3.8 k-Space and Data Acquisition

3.8 k-Space and Data Acquisition

At the heart of MRI image reconstruction is k-space, a spatial frequency domain where raw digitized signals are stored during data acquisition. Understanding how k-space is structured, how it is filled by the gradients, and how sampling limits and acceleration techniques affect the final image is a core requirement for the ARRT MRI registry.

What is k-Space?

k-space is a temporary digital storage matrix of spatial frequencies. It is not an actual image, but rather a grid of data points. Each point in k-space contains phase and frequency information from the entire imaging slice.

• Units of k-space: Spatial frequency is measured in radians per meter (rad/m) or cycles per millimeter (cycles/mm).
• Image Reconstruction: The raw data stored in k-space is converted into a recognizable anatomical image using a mathematical process called the Two-Dimensional Fast Fourier Transform (2D FFT).

k-space exhibits a symmetrical property known as conjugate symmetry. The data points in the top half of k-space are mirror images of the bottom half, and the left half is a mirror image of the right half.

Spatial Frequencies: Contrast vs. Resolution

The location of data within the k-space matrix determines its contribution to the final image:

• The Center of k-space: Contains low spatial frequencies. This area holds the information regarding the signal-to-noise ratio (SNR), general anatomy shapes, and tissue contrast (e.g., T1, T2, or proton density weighting). If the center of k-space is omitted or damaged, the image will lose contrast and appear as a low-contrast outline.
• The Periphery of k-space: Contains high spatial frequencies. This region stores the data that determines spatial resolution, fine details, sharp boundaries, and edge definition. If the outer edges of k-space are omitted, the image will appear blurry, though its overall contrast will remain intact.

Frequency and Phase Axes in k-Space

The axes of k-space correspond to the spatial encoding gradients:

• Horizontal Axis (kx): Governed by the frequency encoding gradient. During the readout window, the analog-to-digital converter (ADC) samples the signal at high speed, filling a horizontal row in k-space from left to right.
• Vertical Axis (ky): Governed by the phase encoding gradient. Each phase encoding step determines which horizontal line of k-space will be filled. If a protocol has a phase matrix of 256, then 256 separate phase encoding steps must be performed, filling 256 lines from top to bottom.

The Nyquist Theorem and Aliasing

The Nyquist Theorem dictates that to accurately digitize an analog signal without distortion, the sampling frequency (fs) must be at least twice the highest frequency component (f_max) present in the signal:

fs ≥ 2 * f_max

If this condition is not met, a phenomenon called aliasing (wrap-around) occurs.

• Frequency Axis Aliasing: Prevented by utilizing bandpass filters (analog or digital filters) to remove frequencies outside the desired Field of View (FOV), and by oversampling the signal along the frequency encoding axis.
• Phase Axis Aliasing: Occurs when the FOV is smaller than the physical dimensions of the object being imaged. Because phase encoding cannot be filtered in the same way as frequency, spins outside the FOV acquire phase shifts that match spins inside the FOV, causing the anatomy outside the FOV to fold over and superimpose on the opposite side of the image. This is corrected by increasing the FOV, using sat bands, or employing no-phase-wrap (phase oversampling), which increases the phase FOV and discards the extra data.

k-Space Filling Strategies

Gradients steer the acquisition through k-space. Different trajectories can be used to optimize scan speed or contrast:

• Linear (Sequential) filling: k-space is filled line-by-line from top to bottom. This is the default trajectory for standard spin echo and gradient echo sequences.
• Centric filling: The center lines of k-space are filled first, and the acquisition then progresses outward toward the periphery. This is crucial for contrast-enhanced MRA because the center of k-space (which determines contrast) is filled during the peak concentration of the contrast agent.
• Partial Fourier (Half-Fourier / Fractional NEX): Exploiting conjugate symmetry, the scanner only acquires a portion of k-space (typically 50% to 60%) along the phase axis or frequency axis. The remaining lines are mathematically synthesized. This reduces scan time by up to 40% at the cost of a square root of 2 reduction in SNR.
• Radial and Spiral filling: Non-Cartesian trajectories.
  • Radial filling: Lines of k-space are acquired as spokes of a wheel passing through the center. Because the center is sampled repeatedly with every line, this method is highly resistant to patient motion, making it useful for pediatric or uncooperative patients.

Parallel Imaging Acceleration

Parallel Imaging is an acquisition acceleration technique that reduces scan times by under-sampling k-space. It requires a multi-channel receiver coil array (phased-array coil).

Mechanism

Instead of acquiring every phase encoding line in k-space, the scanner skips lines (e.g., acquiring only every second or third line). This reduces the scan time by a factor equal to the Acceleration Factor (R). Under-sampling k-space normally causes severe aliasing. However, because each coil element in the phased-array has a unique spatial sensitivity profile, the scanner uses these known sensitivity maps to mathematically separate and unwrap the aliased pixels.

Common Parallel Imaging Algorithms

• SENSE (Sensitivity Encoding): An image-domain reconstruction method. The aliased image is reconstructed first, and the coil sensitivity profiles are used to unwrap the image pixel-by-pixel.
• GRAPPA (GeneRalized Autocalibrating Partially Parallel Acquisitions): A k-space domain method. The missing k-space lines are calculated from the acquired lines using a calibration fit before the Fourier transform is applied.

Tradeoffs

• Scan Time Reduction: Reduces scan time directly by the factor of R.
• SNR Penalty: SNR decreases due to the reduction in data points collected. The new SNR is calculated as:

SNR_parallel = SNR_base / (g * √R)

Where g is the geometry factor (a measure of the coil elements' spatial arrangement and overlap) and R is the acceleration factor.

• Artifacts: If the coil sensitivity maps are inaccurate (e.g., due to patient motion), parallel imaging artifacts (central band-like noise or residual aliasing) will occur.