ protons) possessing non-zero quantum magnetic dipole moments.

1. Nuclear Magnetization & The Larmor Precession

When placed inside a static external magnetic field $\mathbf{B}_0$ (typically $1.5\text{ T}$ or $3.0\text{ T}$ produced by liquid-helium-cooled niobium-titanium superconducting coils), proton magnetic moments precess around $\mathbf{B}_0$ at the Larmor frequency:

$$\omega_0 = \gamma B_0$$

Where $\gamma$ is the gyromagnetic ratio ($42.58\text{ MHz/Tesla}$ for $^1H$). In a $3\text{ T}$ scanner, $\omega_0 \approx 127.74\text{ MHz}$ (VHF radio band).

A transient $90^\circ$ Radiofrequency (RF) pulse tuned to $\omega_0$ tips net magnetization into the transverse $xy$-plane. As protons relax back to thermal equilibrium, they induce a microvolt Free Induction Decay (FID) signal inside receiver coil loops.

       Nuclear Magnetization Precession and RF Excitation
             Z (B0 Field)                         Z
             |                                    |
             |  Net Magnetization                 |
             |  / M_z                             |
             | /              RF 90 deg Pulse     |
             |/               ------------->      |
             +------------> Y                     +------------> Y (M_xy Transverse)
            /                                    /
           /                                    /
          X                                    X

2. Spatial Encoding & K-Space Matrix

To determine where signals originate anatomically in 3D space, magnetic gradient coils ($G_x, G_y, G_z$) superimpose linear magnetic field variations:

1. Slice Selection ($G_z$): Applied concurrently with the RF pulse so only a 2 mm axial slice resonates at frequency $\omega$.
2. Phase Encoding ($G_y$): Momentarily energized to induce spatial phase shifts along the vertical axis.
3. Frequency Encoding ($G_x$): Active during ADC readout, frequency-tagging horizontal pixel coordinates.

Data fills K-space ($k_x, k_y$), representing spatial frequency domains:

$$k_x(t) = \gamma \int G_x(\tau) d\tau, \quad k_y(t) = \gamma \int G_y(\tau) d\tau$$

Image reconstruction is completed by computing the 2D Inverse Fast Fourier Transform (2D-IFFT):

$$I(x, y) = \iint S(k_x, k_y) e^{j 2\pi (k_x x + k_y y)} dk_x dk_y$$

import numpy as np

def reconstruct_mri_kspace(kspace_matrix):
    '''
    Performs 2D Inverse FFT reconstructing anatomical cross-section image from raw k-space.
    '''
    # Shift low spatial frequencies from matrix corners to center
    shifted_kspace = np.fft.ifftshift(kspace_matrix)
    # Execute 2D inverse Fourier transform
    reconstructed_complex = np.fft.ifft2(shifted_kspace)
    # Shift back
    image = np.fft.fftshift(np.abs(reconstructed_complex))
    return image

3. Relaxation Contrast: T1 vs T2 Weighting

Different human tissues exhibit divergent intrinsic molecular relaxation time constants:

• T1 Relaxation (Spin-Lattice): Energy transfer back to surrounding molecular framework. Fat aligns rapidly (short T1, bright on T1-weighted images); cerebrospinal fluid aligns slowly (long T1, dark).
• T2 Relaxation (Spin-Spin): Loss of transverse phase coherence due to local magnetic field inhomogeneities. Pathological edema and tumors retain free water, showing bright signals on T2-weighted scans.

Combining accelerated compressed sensing with parallel RF receiver channels enables modern diagnostic MRI to capture crisp volumetric scans of beating hearts and vascular brain aneurysms within seconds.