MRI Parameters — Overview and Classification

MRIninja Knowledge Base | Master / Reference Page Version 1.0 — May 2026

MRIninja Knowledge Base | Master / Reference Page Version 1.0 — May 2026

1. Introduction and Purpose of This Page

Every MRI image is the product of a precisely defined set of numerical values — the acquisition parameters — that govern how the scanner interacts with tissue, how long each acquisition takes, and what the resulting image looks like. Changing a single parameter rarely changes only one thing: in MRI, nearly every parameter is interdependent with at least two others, and optimising a protocol requires understanding these relationships simultaneously rather than one variable at a time.

This page provides the classification framework, the physical logic, and the practical decision rules for MRI acquisition parameters across the MRIninja knowledge base. It does not reproduce the scanner-manual-level parameter lists that are vendor-specific and protocol-specific; instead, it provides the conceptual architecture that allows a radiologist or MRI technologist to understand why a parameter value appears in a protocol, what happens when it is changed, and which trade-offs are acceptable in clinical practice.

Each parameter group described here has, or will have, a dedicated child page on MRIninja covering its specific optimisation strategies, field-strength dependencies, and artefact implications. This master page provides the taxonomy and the inter-parameter relationships.

The companion page MRI Sequences — Overview and Classification covers the sequence families that these parameters control. The two pages together constitute the technical foundation of the MRIninja protocol knowledge base.

2. The Fundamental Framework: What Parameters Control

MRI acquisition parameters control four dimensions simultaneously:

1. Contrast — the relative signal intensity difference between tissues; what the image is designed to show. Parameters controlling contrast: TR, TE, TI, flip angle, b-value, diffusion direction, inversion time, echo train length (ETL), and fat suppression method.

2. Signal-to-noise ratio (SNR) — the ratio of signal from tissue to random noise; the fundamental currency of image quality. Parameters controlling SNR: voxel volume (FOV, matrix, slice thickness), number of signal averages (NSA/NEX), bandwidth, field strength, and coil configuration.

3. Acquisition time — how long the sequence takes. Parameters controlling time: matrix size, phase-encoding steps, TR, ETL, parallel imaging acceleration (R), partial Fourier factor, number of averages, and 3D vs 2D acquisition.

4. Artefact profile — what image degradation the sequence is vulnerable to. Parameters controlling artefact vulnerability: bandwidth (chemical shift, geometric distortion), TR (flow artefacts), phase-encoding direction (motion, aliasing), echo time (susceptibility), ETL (T2 blurring), and spatial resolution (partial volume).

The fundamental tension in MRI parameter design is that improving one of these dimensions almost always degrades another. Understanding which trade-offs are clinically acceptable for a given diagnostic question is the practical skill that separates optimised protocols from default scanner presets.

3. Classification of MRI Parameters

MRI parameters are most usefully classified into five groups based on their primary function in the acquisition:

4. Group 1 — Timing Parameters

4.1 TR (Repetition Time)

TR is the interval between successive RF excitation pulses for the same slice or slab. It is the primary determinant of T1 contrast in spin echo and gradient echo sequences.

Physical logic: after an RF excitation, longitudinal magnetisation (Mz) recovers toward M₀ at a rate determined by T1. If TR is short relative to T1, tissues with long T1 have not recovered much Mz before the next excitation — they produce less signal. Tissues with short T1 recover more fully — they produce more signal. This T1-dependent signal difference is T1 contrast.

Short TR (< T1 of relevant tissue): maximises T1 contrast; tissues with long T1 appear dark relative to tissues with short T1. Used for T1-weighted imaging.

Long TR (>> T1 of relevant tissue): all tissues approach full Mz recovery before each excitation; T1 differences are minimised; contrast is driven by other parameters (TE, flip angle). Used for T2-weighted, PD-weighted imaging.

Practical interdependences:

• TR × number of slices = maximum per-slice acquisition time available. Increasing TR allows more slices for the same acquisition time.
• Short TR required for T1 contrast → fewer slices per acquisition → may require separate stack for full coverage.
• SAR increases proportionally with flip angle squared and inversely with TR. Very short TR with large flip angles triggers SAR limits, particularly at 3T.
• In inversion recovery (STIR, FLAIR): TR must be sufficiently long for the tissue with the longest T1 to adequately recover before the next inversion pulse. Insufficient TR shifts the effective null point (see STIR and FLAIR sequence pages).

4.2 TE (Echo Time)

TE is the interval between the RF excitation pulse and the centre of the signal readout (the echo). It is the primary determinant of T2 or T2* contrast in SE and GRE sequences respectively.

Physical logic: after excitation, transverse magnetisation (Mxy) decays. In spin echo sequences, decay follows T2 (true transverse relaxation); in gradient echo sequences, decay follows T2 (T2 plus field inhomogeneity contributions). At longer TE, tissues with short T2/T2 have decayed more and produce less signal relative to tissues with long T2/T2*.

Short TE (< T2 of relevant tissue): minimises T2 differences; contrast driven by T1 (if TR is also short) or proton density (if TR is long). Used for T1-weighted, PD-weighted, and anatomical imaging.

Long TE (comparable to T2 of relevant tissue): maximises T2 differences. Used for T2-weighted imaging; optimal TE ≈ T2 of the tissue of interest for maximum CNR.

TE and susceptibility: in gradient echo sequences, TE controls the degree of T2* weighting and therefore the sensitivity to susceptibility effects (iron, haemosiderin, calcification, deoxyhaemoglobin). Longer TE at a given field strength increases susceptibility contrast (and artefact). This is exploited in SWI (long TE) and avoided in DCE/VIBE (minimum TE).

TE and chemical shift (Dixon): the fat-water phase relationship depends on TE. At specific TE values, fat and water protons are in-phase (IP) or opposed-phase (OP), enabling Dixon fat-water separation. IP and OP TE values are field-strength-dependent (1.5T: IP at 4.6 ms multiples; OP at 2.3 ms multiples; 3T: IP at 2.4 ms multiples; OP at 1.2 ms multiples).

Effective TE (TE_eff) in TSE: in turbo spin echo sequences, the central k-space lines — which determine the primary image contrast — are filled at a specific echo in the train. This echo's timing is the effective TE, regardless of when the peripheral k-space lines are acquired. Selecting the TE_eff determines whether the TSE image is T1-weighted (short TE_eff), PD-weighted (medium TE_eff), or T2-weighted (long TE_eff).

4.3 TI (Inversion Time)

TI is the interval between the 180° inversion pulse and the readout excitation in inversion recovery (IR) sequences. It determines which tissue's longitudinal magnetisation passes through zero at the time of the readout — nulling that tissue's signal.

Null point: Mz(TI) = 0 when TI = T1 × ln(2) ≈ 0.693 × T1 of the target tissue.

Field-strength dependence: T1 values are longer at 3T than at 1.5T; therefore, TI values must be recalibrated when changing field strength. The most common clinical error in IR sequence migration is using 1.5T TI values at 3T, which prevents appropriate tissue nulling.

FLAIR at 3T shows the TI paradox: despite CSF T1 being longer at 3T (4000 ms vs 3600 ms at 1.5T), the optimal TI is shorter. This results from the finite TR correction: the effective null TI = −T1_CSF × ln[(1−e^(−TR/T1))/(1+e^(−TR/T1))], which produces shorter values when T1 is long relative to TR. See the FLAIR sequence page for the complete derivation.

4.4 Echo Spacing (ES) and Echo Train Length (ETL) Interaction

In TSE/FSE, the effective TE is a product of the echo spacing (the time between consecutive echoes in the train) and the position of the k₀ echo within the train. Short echo spacing reduces T2 blurring (see Section 5 on ETL). ES is primarily controlled by the readout bandwidth — wider bandwidth → shorter readout → shorter ES.

5. Group 2 — Spatial Parameters

5.1 Field of View (FOV)

Dedicated parameter page: FOV — Field of View provides the complete optimisation reference for FOV, voxel size, aliasing, SNR/CNR and reduced-FOV EPI behaviour.

