MRI Physics, Part II: Interactive Companion Experimental

Teaching page. A companion to Biglands 2012, read alongside the paper. The drawings are schematic models, not patient images; values marked illustrative are model values, not reference values.

Faster imaging

Fewer phase-encoding lines: reduced matrix, rectangular field of view, half-Fourier

Every technique in the paper's Table 1 shortens the scan the same way: it acquires fewer phase-encoding steps, so fewer lines of k-space. The long-established ones reduce the acquisition matrix in the phase-encoding direction (fewer, outer lines dropped: coarser pixels in that direction), reduce the field of view in that direction (a rectangular field of view: anything outside it wraps round), or acquire only just over half of k-space and fill the rest from its symmetry (half-Fourier).

Newer techniques leave out further lines where the information is redundant: parallel imaging (next section) and, for cine and dynamic imaging, methods that use information repeated over time.

Read in the paper: Biglands 2012, section “More acceleration techniques”, Table 1.

k-space (log scale), acquired lines
Reconstructed image

The phantom is part I's drawn 64 × 64 short-axis slice, phase encoding top to bottom; it is transformed in your browser. The phantom, its 64 lines and the half-Fourier extra lines are illustrative.

Acceleration techniques and some of their names: Biglands 2012, Table 1
TechniqueNames in the table
Reduced matrix in the phase-encoding directionScan %, Phase resolution, Zero Padding
Reduced field of view in the phase-encoding directionRectangular FOV (RFOV), Phase FOV, Asymmetric FOV
Half-Fourier acquisitionHalf Fourier, Partial Fourier, Half Scan, 0.5 NEX
Parallel imaging, image-based reconstructionSENSE, mSENSE (self-calibrating), ASSET, SPEEDER
Parallel imaging, k-space-based reconstructionSMASH, GRAPPA (self-calibrating), ARC (self-calibrating)
Temporal under-samplingView-sharing, Keyhole, UNFOLD, TSENSE, Kt-BLAST

Parallel imaging: folding over and unfolding with coil sensitivities

Parallel imaging needs a receiver coil array. It acquires the phase-encoding lines spaced further apart, skipping lines by the reduction factor R, over the same extent of k-space, so the resolution is kept. On its own that would alias: signal wraps from one side of the image to the other. With at least two coil elements spread along the phase-encoding direction, each element sees the folded copies at different strengths; knowing each element's sensitivity map (made from the central k-space lines, in a reference scan or within the acquisition) the reconstruction puts each signal back in its place. SENSE and ASSET do this in image space; SMASH, GRAPPA and ARC in k-space.

The cost: SNR falls with the square root of R, about 30% for R = 2, so it suits images with SNR to spare. If the field of view in the phase-encoding direction is too small, fold-over remains at the centre of the image that the reconstruction cannot remove. And R must be applied along a direction with a favourable spread of coil elements, or the reconstruction fails, with high noise at the centre of the image.

Read in the paper: Biglands 2012, section “More acceleration techniques”, Figure 1.

Reduction factor R
Anterior element
Posterior element
Unfolded

The two element images are what each element records (folded when R > 1); the third is the reconstruction with noise added in proportion to its noise amplification. Illustrative model: two elements, anterior and posterior; sensitivity falls as 1 / (1 + (distance / 0.45 × image width)²); pixel 5 mm.

A section-by-section companion to Biglands 2012, part II of the physics review: each section condensed, with a model you can move. The drawings are schematic simulations, not patient images.

Method and sources

What this page is

A teaching companion to Biglands 2012, in the paper's order. Each section is condensed and paraphrased in our own words, and its mechanism is redrawn as an interactive model. Nothing here is a clinical threshold. The constants and ranges the page prints as the paper's (the chemical shift, relaxivities, blood T1 and T2 estimates, the bottle concentration and cardiac output, TS, doses, the EGE and LGE timings and TIs, Tables 1–4) are from it; the gyromagnetic ratio is from part I (Ridgway 2010). Every other parameter is an illustrative model value, labelled where it is used.

Models

Where the paper disagrees with itself

Its text says non-linearity yields underestimates of MBF, its Figure 16 legend an overestimate (this page follows the figure, which the model reproduces). The test-bolus volume is 2–3 ml in the text and 1–2 ml in Figure 29 (both given).

Left out

The paper's conclusion, acknowledgements and reference list; its patient images (none copied); the full mathematics of quantitative perfusion models, which the paper does not give.

Verification

Experimental. The pure functions (fold-over and SENSE unfolding, half-Fourier symmetry, null points, chemical shift, SPAMM rotations, relaxivity, the CE-MRA dilution arithmetic, perfusion slice timing, truncation, IR signals against Figures 22 and 24, navigator efficiency, bipolar phase and VENC aliasing) are unit-tested against hand-checked values; every sentence the page writes about a simulation is computed from the model at the current setting. The models are not verified against scanner data.

Source and licence

  1. Biglands JD, Radjenovic A, Ridgway JP. Cardiovascular magnetic resonance physics for clinicians: part II. J Cardiovasc Magn Reson 2012;14:66. doi:10.1186/1532-429X-14-66
  2. Ridgway JP. Cardiovascular magnetic resonance physics for clinicians: part I. J Cardiovasc Magn Reson 2010;12:71. doi:10.1186/1532-429X-12-71 (the gyromagnetic ratio and the phantom).

The paper is open access under the Creative Commons Attribution License 2.0 (CC BY 2.0), © 2012 Biglands et al. This page adapts it: the text is condensed and paraphrased, Tables 1–4 are condensed, and the figures are redrawn as interactive models; no figure is copied. The adaptation is ours, not the authors'.