MRI Signal Acquisition, k-Space Navigation & Image Artifacts

Fundamentals: MR Signal, Echoes & k-Space

• Data are collected as echoes generated by a pulse sequence (spin-echo used as canonical example; gradient-echo & other sequences introduced later).
• Each echo is sampled over time; the set of time samples populates k-space (frequency domain).
  • 2-D inverse Fourier Transform (2D-FT) → image space.
  • Spatial position is encoded in the frequency of precession imposed by magnetic field gradients.
• k-space coordinates quantify accumulated phase (dephasing):
  • Centre of k-space = fully rephased magnetisation (max signal amplitude).
  • Peripheral k-space = greater dephasing (high-spatial-frequency information).
• Practical restriction: a readout gradient lets us traverse only one k-space axis at a time (frequency-encode axis). A second axis is filled by repeating echoes with stepped phase-encode gradients.

Fourier Transform: Intuitive Audio Demo

• Lecturer visualised own voice:
  • Time-domain waveform (pressure vs. time) at top of screen.
  • FFT shows simultaneous component frequencies.
• MRI analogy: acquired FID/echo is time-domain; gradients embed many spatial frequencies; FT sorts them so each frequency → spatial location.

Sampling an Echo & Nyquist Criterion

• Echo is not recorded continuously; receiver samples at fixed interval $\Delta t$ → discrete k-space points spaced by $\Delta k$.
• If $\Delta k$ too large (undersampling) high frequencies are mis-represented (aliasing).
• Nyquist theorem: $f<em>s = \frac{1}{\Delta t} \ge 2\,f</em>{\text{max}}$.
  • In MRI, “frequency’’ ↔ spatial frequency, so criterion dictates maximum resolvable spatial frequency.
• Receiver bandwidth (RBW) = accessible frequency range set by $\Delta t$.
  • Wider RBW (smaller $\Delta t$) → higher max frequency, larger FOV but lower SNR (noise power (\propto) RBW).
  • Narrow RBW (larger $\Delta t$) → smaller FOV, higher SNR.

Field of View (FOV)

• Frequency gradient maps frequency spread to specimen width.
  • Edge pixels ↔ highest |frequency|.
• $\text{FOV}=\frac{1}{\Delta k}$ in each axis.
  • Increase RBW or decrease gradient slope to enlarge FOV; inverse to shrink FOV.

Building 2-D k-Space (Image Matrix)

• During each echo readout:
  • Frequency-encode gradient active → sample a horizontal line (kx axis).
  • Phase-encode gradient applied briefly beforehand to step vertically (ky axis).
• Repeating sequence with N phase steps yields N lines; common matrices $128\times128$, $256\times256$, $512\times512$, etc.
• Matrix size + FOV → in-plane voxel size $=\frac{\text{FOV}}{N}$.

Gradient Moments & k-Space Trajectories (Illustrative Walk-through)

• Area (moment) under a gradient pulse moves the receiver point in k-space.
  • Sign defines direction, magnitude defines distance; concurrent gradients sum vectorially → diagonal paths.
• Example trajectories dissected:
  1. Simple FID: positive read gradient only → walk from centre to +kx while signal decays.
  2. Gradient-Echo (GRE): pre-dephase lobe (−Gx) pushes to −kx edge; readout lobe (+Gx) brings trajectory back through centre (echo forms) to +kx.
  3. Phase-encode pulse alone: shifts position along +ky without readout.
  4. Combined phase (+Gy) & read (−Gx/+Gx) → diagonal move out, then horizontal read.
  5. Spin-Echo: 90°-180° pair; 180° pulse flips phase → instantaneous jump to opposite k-space side before readout.

Sampling Strategies

• Cartesian (rectilinear) sampling
  • Most common; lines acquired sequentially → regular grid → direct 2D-FT.
• Non-Cartesian examples shown:
  • Spiral: concurrent sinusoidally varying gradients spiral outward from centre.
  • Radial (projection): constant-magnitude gradient rotates to acquire spokes.
• Non-Cartesian requires re-gridding: interpolate measured points onto Cartesian grid before FT.

Signal-to-Noise Ratio (SNR) & Methods to Boost Signal

• Random noise degrades k-space and propagates into image.
• Strategies:
  • Signal averaging (NEX/NSA): repeat acquisition, add coherently.
  – $\text{SNR}\propto\sqrt{N<em>{\text{avg}}}$, scan time (\propto N{\text{avg}}) → diminishing returns.
  • Increase voxel volume: thicker slice and/or larger in-plane pixel raises total magnetisation.
  • Reduce receiver bandwidth: concentrates noise over narrower spectrum → higher SNR, but increases chemical-shift & susceptibility artefacts and lengthens echo.

Artefact 1: Aliasing / Phase-Wrap / Wraparound

• Cause: object size exceeds selected FOV in phase (or freq) axis; frequencies outside sampled range fold back inside (modulo sampling window).
• Visual: posterior head appears over anterior brain (white stripe example).
• Remedies:
  • Enlarge FOV (reduce $\Delta k$).
  • Apply spatial saturation bands to suppress signal outside region.
  • Employ oversampling or anti-alias options offered by scanner.

Artefact 2: Chemical Shift (Fat–Water Mis-registration)

• Fat and water protons have different resonance frequencies (≈ $\Delta f =220\text{ Hz}@1.5\text{ T},\;440\text{ Hz}@3\text{ T}$).
• Frequency-encode gradient converts $\Delta f$ into spatial offset:
  • Pixel shift $= \frac{\Delta f}{\text{RBW per pixel}}$.
  • Low RBW → small Hz/pixel → larger shift → bright/dark bands at fat–water interfaces (kidney rim example).
• Countermeasures:
  • Increase RBW (steepen read gradient) so $\Delta f$ < ½ pixel.
  • Use fat-suppression techniques (frequency-selective or inversion).
  • Swap frequency/phase axes so shift projects out of diagnostic plane.
• Trade-off: increasing RBW lowers SNR; mitigation must balance artefact removal vs. image noise.

Other Image-Quality Concepts Mentioned

• k-space exhibits approximate Hermitian symmetry (complex conjugate symmetry); can be exploited for acceleration (half-Fourier) but limited by noise.
• Data points are complex numbers (magnitude & phase), although visual examples use magnitude only.

Operational / Course Logistics

• Upcoming lab: three MR-scanner demo sessions (Tue-Wed-Thu).
  • Students sign up; six participants per slot; email lecturer if unavailable at specific times.
• Lab goals: witness scanner environment, manipulate FOV, RBW, matrix size, observe SNR & artefact changes in real time.

Connections & Real-World Relevance

• Fourier & sampling principles are universal across imaging, audio, telecommunications.
• Aliasing analogy: 12-hour wall clock wraps 24-hour time; MRI wraps spatial info.
• Chemical shift relates to MR spectroscopy where frequency differences intentionally reveal chemical composition.
• Bandwidth–noise trade-off parallels camera ISO vs. grain, CT tube current vs. dose, etc.

Ethical / Practical Considerations

• Scan-time inflation (e.g., high averages) impacts patient comfort, motion artefacts, throughput.
• Artefact management critical for diagnostic accuracy; mis-registration can mimic pathology.
• Hardware limits (gradient strength, slew rate, receiver sampling speed) constrain achievable RBW, FOV, and non-Cartesian trajectories.