FOV defines the physical dimensions of the image in the phase and frequency encoding directions. It determines, together with the matrix, the voxel size in those dimensions.

FOV_phase × FOV_frequency × Slice thickness = voxel volume

Reducing FOV reduces voxel volume (higher spatial resolution) but also reduces SNR (SNR ∝ voxel volume × √acquisition_time). The critical constraint: if the FOV in the phase direction is smaller than the anatomy in that direction, aliasing (wraparound or fold-over artefact) occurs — anatomy outside the FOV folds back onto the image. This is prevented by phase oversampling (zero-filling beyond the FOV in k-space) at the cost of increased acquisition time, or by phase encoding direction selection (choosing the axis where less anatomy exceeds the FOV).

Rectangular FOV (rFOV): using a smaller FOV in the phase direction than the frequency direction, tailored to the anatomy shape (e.g., sagittal knee: AP-shorter than SI). Reduces phase-encoding steps proportionally → reduces acquisition time for the same spatial resolution. Standard for all extremity and spine protocols.

Dedicated parameter page: Phase Oversampling — Fold-over Suppression / No Phase Wrap provides the complete optimisation reference for aliasing prevention, rFOV interaction, No Phase Wrap/fold-over suppression, sequence-specific time costs and 3D slab oversampling logic.

Dedicated parameter page: REST Slab / Presaturation Band provides the complete optimisation reference for spatial saturation pulses, flow artefact suppression, fold-over control, spectroscopy outer-volume suppression and SAR-aware REST planning.

5.2 Matrix

Dedicated parameter page: Acquisition Matrix provides the complete optimisation reference for matrix, k-space sampling, spatial resolution, SNR/CNR, acquisition time and sequence-specific matrix behaviour.

Matrix defines the number of data points collected in the frequency (Nx) and phase (Ny) encoding directions. Together with FOV, it determines in-plane spatial resolution:

In-plane resolution = FOV / Matrix

Frequency matrix (Nx): collected continuously during the readout gradient; increasing Nx directly increases acquisition bandwidth (not time, unless bandwidth is held constant). In practice, Nx can be increased at no time cost; the trade-off is SNR reduction from the wider bandwidth required.

Phase matrix (Ny): each line of k-space requires a separate TR interval; increasing Ny directly increases acquisition time proportionally. Ny is the most directly time-expensive parameter to increase.

Asymmetric matrix: using a higher frequency resolution than phase resolution — very common in clinical MRI. A 512 × 256 matrix acquires 512 frequency points and 256 phase steps, giving anisotropic voxels that are twice as long in the phase direction. Saves acquisition time (256 vs 512 TR intervals) at the cost of anisotropic resolution.

Minimum resolutions for diagnostic structures (from MRIninja sequence and protocol pages):

5.3 Slice Thickness

Dedicated parameter page: Slice Thickness provides the complete optimisation reference for slice selection, through-plane resolution, partial volume, cross-talk, SNR/CNR and 2D versus 3D slice behaviour.

Dedicated parameter page: 2D vs 3D Acquisition provides the complete optimisation reference for the acquisition-mode choice between slice-selective 2D imaging and Fourier-encoded 3D volume imaging, including SNR efficiency, MPR capability, motion sensitivity and sequence-specific behaviour.

Slice thickness determines the through-plane spatial resolution and contributes the third dimension of voxel volume. Thinner slices provide:

• Higher through-plane spatial resolution
• Less partial volume averaging (more accurate signal for small structures)
• Lower SNR (less tissue signal per voxel)
• More slices required for a given coverage → longer scan time or more acquisitions

Minimum useful slice thickness is determined by: SNR at the available field strength; structure size (slice must be smaller than the structure to avoid critical partial volume); coil sensitivity profile; acquisition time constraints.

Slice gap vs contiguous slices: a positive gap between slices reduces cross-talk (RF from an adjacent slice exciting the target slice before full relaxation), but creates coverage gaps where pathology can be missed. Zero-gap or overlapping slices are preferred for structures where continuity matters; a small positive gap (10–20% of slice thickness) is standard for multi-slice T2 acquisitions in the brain to prevent cross-talk.

3D isotropic: a single voxel size in all three dimensions, enabling post-acquisition MPR in any plane without resolution loss. Required for: brain volumetry (MPRAGE/BRAVO); joint cartilage (3D SPACE/CUBE); brachial plexus (3D CISS); MRCP (3D SPACE T2); breast MRI (3D VIBE). The time cost of 3D acquisition is offset by the phase-encoding SNR advantage (SNR ∝ √N_slices for 3D vs individual 2D slices).

5.4 SNR, Voxel Volume, and the Signal Equation

The fundamental SNR relationship for MRI:

SNR ∝ Voxel_volume × √(NSA) × √(BW_correction) × B₀ × Coil_sensitivity

Where voxel_volume = FOV_x × FOV_y / (Nx × Ny) × Slice_thickness.

This relationship makes explicit the trade-off between spatial resolution (small voxel = low SNR) and signal quality. When the spatial resolution requirement is fixed by the diagnostic target, SNR can be recovered only by: increasing NSA (at linear time cost), increasing field strength (B₀), improving coil efficiency, or reducing bandwidth (at the cost of chemical shift artefact and longer ES).

6. Group 3 — k-Space and Acquisition Parameters

6.1 Echo Train Length (ETL) / Turbo Factor

ETL is the number of echoes acquired per TR interval in a TSE/FSE sequence. Each echo fills one k-space line. Therefore, ETL directly divides the acquisition time compared with a conventional SE: a TSE with ETL=16 acquires a matrix of 256 phase steps in 256/16 = 16 TRs instead of 256 TRs — a 16× time reduction.

T2 blurring: echoes later in the train have been T2-weighted longer. Their signal is lower. Since they fill the higher spatial-frequency k-space lines (outer k-space = fine detail), the contribution of short-T2 tissue to fine detail is reduced — producing a T2-dependent blurring effect on structures with short T2 (cartilage, tendons, ligaments, cortex). This is the primary image quality cost of long ETL.

ETL selection rules:

• Short ETL (4–8): for high-resolution fine-structure sequences (cartilage, plantar plate, Lisfranc ligament); acceptable T2 blurring
• Medium ETL (12–20): standard T2/PD brain and MSK imaging; acceptable for most clinical structures
• Long ETL (30–100): STIR whole body, rapid surveys; significant blurring — only acceptable when fine structure is not the diagnostic target
• VFA (Variable Flip Angle) ETL (50–300): 3D SPACE/CUBE/VISTA for volumetric T2 — ETL extended with flip angle modulation to maintain signal throughout the echo train

6.2 Partial Fourier (Partial k-Space)

Partial Fourier exploits the conjugate symmetry of k-space: the upper and lower halves of k-space are complex conjugates of each other (for a perfectly symmetric object). Acquiring only 5/8, 6/8, or 7/8 of the phase-encoding steps in k-space and estimating the unacquired portion reduces scan time proportionally.

Penalty: phase errors (from B0 inhomogeneity, motion, flow) break the conjugate symmetry assumption → Gibbs ringing, ghosting, and resolution loss in the phase direction. The shorter the Fourier fraction (5/8 < 6/8 < 7/8), the more the conjugate symmetry assumption is exploited and the more vulnerable the reconstruction to phase errors.

Clinical application: partial Fourier 6/8 is routinely used in 3D acquisitions (MRCP, brain MPRAGE, body VIBE) to reduce scan time without significant resolution loss. Partial Fourier 5/8 is used in EPI-DWI to shorten the echo train and reduce geometric distortion, at the cost of increased Gibbs ringing.

6.3 Parallel Imaging

Parallel imaging uses the spatial encoding information from multiple receiver coil elements to partially replace phase-encoding gradient steps, reducing the number of TR intervals needed and directly reducing acquisition time by the acceleration factor R.

