K-Space

The center of k-space contains low spatial frequency information, which determines the overall image contrast and signal-to-noise ratio (SNR). Data points near the origin represent slowly varying features across the field of view.

• Dominates perceived tissue brightness
• Defines gross anatomical shapes
• Sampling errors here cause severe shading artifacts

In Cartesian acquisitions, the central phase-encoding lines are typically acquired near the echo time (TE), making them most sensitive to T1, T2, or T2* weighting.

The outer regions of k-space contain high spatial frequency information, which defines fine detail and edge sharpness. These data points represent rapidly varying intensity transitions across the image.

• Determines spatial resolution
• Defines boundaries between structures
• Truncation of outer k-space causes Gibbs ringing artifacts

Acquiring more peripheral data extends the maximum sampled frequency, enabling sharper delineation of small anatomical features at the cost of increased scan time.

For a purely real-valued object, k-space exhibits Hermitian symmetry: data at position (kx, ky) is the complex conjugate of data at (-kx, -ky). This property is exploited in partial Fourier techniques.

• S(kx, ky) = S*(-kx, -ky)
• Enables acquisition of only ~60-75% of k-space
• Remaining data synthesized via homodyne reconstruction

Partial Fourier imaging reduces scan time but introduces phase sensitivity, requiring careful phase correction to avoid artifacts in the final image.

K-space and image space form a Fourier transform pair. The 2D inverse Fourier transform converts raw k-space data into the anatomical image, while the forward transform decomposes an image into its spatial frequency components.

• Image = IFFT(K-space)
• Each k-space point contributes to every pixel in the image
• Localized artifacts in k-space produce global image effects

This global relationship means that motion during acquisition of any single k-space line corrupts the entire reconstructed image, not just a localized region.

The path traversed through k-space during acquisition is called the trajectory, and its design critically impacts image quality, speed, and artifact behavior.

• Cartesian: Rectilinear grid, robust and standard
• Radial: Lines through center, motion-robust, oversamples center
• Spiral: Single-shot efficient, sensitive to off-resonance
• PROPELLER/BLADE: Rotating blades, self-navigating motion correction

Non-Cartesian trajectories require gridding or iterative reconstruction rather than simple FFT, increasing computational complexity.

Modern acceleration techniques like SENSE and GRAPPA deliberately skip phase-encoding lines, creating an under-sampled k-space. The resulting aliasing is resolved using coil sensitivity profiles.

• Reduction factor R = number of skipped lines + 1
• SENSE: Image-domain unfolding
• GRAPPA: K-space domain interpolation of missing lines
• Enables 2-4x acceleration with minimal SNR penalty

This principle extends to Compressed Sensing, which uses incoherent undersampling and non-linear iterative reconstruction for even higher acceleration factors.