Mechanism: in GRAPPA (Siemens) and SENSE (Philips/GE), every Rth phase-encoding line is acquired (1/R of the full k-space). The missing lines are reconstructed from the coil sensitivity profiles of the array coil elements. The SNR penalty from parallel imaging is √R from the reduced data, multiplied by the g-factor (a geometry-dependent noise amplification factor from the coil array configuration).

SNR_parallel = SNR_full / (g × √R)

For R=2 and typical g-factor 1.1–1.3, SNR is reduced to approximately 60–70% of the full-sampled acquisition. For R=4, SNR is approximately 25–35% of full-sampled — often too low for diagnostic quality without compensatory measures.

Practical limits: R=2 is universally used in clinical MRI with minimal quality compromise. R=3 is used in 3D acquisitions where the full-sampled SNR provides headroom. R=4+ requires high-field scanners (3T), dense coil arrays (≥ 32 channels), and careful application.

Simultaneous multi-slice (SMS/multiband): applies parallel imaging in the slice direction rather than the phase-encoding direction. Multiple slices are acquired simultaneously by applying simultaneous RF pulses at multiple frequencies, then un-aliased by the coil array. Used for: brain fMRI (factor 2–6), DWI (factor 2–4), whole-body DWI. Reduces scan time without the through-plane SNR penalty that standard parallel imaging imposes in the phase direction.

Compressed sensing (CS): exploits the sparsity of MRI data in an appropriate transform domain (wavelet, etc.) to reconstruct undersampled k-space acquisitions. Achieves higher acceleration factors (4–8×) than parallel imaging alone. Requires iterative reconstruction algorithms, longer reconstruction time, and images that may have a different texture from fully sampled acquisitions. Validated for several clinical applications (3D breast MRI, DCE liver, brain MPRAGE).

6.4 2D vs 3D Acquisition

2D acquisition: each slice is acquired independently. Phase encoding in 2D occurs in the in-plane direction (Ny steps per slice). The z-dimension (slice) is selected by the RF pulse bandwidth and gradient strength. Cross-talk between adjacent slices (if gap = 0) requires slice interleaving or TR-based spacing. SNR per voxel ∝ √1 (no slice-encoding benefit).

3D acquisition: a slab is excited and phase-encoded in both in-plane and through-plane directions. The z-dimension is encoded by a second phase-encoding gradient. SNR advantage: √N_z (where N_z is the number of through-plane encoding steps) compared with 2D — this is the primary SNR advantage of 3D. All voxels in the 3D acquisition benefit equally from this SNR, enabling thinner slices at equivalent SNR compared with 2D.

3D advantages: isotropic resolution; no inter-slice gaps; superior SNR for thin slices; no cross-talk; full MPR capability; compatible with high parallel imaging factors (2D × 1D acceleration available).

3D disadvantages: longer acquisition time (all z-steps required per TR × Ny × Nz); more vulnerable to motion during the long acquisition window; higher SAR for sequences requiring many RF pulses (TSE at 3T).

When 2D is preferred: when motion is severe (free-breathing body acquisitions); when per-slice motion robustness matters (HASTE/SSFSE); when the acquisition must be very fast per slice (dynamic cardiac); when the FOV depth (z-coverage) is small and SNR is not limiting.

7. Group 4 — RF and Contrast Preparation Parameters

7.1 Flip Angle (α)

The flip angle is the degree by which the 180° inversion pulse (in SE-based sequences) or the excitation pulse (in GRE sequences) tips the longitudinal magnetisation toward the transverse plane. In spin echo sequences, the excitation pulse is always 90° and the refocusing pulse is always 180°. In gradient echo sequences, the flip angle is a free parameter ranging from 5° to 90°.

Ernst angle optimisation: for a spoiled GRE at a given TR and T1, the flip angle that maximises SNR is the Ernst angle:

α_Ernst = arccos(e^(−TR/T1))

For TR = 5 ms and T1 = 900 ms (brain white matter at 1.5T): α_Ernst ≈ 12°. Operating at the Ernst angle maximises SNR efficiency. Operating above the Ernst angle increases T1 contrast (at SNR cost); operating below it increases signal but reduces T1 differentiation.

Flip angle and SAR: SAR scales as α². Large flip angles at 3T produce substantial RF heating, particularly for TSE sequences with 180° refocusing pulses. VFA echo trains (SPACE/CUBE) use variable flip angles along the echo train to reduce SAR while maintaining signal throughout long echo trains.

7.2 Receiver Bandwidth (BW)

Receiver bandwidth defines the range of frequencies sampled during the readout gradient. It is expressed in Hz/pixel (Hz per pixel in the frequency-encoding direction).

Wide bandwidth: shorter readout duration → shorter minimum TE → less T2* dephasing during readout → less susceptibility artefact near metal and air-tissue interfaces → shorter echo spacing (ES) in TSE → less T2 blurring. SNR penalty: SNR ∝ 1/√BW — doubling the bandwidth reduces SNR by 30%.

Narrow bandwidth: longer readout → larger chemical shift artefact (more Hz displacement of fat relative to water → larger spatial shift). Chemical shift displacement = (Δν × Δx) / BW, where Δν = fat-water frequency difference (220 Hz at 1.5T, 440 Hz at 3T). At 100 Hz/px bandwidth, the fat-water shift is 2.2 pixels at 1.5T and 4.4 pixels at 3T — diagnostically significant near fat-water interfaces. Recommendation: use ≥ 200 Hz/px at 3T for T1 sequences to keep chemical shift below 1 voxel.

Bandwidth and SNR trade-off in context: the bandwidth is often misunderstood as a simple SNR knob. In practice, the bandwidth must be set high enough to: (a) prevent diagnostic chemical shift artefacts; (b) achieve the required TE (for T2*/susceptibility control); (c) maintain the desired echo spacing in TSE (for T2 blurring control). Within these constraints, bandwidth can be minimised to maximise SNR.

7.3 Fat Suppression Methods

Fat produces T1-bright and T2-bright signal that competes with pathological signal in many clinical contexts. The choice of fat suppression technique is one of the most clinically important parameter decisions in protocol design because different techniques have very different B0 dependency profiles.

CHESS / Chemical Saturation (ChemSat): a spectrally selective RF pulse saturates the fat resonance before the imaging excitation. Requires accurate knowledge of the local fat resonance frequency → fails when B0 is inhomogeneous. Effective near isocentre (brain, knee at isocentre), unreliable off-isocentre (extremities > 20 cm from isocentre, cervicothoracic junction, parotid with metallic fillings).

SPAIR (Spectral Adiabatic Inversion Recovery): uses an adiabatic inversion pulse to invert fat magnetisation, then waits for a short TI for fat to pass through zero, then applies the imaging sequence. More B0-robust than CHESS because the adiabatic inversion is partially B1-insensitive. Standard spectral fat suppression for body and MSK MRI at 3T. Fails in regions of severe B0 inhomogeneity (near metallic implants, cervicothoracic junction, distal extremities).

STIR (Short TI Inversion Recovery): T1-based fat null — independent of B0 field homogeneity. Reliable fat suppression at any position within the scanner. Mandatory for: off-isocentre extremity MRI (wrist, fingers, toes, ankles); whole-body DWI; brachial plexus; cervicothoracic junction; parotid with dental metalwork. Contraindicated post-gadolinium (Gd shortens T1, shifting the null point). SNR lower than SPAIR for equivalent scan time.

Dixon (2- or 3-point): acquires images at two or more specific TE values where fat and water are in-phase and out-of-phase, then computationally separates fat and water. B0-independent (uses B0 map correction). Provides simultaneously: water-only, fat-only, in-phase, and out-of-phase images. Fully compatible with post-gadolinium imaging. Preferred technique for body T1 sequences at 3T (VIBE-Dixon, LAVA-Flex, mDixon). Requires TE set at specific values; slightly longer minimum TE than single-echo alternatives.

The STIR post-gadolinium contraindication is absolute across all MRIninja protocols: documented in the STIR, FLAIR, WB-MRI, and all relevant protocol pages. This rule is enforced regardless of clinical context.

7.4 Magnetisation Transfer Contrast (MTC)

Magnetisation transfer (MT) pre-pulses apply an off-resonance RF pulse that saturates the bound proton pool (macromolecular protons in myelin, proteins) and transfers this saturation to the free water pool, reducing water signal in tissue. Structures with high bound-to-free proton exchange (myelinated white matter, muscle, liver parenchyma) show greater MT suppression than structures with low exchange (free fluid, cartilage).

Clinical applications of MTC: (a) MR angiography — MT suppresses background tissue signal while flowing blood (no exchange) remains bright, improving vessel-to-background CNR; (b) brain MS plaques — MT ratio (MTR) quantifies demyelination severity; (c) cartilage biochemistry — MT-sensitised sequences for glycosaminoglycan content assessment.

SAR cost: MT pre-pulses add significant SAR, particularly at 3T. May limit the number of slices achievable in a given TR.

8. Group 5 — Derived and Functional Parameters

8.1 b-Value (Diffusion Weighting)

The b-value quantifies the degree of diffusion weighting applied in a DWI sequence. It is a function of the diffusion gradient amplitude (G), duration (δ), and separation (Δ):

b = γ² G² δ² (Δ − δ/3)

where γ is the gyromagnetic ratio. The SI (signal intensity) in a DWI sequence is:

SI(b) = SI(0) × e^(−b × ADC)

where ADC is the apparent diffusion coefficient. The signal decreases exponentially with b-value; the rate of decrease is proportional to the ADC of the tissue.

b = 0: no diffusion weighting; T2-weighted image. Free water (CSF, necrosis) appears bright; restricted water (dense cells) also appears bright but for a different reason. The b=0 alone cannot discriminate T2-bright from restricted diffusion without a high-b comparison.

Low b (50–200): primarily suppresses vascular signal (blood moving in capillaries has high "pseudo-diffusion" ADC from incoherent microcirculation — IVIM component). Used as the low-b reference for IVIM-DWI and as the substitute for b=0 in WB-DWI (DWIBS) to reduce vascular background.

Diagnostic b (800–1000): the standard clinical diagnostic b-value. Provides sufficient diffusion weighting to suppress normal free water while retaining signal in restricted-diffusion tissue (acute infarcts, dense tumours, abscesses, malignant nodes). SNR is adequate at 1.5T and 3T with standard coil configurations.

High b (1500–2000): increases sensitivity for restriction at the cost of SNR. Useful for prostate (b=1500–2000 for PI-RADS DWI), paediatric brain tumours, and CJD cortical ribbon sign detection. Often calculated as a mathematical extrapolation from lower b-values rather than directly acquired, to maintain SNR.

ADC calculation: requires a minimum of two b-values. ADC = −ln(SI_b2 / SI_b1) / (b2 − b1). Multiple b-values improve ADC accuracy by fitting an exponential decay curve.

Clinical ADC reference ranges (approximate; b=0 and b=1000 s/mm²):

8.2 Diffusion Directions (DTI)

Standard DWI acquires diffusion in three orthogonal directions and averages (trace DWI) to produce the isotropic ADC map. Diffusion Tensor Imaging (DTI) acquires diffusion in ≥ 6 non-collinear directions to characterise the full diffusion tensor — providing:

• FA (Fractional Anisotropy): the degree of directionality of diffusion (0 = fully isotropic; 1 = fully anisotropic). High FA in myelinated white matter tracts; reduced FA in demyelination, axonal injury, or oedema
• Tractography: 3D reconstruction of white matter pathways from the tensor field
• Mean diffusivity (MD): equivalent to the ADC but calculated from the full tensor

Minimum directions for clinical DTI: 6 (adequate for clinical FA maps); 12–30 for improved precision; 60–100+ for research-grade tractography.

8.3 Perfusion Parameters

Perfusion MRI quantifies the delivery of blood to tissue and is used in neurological, oncological, and cardiac applications.

DSC (Dynamic Susceptibility Contrast): T2-weighted EPI series acquired during gadolinium bolus passage. The T2 signal drop is related to cerebral blood volume (CBV), cerebral blood flow (CBF), and mean transit time (MTT). Key parameter: the temporal resolution of each 3D volume acquisition (typically 1.5–2.5 seconds per volume); TR and TE of the EPI readout.

DCE (Dynamic Contrast Enhanced): T1-weighted 3D GRE series acquired during and after gadolinium bolus. The T1 signal increase is related to tissue perfusion parameters (Ktrans, kep, ve) via pharmacokinetic modelling. Key parameter: temporal resolution (< 10–60 seconds per volume depending on the model); flip angle (usually 10–15° for saturation-recovery-like T1 sensitivity).

ASL (Arterial Spin Labelling): non-contrast perfusion using magnetically labelled blood water as the endogenous tracer. Key parameters: labelling duration (PCASL: 1.5–2 seconds), post-labelling delay (PLD: 1.5–2.5 seconds for adult brain), background suppression. No gadolinium required; lower SNR than DSC/DCE; quantitative CBF in mL/100g/min.

9. Parameter Interdependence — The Trade-Off Matrix

The following matrix summarises the most clinically important trade-offs between parameter changes. Each cell shows the direction of effect on the output when the row parameter is increased.

11. Vendor Naming Conventions for Parameters

The core parameters (TR, TE, TI, FOV, matrix, slice thickness, bandwidth, flip angle) are named consistently across vendors. Advanced parameters and acceleration techniques carry vendor-specific names:

12. The Parameter Optimisation Workflow

When a protocol is designed or revised, parameters should be adjusted in the following logical sequence:

Step 1 — Define the diagnostic target: what structure? what resolution is needed? what contrast? This defines the non-negotiable constraints: minimum in-plane resolution, mandatory fat suppression type, required contrast weighting.

Step 2 — Select the sequence family: SE/TSE/GRE/EPI/IR based on the required contrast mechanism and speed requirement (see companion page 9003).

Step 3 — Set the contrast parameters first (TR, TE, TI, flip angle, b-value): choose values that provide the required tissue contrast for the diagnostic target. This is the most important step and the most commonly wrong step in default protocols.

Step 4 — Set the spatial parameters (FOV, matrix, slice thickness): choose the minimum voxel size that the diagnostic target requires, consistent with acceptable SNR at the available field strength.

Step 5 — Set the acceleration parameters (ETL, parallel imaging, partial Fourier): choose the maximum acceleration consistent with acceptable T2 blurring (ETL), g-factor SNR loss (parallel imaging), and reconstruction artefacts (partial Fourier) — constrained to meet the required acquisition time.

Step 6 — Verify the trade-offs: check that the parameter combination simultaneously achieves adequate SNR, acceptable artefact profile, required contrast, and target acquisition time. Use the trade-off matrix in Section 9 for guidance.

Step 7 — Test and validate: acquire the sequence on a volunteer or phantom and verify that the expected contrast, resolution, and artefact profile are achieved. Adjust bandwidth first (artefact control), then NSA (SNR), then matrix (resolution vs time).

13. Evidence Gaps and Ongoing Debate

Optimal b-value combinations for specific pathologies: the b-values recommended for clinical DWI (liver, prostate, brain, lymph node) vary between guidelines and between vendors. No universally accepted single standard exists. The PI-RADS v2.1 [1] and ESUR prostate MRI guidelines [2] represent the most standardised recommendations currently available for one specific anatomical region.

Deep learning reconstruction and parameter recalibration: DLR algorithms trained on specific acquisition parameters may produce non-standard noise texture and contrast when applied to acquisitions with different parameters. The effect of DLR on apparent ADC values, SNR quantification, and volumetric measurements has not been systematically characterised across all clinical applications. Parameter optimisation protocols developed without DLR may require revalidation when DLR is enabled.

Compressed sensing clinical validation: CS-accelerated acquisitions have been validated for specific applications (brain MPRAGE, DCE breast, liver VIBE) but the validation methodology differs substantially across studies. The specific CS reconstruction algorithm, regularisation parameters, and acceleration factor interact to produce image quality that may not be predictable from acceleration factor alone.

1.5T vs 3T parameter equivalence: translating protocols between 1.5T and 3T is not simply a matter of adjusting TI and TE. The full set of parameter changes required for equivalent diagnostic quality — including TR, bandwidth, flip angle, ETL, and fat suppression technique — has not been formally published as a comprehensive equivalence table for all protocol types.

SAR management strategies at 3T: the SAR limits at 3T constrain which protocols are feasible at short TR and high flip angle. Scanner-specific SAR management solutions (TR extension, flip angle reduction, RF hypersensitivity mode) vary by vendor and do not have evidence-based comparison data.

14. Group 6 — Physiological Synchronisation Parameters

Physiological motion — from respiration, cardiac pulsation, and swallowing — is one of the primary sources of image degradation in body, cardiovascular, and head-and-neck MRI. The parameters controlling physiological synchronisation determine whether motion is avoided, gated, triggered, or averaged out, and represent a distinct group of acquisition decisions that interact with all timing and spatial parameters described in the preceding sections.

14.1 Respiratory Triggering and Gating

The motion problem: the liver, kidneys, spleen, adrenal glands, and surrounding structures move 10–30 mm in the superior-inferior direction with normal quiet breathing. At the spatial resolution required for MRI of these structures (0.5–1.5 mm in-plane), even 2–3 mm of inter-TR motion produces ghosting artefacts (in the phase-encoding direction), blurring, and misregistration between pre-contrast and post-contrast acquisitions.

Four strategies address respiratory motion in body MRI, each with distinct parameter implications:

Breath-hold acquisition: the patient holds respiration during a single acquisition. Used for: DCE liver (arterial phase 15–25 s); 2D T1 body acquisitions (10–20 s per stack); MRCP 2D thick-slab (3–5 s per projection). The breath-hold duration determines the maximum number of phase-encoding steps and therefore the achievable spatial resolution within the breath-hold window. The equation governing breath-hold-limited matrix: N_phase_max = T_breathhold / TR. At TR = 5 ms: N_phase = 20 s / 5 ms = 4000 phase steps — well exceeding typical matrices; at TR = 6000 ms (STIR): N_phase = 20 s / 6000 ms = 3 — entirely impractical. Breath-holding is only feasible for sequences with TR ≪ 1 second.

Free-breathing averaging (NSA-based): random respiratory motion is partially cancelled by signal averaging over multiple acquisitions. Effective when: many averages are used (NSA ≥ 4–8); the respiratory motion is truly random and not periodic; the anatomical target does not require high in-plane resolution. Used for: WB-DWI/DWIBS (free-breathing, NSA = 4–8); whole-body STIR; paediatric body MRI. Motion is not eliminated — it is reduced. Residual ghosting is visible at NSA=4; substantially less at NSA=8.

Respiratory triggering (prospective): a respiratory motion monitor (bellows, navigator, optical tracking) detects the respiratory cycle and triggers the next TR at a defined phase of the cycle — typically at end-expiration, when the diaphragm is in its lowest, most reproducible position. Each TR starts at the same respiratory phase → data acquired from the same spatial position → no motion blurring. Parameters:

• Trigger window (acceptance window): the duration around the trigger point during which acquisition is allowed. Narrow window (5–10 ms): highest position reproducibility, lowest efficiency (many TRs rejected). Wide window (20–50 ms): higher efficiency but more positional variation within the window.
• Trigger phase: acquisition at end-expiration is standard (diaphragm lowest). Some sequences use mid-expiration for longer acquisition windows.
• Navigator efficiency: for navigator-triggered acquisitions, the fraction of accepted TRs. Target efficiency ≥ 50%. Below 30%, effective scan time more than doubles; consider breath-hold instead.
• Effective TR: because many respiratory cycles are rejected, the effective TR between consecutive accepted TRs equals the respiratory cycle duration (typically 4–6 seconds) regardless of the nominal TR set on the scanner. This effective TR determines T1 contrast — it is always long enough to approximate full T1 recovery for most body tissues.

Respiratory navigator (pencil-beam navigator): a 1D column of signal is continuously sampled through the liver dome. The position of the diaphragm-liver interface is tracked in real time. When the interface is within the acceptance window (typically ±3–5 mm of end-expiration), the subsequent TR is accepted. Used for: 3D MRCP (standard implementation); 3D liver SPACE/CUBE; DWI with navigator correction. Navigator placement is the technologist's most critical positioning task for navigator-triggered acquisitions: the pencil beam must intersect the liver-lung interface (bright-dark boundary) at a perpendicular angle; placement on lung tissue alone or within hepatic vessels produces navigator failure.

Prospective motion correction (continuous navigator): in 3D volumetric acquisitions (brain, cardiac), continuous navigator information can be used not just to gate data but to correct the k-space trajectory or the slice prescription in real time for measured motion. Vendor implementations: Siemens PROMO (brain); GE PROBE; Philips Navigator with real-time correction. Reduces motion sensitivity without the SNR penalty of triggering, but requires additional scan time for navigator acquisitions.

14.2 Cardiac Triggering and Gating

The cardiac motion problem: the heart moves 10–20 mm during the cardiac cycle. Cardiac structures — myocardium, valves, coronary arteries — move at velocities up to 10 cm/s during systole. Additionally, cardiac pulsation causes aortic and CSF pulsation artefacts visible in brain, spine, and body MRI even when cardiac pathology is not the imaging target.

Electrocardiographic (ECG) triggering: the ECG R-wave is the physiological reference event. Each TR or sequence acquisition is synchronised to the R-wave.

• Prospective triggering: the acquisition starts at a defined delay after the R-wave (trigger delay, TD). The acquisition window must end before the next R-wave. Used for: cardiac T1 mapping (MOLLI — multiple TIs sampled across multiple cardiac cycles); cardiac cine with segmented k-space filling; coronary MRA.
• Retrospective gating: data is acquired continuously; the ECG signal is recorded simultaneously. During reconstruction, each k-space line is assigned to a cardiac phase based on its timing relative to the preceding R-wave. All cardiac phases are reconstructed post-acquisition. Used for: cardiac function cine (bSSFP); flow-velocity mapping (PC-MRA); 4D flow MRI.

Key cardiac triggering parameters:

Peripheral pulse triggering (PPG/PPU): for patients where ECG is technically inadequate or unsafe, peripheral pulse (finger or toe plethysmograph) provides a surrogate trigger. PPG delay relative to the R-wave is approximately 200–400 ms depending on patient circulation. This delays the effective trigger and shifts the cardiac phase actually acquired. Most modern scanners automatically account for the PPG delay; the technologist must verify the stated cardiac phase corresponds to the clinical target.

Cardiac gating in non-cardiac MRI: pulsatile flow artefacts from the carotid arteries and aorta propagate into adjacent structures. In brain and spine MRI, CSF pulsation in the phase-encoding direction produces ghost artefacts. Cardiac gating of brain and spine T2 TSE sequences reduces these artefacts, at the cost of longer, variable scan times. Most modern departments use saturation bands and flow compensation (GMN) rather than cardiac gating for non-cardiac examinations, accepting residual pulsation artefacts rather than the time and complexity of cardiac gating.

MOLLI (Modified Look-Locker Inversion Recovery) for myocardial T1 mapping: a series of IR-GRE acquisitions at different effective TI values, acquired across multiple cardiac cycles, fits a T1 recovery curve for each pixel of the myocardium. Parameters: number of heartbeats per IR preparation; number of images per preparation; trigger delay; flip angle for the GRE readout. See the IR sequence page and cardiac MRI protocol pages for the full parameter specification.

15. Group 7 — Breath-Hold Timing Management

15.1 Breath-Hold Duration and Sequence Design

Breath-hold MRI requires the technologist to manage the timing of instruction delivery, patient compliance assessment, and sequence triggering in real-time. The practical parameters controlling breath-hold acquisition quality are as operationally important as the technical acquisition parameters.

Maximum breath-hold duration: motivated, trained patients can sustain a breath-hold for 18–25 seconds. Elderly, obese, or dyspnoeic patients may sustain only 8–12 seconds. Paediatric patients (< 6 years) typically cannot cooperate with breath-hold commands. The protocol must be designed with the weakest patient in mind: if the clinical indication requires contrast enhancement and the patient cannot hold for 18 seconds, the temporal resolution of the arterial phase must be compressed to fit within 12–15 seconds.

Hyperventilation (pre-oxygenation): instructing the patient to take 2–3 deep breaths before the breath-hold significantly extends sustainable apnea duration, typically by 5–8 seconds. This is standard practice for the arterial phase of liver DCE acquisitions, where the temporal window is narrow and temporal resolution cannot be compromised. Not appropriate for all patients (contraindicated in severe COPD, recent cardiac event).

Breath-hold instruction standardisation: the phase of the respiratory cycle at which the breath-hold is performed affects the position of abdominal organs. End-expiration breath-hold provides the most reproducible organ position across multiple acquisitions and across sessions (critical for serial comparison). Inspiration breath-hold allows a slightly larger diaphragm excursion and longer comfortable duration but produces more variable organ position. The standard for body MRI is end-expiration breath-hold for all non-cardiac acquisitions.

Multiple breath-hold acquisitions: when a single breath-hold cannot cover the required volume at the required resolution, multiple overlapping breath-hold acquisitions are used. This strategy requires:

• Consistent breath-hold position between acquisitions (end-expiration standard)
• Short recovery period between breath-holds (minimum 30–60 seconds for adequate recovery)
• Geometric matching of overlapping acquisitions (same FOV, same slice orientation)
• Verification that organ position is reproducible before continuing to the next critical sequence (e.g., before injecting contrast)

Automated breath-hold detection: some scanner platforms (GE SilentScan features, Siemens navigator-based detection) can monitor respiratory position in real-time and delay sequence start until the navigator confirms end-expiration position, then automatically trigger the acquisition. This removes operator-dependent timing variability and is particularly useful for arterial phase dynamic acquisitions where timing is critical.

15.2 Dynamic Contrast Timing and Breath-Hold Sequencing

The most demanding real-time management task in body MRI is the coordination of contrast injection timing with breath-hold commands and sequence triggering for dynamic liver acquisitions.

Standard DCE liver timing protocol:

Bolus tracking (fluoroscopic triggering): a rapid low-resolution 2D GRE is acquired repeatedly over the aorta during contrast injection. When the aortic signal exceeds a threshold (> 100% increase from baseline), the scanner automatically or semi-automatically triggers the arterial phase acquisition. This adjusts for patient-specific circulation times (range approximately 12–30 seconds in clinical populations) and is more reliable than a fixed injection-to-scan delay. Parameters: monitoring location (aorta at the level of the celiac axis); trigger threshold (institution-dependent; typically 100–150% signal increase); preparation delay before acquisition start (scanner-specific; typically 4–8 seconds).

SmartPrep / CARE Bolus / BOLUS TRAK: vendor-specific implementations of fluoroscopic bolus tracking. All use the same principle; the threshold and trigger delay are adjustable.

Test bolus: a small (2 mL) gadolinium injection followed by 20 mL saline, with a rapid T1-GRE series over the aorta, measures the individual circulation time for the specific patient. The time-to-peak is used to set the scan delay for the diagnostic full-dose injection. Rarely used in routine practice but essential for coronary MRA and when bolus tracking is not available.

16. Group 8 — Real-Time Monitoring and Sequence Execution Timing

16.1 Real-Time Display and Decision Points

Modern MRI scanners provide real-time image display during acquisition. The technologist must perform active quality monitoring during the examination — not only after completion — and make in-sequence decisions that affect image quality.

Navigator efficiency monitoring: during a navigator-triggered 3D acquisition (MRCP, liver SPACE), the navigator efficiency is displayed in real-time. If efficiency drops below 40–50% after 2–3 minutes of acquisition (typically due to drifting respiratory baseline or irregular breathing), the technologist should pause the acquisition and reassess: (a) re-position the navigator; (b) instruct the patient to regulate breathing; (c) widen the acceptance window (at the cost of positional accuracy); (d) switch to a breath-hold strategy. Waiting for the full acquisition to complete with 20% efficiency produces an acquisition time 5× the expected duration and often degraded image quality.

Fat suppression quality check: the first acquired slice of a fat-suppressed sequence (STIR, SPAIR, Dixon) should be immediately reviewed on the scanner console. If fat suppression has failed (visible bright fat signal where suppression is expected), the acquisition should be paused, the shimming re-optimised, and the fat suppression method reconsidered (STIR vs Dixon vs SPAIR) before continuing. Allowing a full 20-minute acquisition to complete with failed fat suppression is one of the most common avoidable errors in MRI.

DWI geometry check: immediately after the first DWI slab is acquired, verify that geometric distortion at the tissue-air interfaces (particularly the posterior lung-liver interface and the sacrum-bowel interface in pelvic DWI) is within acceptable diagnostic limits. If distortion is excessive, reduce the echo train length (use higher parallel imaging R to reduce EPI readout length), or switch to a readout-segmented EPI technique. Confirm that the DWI ADC map is computing correctly and that the signal at b=0 and b=high are appropriately distinct before completing the acquisition.

SWI phase and magnitude check: for SWI ARIA monitoring in patients on anti-amyloid therapy, verify that the phase and magnitude images are both generated and available for review before ending the examination. Missing phase images cannot be retrospectively reconstructed from the magnitude alone.

Motion artefact severity assessment: after each sequence, assess the phase-encoding direction for ghosting artefacts. A rapid visual check of the first and last slices of each acquisition identifies whether motion worsened during the sequence. If the last slices of a long TSE acquisition show significantly more ghosting than the first (patient fatigue-related motion increase), the sequence may need to be repeated with patient re-instruction or the protocol shortened.

16.2 Real-Time Adaptive Protocol Management

The modern MRI technologist is not a passive button-presser executing a fixed protocol — they are an active diagnostic partner making real-time decisions that affect the clinical outcome of the examination.

Protocol modification triggers that the technologist should recognise and act upon:

17. Group 9 — Image Reconstruction, Interpolation, and Deep Learning

This section substantially extends the brief reconstruction discussion in Section 6. Image reconstruction encompasses all post-acquisition mathematical operations that transform the raw k-space data into the diagnostic image presented to the radiologist.

17.1 Fourier Transform and Raw Data Reconstruction

The fundamental reconstruction step in all standard MRI is the 2D or 3D Fourier transform. The k-space data (complex-valued, Nx × Ny × Nz array) is transformed into image space by the inverse Fourier transform. The properties of the Fourier transform directly determine several image characteristics:

k-space density and image contrast: the central k-space region (low spatial frequencies, near k=0) determines the overall image contrast and bulk signal levels. The peripheral k-space region (high spatial frequencies, far from k=0) determines the fine structural detail — edges, small lesions, fine anatomical lines. This duality is exploited in partial Fourier (see Section 6.2) and in centric k-space ordering (k₀ filling strategy).

k₀ ordering strategy: in TSE sequences, the echo in the train assigned to k₀ (the central k-space line) determines the effective TE and therefore the primary image contrast. Early k₀ assignment (k₀ = first echo) → T1-weighted; late k₀ (k₀ = last echo) → T2-weighted. In SWI, the k₀ is typically assigned to the echo that provides the target T2* contrast.

Zero-filling (interpolation in k-space): raw k-space data can be zero-padded (additional zero-valued k-space lines added beyond the acquired matrix) before Fourier transformation. This is equivalent to sinc interpolation in image space — it increases the apparent image matrix (and therefore displayed resolution) without adding new physical information. Commonly used: 512 × 512 zero-filled display from a 256 × 256 acquired matrix. The apparent improvement in edge sharpness is real (the sinc interpolation eliminates the Gibbs ringing pattern at the Nyquist frequency) but does not improve the actual spatial resolution or SNR. Technologists should be aware that the "matrix" shown in the scanner display or DICOM header may reflect the interpolated matrix rather than the acquired matrix — important for reporting true spatial resolution.

Homodyne reconstruction (partial Fourier): when only a fraction of k-space is acquired (e.g., 5/8 partial Fourier), the reconstruction algorithm estimates the missing k-space data using the conjugate symmetry relationship and combines it with the acquired data using a weighting filter. This produces an image with nominally full matrix at the cost of mild Gibbs ringing from the imperfect conjugate symmetry estimation. The homodyne algorithm is the standard partial Fourier reconstruction on all major vendor platforms.

17.2 Parallel Imaging Reconstruction

Parallel imaging reconstruction (Section 6.3) operates on undersampled k-space data. Two fundamentally different architectures:

k-space-based (GRAPPA): missing k-space lines are estimated from the acquired lines using the calibration data (auto-calibration signal, ACS) acquired at the centre of k-space. The GRAPPA kernel is calibrated from the ACS. The quality of the kernel calibration determines the degree of reconstruction artefact (coherent residual aliasing or noise amplification). More ACS lines → better kernel → lower g-factor but more ACS acquisition time. Typical ACS: 24–32 central k-space lines.

Image-space-based (SENSE): the aliased (undersampled) image is unaliased in image space using the measured coil sensitivity maps. Requires a separate coil sensitivity calibration scan (body coil reference or pre-scan). SENSE has a well-defined g-factor that quantifies local SNR loss from the reconstruction geometry.

Reconstruction artefacts specific to parallel imaging:

• g-factor noise amplification: spatially heterogeneous SNR reduction; worst in regions of poor coil geometry (centre of large bodies with high R)
• Residual aliasing (coherent ghost): when the GRAPPA kernel calibration is poor (few ACS lines, subject motion during calibration) or when SENSE coil maps are inaccurate (patient moving between map acquisition and imaging)
• Artificial sharpening: some implementations apply post-reconstruction filtering that increases apparent edge sharpness, simulating higher resolution; this may affect lesion characterisation

17.3 Compressed Sensing Reconstruction

Compressed sensing (CS) exploits the compressibility of MRI data: most MRI images contain significant redundancy (smooth regions, sparse structures), and can be accurately reconstructed from fewer measurements than required by the Nyquist criterion if appropriate mathematical constraints are applied.

CS reconstruction algorithm: an iterative optimisation that minimises:

• Data consistency: the reconstructed image, when re-sampled, should match the acquired k-space data
• Sparsity penalty: the image, when transformed into a sparse representation (wavelet transform, finite differences), should have as few non-zero coefficients as possible

This dual constraint is solved iteratively, typically requiring 10–50 iterations per image. Reconstruction time is substantially longer than standard FFT reconstruction (seconds to minutes per volume vs milliseconds).

Parameters that control CS image quality:

• Acceleration factor (R_CS): total k-space undersampling. Achievable R: 3–8× in clinical practice (vs 2–4× for GRAPPA/SENSE)
• Random vs pseudo-random undersampling pattern: CS requires an incoherent (non-periodic) sampling pattern to avoid aliasing artefacts. Variable-density random undersampling (denser at k=0, sparser at periphery) is standard
• Regularisation strength (λ): the relative weighting of the sparsity penalty vs data fidelity. Too high λ: over-smoothed, blurry images that may miss small lesions. Too low λ: inadequate regularisation, residual aliasing. The optimal λ is application-specific and is typically fixed by the vendor for each approved CS application
• Number of CS iterations: more iterations → better convergence → longer reconstruction time → higher image fidelity

Clinical applications with vendor-approved CS implementations:

Critical CS caution: CS texture artefacts produce characteristic image appearance (cartoon-like smoothness in homogeneous regions, edge ringing) that differs from noise-limited or parallel imaging images. Radiologists and technologists must be aware that:

• Lesion characterisation may differ between CS and fully-sampled acquisitions at equivalent acquisition time
• Quantitative measurements (ADC, T1 maps, T2 maps) may be systematically affected by the regularisation
• No large prospective validation study has confirmed diagnostic equivalence for CS vs fully-sampled for all clinical applications; vendor approvals are based on limited validation datasets

17.4 Deep Learning Reconstruction (DLR)

Deep learning reconstruction represents the most significant change in MRI image formation since parallel imaging. DLR applies neural networks trained on large datasets to either: (a) denoise noisy images acquired at standard acquisition parameters; or (b) reconstruct undersampled k-space data, replacing or supplementing standard iterative algorithms.

Denoising DLR (Siemens Deep Resolve, GE AIR Recon DL, Philips SmartSpeed, Canon AiCE): a convolutional neural network (CNN) is trained on paired noisy–clean image sets. The network learns to remove noise while preserving structural detail. Applied at the image reconstruction stage after standard Fourier transform reconstruction.

Parametric DLR (k-space-based): the network operates on k-space data directly or on the unaliased parallel imaging output, learning to reduce reconstruction artefacts and noise simultaneously.

Effect on acquisition parameters: because DLR substantially improves SNR and sharpness, it allows acquisition parameters to be modified:

Critical parameters and clinical cautions for DLR:

• Training data dependency: DLR networks are trained on specific anatomical regions, field strengths, and coil configurations. Applying a brain DLR model to body imaging, or a 3T model to 1.5T data, may produce unexpected results. Use DLR models only within their validated application scope.

• Over-sharpening artefact: some DLR implementations produce images with exaggerated edge enhancement that simulates higher resolution than physically achieved. Small lesion conspicuity may be artificially improved in training-familiar contexts and artificially degraded in unfamiliar contexts.

• Smoothing of genuine small structures: the regularisation inherent in DLR tends to smooth small-amplitude high-spatial-frequency features. Small lesions (< 3 mm) or fine structures (e.g., articular cartilage tears, plantar plate partial tears at the detection threshold) may be smoothed away. This effect is application-specific and should be assessed during DLR implementation in any high-resolution MSK or small-structure programme.

• Quantitative MRI and DLR: ADC values, T1 maps, T2 maps, and fat fraction measurements may be systematically altered by DLR processing. Quantitative protocols should be validated separately with DLR enabled and DLR disabled to ensure measurement consistency.

• DLR and the radiologist: the radiologist should be informed when DLR is applied to an examination. The texture of DLR images differs from conventionally reconstructed images, and comparison with prior studies acquired without DLR may produce apparent changes in image quality that are reconstruction-related rather than pathology-related.

Vendor-specific DLR parameters:

17.5 Quantitative MRI Reconstruction and Mapping

Beyond qualitative imaging, modern scanners increasingly offer quantitative tissue parameter mapping as routine reconstructions.

T1 mapping: inversion recovery or variable flip angle (VFA) methods produce a pixel-by-pixel T1 value in milliseconds. Requires multiple acquisitions at different TI (IR method) or different flip angles (VFA). Key parameters: number of TI values (≥ 5 for good accuracy); accuracy of flip angle (B1 correction required for VFA at 3T); fitting algorithm (Levenberg-Marquardt or similar).

T2 mapping: multi-echo TSE with variable TE acquires multiple T2-weighted images. Pixel-by-pixel T2 fitting (mono-exponential or bi-exponential). Key parameters: minimum number of echoes (≥ 4); echo spacing; Rician noise correction for short-T2 tissues.

*T2 mapping*: multi-echo GRE produces T2 maps. Used for liver iron quantification, myocardial iron, and brain iron. Key parameters: minimum 3 echoes; field-strength-specific TE range; Dixon correction for fat-water interference.

Fat fraction (PDFF) mapping: multi-echo GRE with Dixon processing produces proton density fat fraction maps calibrated in percent (0–100%). Key parameters: 6 echoes at alternating IP/OP TE values; T1 bias correction (low flip angle); T2* correction for iron co-pathology. Cross-validated against MR spectroscopy as the reference standard for liver fat quantification.

18. Group 10 — Scanner Console Software and Post-Processing Tools

18.1 Integrated Console Software

Modern MRI scanners are equipped with integrated software modules that operate at the scanner console, either as part of the standard acquisition workflow or as optional licensed additions. These tools are part of the operational parameter set — they affect what images are generated, how they are processed, and what information is available to the radiologist.

Automatic geometry planning (AAA / SmartExam / Automated Planning): AI-assisted tools that automatically detect anatomical landmarks (head position, spine level, knee orientation) from the localiser and propose the optimal slice prescription for the selected protocol. Available on: Siemens (AAAlign, AutoAlign brain/spine); GE (SmartExam); Philips (SmartExam/AutoSurvey); Canon (AutoScan). These tools reduce operator-dependent variability in slice positioning and improve protocol reproducibility between operators, particularly for standardised examinations (brain MPRAGE for volumetry, spine protocols). The radiologist must verify that automated planning has produced the correct geometry before confirming the acquisition.

Real-time image quality feedback: Siemens Dot (Day Optimised Throughput) workflow provides pre-scan checks that alert the technologist to inadequate shimming, incorrect coil configuration, or patient positioning errors before starting a sequence. GE provides pre-scan shimming quality indicators. These real-time quality indicators should be reviewed for every sequence, not dismissed as routine scanner messages.

Integrated shimming tools: B0 shimming (static and dynamic shimming) adjusts gradient coil currents to minimise B0 field inhomogeneity within the imaging volume. The degree of shim correction directly affects fat suppression quality, EPI distortion, and spectroscopy linewidth. Improved B0 shim → better fat suppression → better spectral resolution. Key shim parameters: shim order (1st order = linear gradients; 2nd order = quadratic; higher order shims for localised regions); shim volume (the ROI over which B0 is minimised — should be placed over the target anatomy, not over the entire bore); update frequency (static: once per scan; dynamic: updated per slice in some implementations).

Integrated perfusion post-processing: DSC perfusion post-processing (CBV, CBF, MTT, TTP, Tmax maps) is available at the scanner console on all major platforms (Siemens: syngo.MR Neuro Perfusion; GE: AW FuncTool DSC; Philips: Perfusion Postprocessing). The technologist must ensure that the DSC EPI acquisition is completed, the arterial input function (AIF) is appropriately sampled (the imaging volume must include the major cerebral arteries), and that the post-processing is triggered before releasing the patient. Maps generated at the scanner console may differ from those generated on dedicated workstations due to differences in the AIF selection algorithm, leakage correction method, and normalisation strategy.

Integrated DWI IVIM and ADC mapping: standard ADC maps are generated automatically on all platforms. Multi-b DWI with IVIM modelling (separating the perfusion component D* from the true diffusion coefficient D) requires dedicated software: Siemens syngo.via; GE AW Suite; Philips IntelliSpace Portal. These tools are not universally available and require licensing.

Integrated spectroscopy post-processing: MRS (MR spectroscopy) raw data requires phase correction, eddy current correction, baseline subtraction, peak fitting, and metabolite quantification — none of which produces a diagnostic image without post-processing. Integrated spectroscopy processing is standard on all platforms: Siemens SYNGO MR Spectroscopy; GE FuncTool Spectroscopy; Philips SpectroView. The technologist must verify that the spectroscopy post-processing has completed and produced recognisable metabolite peaks (Cho, Cr, NAA) before releasing the patient — a flat or uninterpretable spectrum usually indicates a technical failure (incorrect voxel placement, inadequate water suppression, severe lipid contamination) that can be corrected at the time of the examination.

Integrated diffusion tensor tractography: DTI tractography requires: diffusion tensor calculation from multi-direction DWI; eigenvector extraction; probabilistic or deterministic tracking. Console-integrated tractography is available on major platforms for pre-surgical planning (Siemens syngo.Neuro DTI; GE AW FuncTool DTI; Philips FiberTrak). These tools produce visual renderings of white matter tracts that must be interpreted with knowledge of their limitations (noise sensitivity, tensor model limitations in crossing fibres).

18.2 AI-Assisted Detection and Characterisation Tools

The integration of AI detection and characterisation algorithms at the scanner console or as cloud-connected services represents an emerging operational parameter that affects protocol execution.

AI-triggered protocol modification: some AI tools analyse the localiser or early diagnostic sequences in real-time and suggest protocol additions. For example: detection of a vertebral compression fracture on the localiser may trigger a STIR sequence addition before the full protocol is confirmed. These tools require the technologist to understand the suggestion, validate it against the clinical indication, and implement it appropriately.

Automated measurement tools: integrated tools on all major platforms provide automated measurements (ventricular volumes, lesion size, organ volumes) that must be verified by the radiologist. Automated hippocampal segmentation for AD volumetry, automated liver volume for pre-surgical planning, and automated cardiac ejection fraction from cine bSSFP are standard integrated tools. Their output is provided as an automatic preliminary calculation, not as a final diagnostic measurement.

Quality scoring tools: Siemens and GE offer integrated image quality scoring that assesses sharpness, noise level, and fat suppression quality per sequence. These scores provide objective documentation of technical quality — useful for quality control reporting, audit, and accreditation. The technologist should review these scores during the examination and use them as an objective supplement to visual quality assessment.

18.3 Post-Processing That Must Be Completed at the Scanner Before Patient Release

Several post-processing operations must be completed before releasing the patient, because they cannot be retrospectively performed if source data has been deleted or if the patient has left:

19. Updated Evidence Gaps and Ongoing Debate

In addition to the evidence gaps described in Section 13, the following specific to the new parameter groups are noted:

Respiratory trigger window standardisation: the optimal navigator acceptance window (5 mm vs 8 mm vs 10 mm) for 3D MRCP and liver SPACE has not been systematically compared in prospective trials. The 5 mm standard is based on early navigator papers and expert consensus, not on controlled outcome data.

DLR and quantitative MRI validation: the impact of DLR on ADC values in clinical populations (vs phantoms) has been documented in limited series. The magnitude of systematic ADC bias from DLR is vendor-specific and sequence-specific. No cross-vendor, multi-site, prospective validation of DLR effect on ADC across all major body applications has been published.

CS regularisation parameter optimisation: the λ (regularisation strength) for clinical CS MRI is typically fixed by the vendor at a single value for each approved application. Whether a single λ is optimal across the full range of patients and pathologies for a given application has not been prospectively validated. The possibility that optimal λ differs between patients with large vs small lesions, or between high vs low SNR acquisitions, is an open research question.

MOLLI and ShMOLLI accuracy at clinical heart rates: T1 mapping accuracy with MOLLI-type acquisitions degrades at heart rates > 90 bpm and with heart rate variability > 20%. The optimal cardiac T1 mapping sequence for clinical populations with arrhythmia or tachycardia has not been established by a definitive head-to-head trial. This is particularly relevant given the increasing use of T1 mapping in atrial fibrillation and heart failure populations where elevated heart rates are common.

New parameter deep dive: Reconstruction Matrix, Pixel Interpolation, and Slice Interpolation.

New parameter deep dive: Parallel Imaging; NEX / NSA.

End of document — MRI Parameters Overview and Classification — MRIninja v1.0 — May 2026 Companion page: MRI Sequences — Overview and Classification (9003) Child pages: TR/TE/TI optimisation by sequence; Parallel imaging deep dive; Dixon fat-water separation; b-value selection for DWI; SAR management at 3T; Respiratory and cardiac synchronisation in body MRI; Deep learning reconstruction clinical guide; Navigator efficiency optimisation; Quantitative MRI parameter standards; Field-strength parameter translation; FOV — Field of View; Acquisition Matrix; Slice Thickness; 2D vs 3D Acquisition; REST Slab / Presaturation Band; Parallel Imaging.