2D vs 3D Acquisition

MRIninja Knowledge Base | MRI Parameter Deep Dive Version 1.0 — May 2026

1. Introduction and General Purpose

The choice between 2D and 3D acquisition is one of the most consequential decisions in MRI protocol design. It determines not only the spatial resolution achievable in the through-plane direction, but also the SNR efficiency, the artefact profile, the susceptibility to motion, the flexibility of post-acquisition reformatting, and — fundamentally — how the MRI scanner encodes the third dimension of space.

In 2D acquisition, each slice is an independent measurement. The z-dimension (through-plane) is selected, not encoded: a frequency-selective RF pulse, in combination with a slice-selection gradient, excites protons only within a defined slab. Each slice is acquired separately, sequentially or in interleaved groups. The z-resolution is fixed at acquisition time and cannot be changed retrospectively. The slice profile is determined by the RF pulse shape — inevitably imperfect, with transition zones and cross-talk consequences.

In 3D acquisition, the entire imaging volume is excited simultaneously and the z-dimension is encoded by a second phase-encoding gradient, exactly as the y-dimension is encoded in 2D. This produces a 3D k-space dataset that is reconstructed by a 3D inverse Fourier transform. The result is a set of contiguous, gap-free, arbitrarily thin partitions with perfectly rectangular z-profiles — reformattable into any plane without loss of resolution.

These two approaches represent genuinely different imaging strategies, not simply different values of the same parameter. Understanding when each is superior — and why — requires understanding the physics of each approach, not just memorising a list of applications.

Historical context: early clinical MRI was exclusively 2D (1980–1988), limited by gradient hardware and computing power. The first clinical 3D protocols emerged in the late 1980s for brain imaging (3D SE T1 at 1.5T) and became clinically widespread with the introduction of MPRAGE (1990 [1]) and 3D FLASH/VIBE for body MRI. The breakthrough for musculoskeletal 3D MRI came with the development of variable flip angle (VFA) TSE sequences (SPACE/CUBE/VISTA, 2006 onwards) that enabled isotropic T2-weighted 3D imaging of joints at diagnostic quality.

2. Physical Foundations

2.1 The Fundamental Encoding Difference

2D acquisition — z-selection by RF pulse:

The slice is defined by the interaction of a selective RF pulse with a slice-selection gradient G_z. The excited slice thickness:

Δz = BW_RF / (γ × G_z)

The z-direction is not phase-encoded. Each slice is an independent measurement. The through-plane spatial information is predetermined at excitation and cannot be reconstructed at different z-resolutions.

Consequences:

• Slice profile is a convolution of the RF pulse profile with the gradient-imposed frequency distribution → imperfect rectangle with transition zones
• Adjacent slices interact: the transition zones of adjacent RF pulses overlap → cross-talk → signal saturation if not managed with interleaving or gaps
• No MPR: the image plane is fixed at acquisition time

3D acquisition — z-encoding by phase-encoding gradient:

The entire imaging volume (slab) is excited by a non-selective or wide-bandwidth RF pulse. A second phase-encoding loop is applied in the z-direction:

Δz = FOV_z / N_z

where N_z is the number of z-phase-encoding steps (partitions). This is exactly analogous to the in-plane phase encoding:

Δy = FOV_y / N_y

The 3D k-space is filled by applying all combinations of (k_y, k_z) phase-encoding steps, for each of which the full readout gradient produces N_x frequency-encoded points.

Consequences:

• Slice profile is the Fourier transform of the rectangular k_z window → perfect rectangle in the ideal case, with z-direction Gibbs ringing at partition boundaries (analogous to in-plane Gibbs ringing)
• No cross-talk: all partitions share the same TR; there is no separate excitation per partition
• Complete MPR freedom: the 3D dataset is reformattable in any plane at the intrinsic voxel resolution

2.2 Mathematical Foundations

Acquisition time comparison

2D: T_acq_2D = TR × N_y × NSA / ETL

3D: T_acq_3D = TR × N_y × N_z × NSA / ETL

The 3D acquisition time is longer by a factor of N_z compared with a single 2D slice — but this covers N_z slices simultaneously. The time per slice is:

T_per_slice_3D = TR × N_y × NSA / ETL (identical to the 2D case per slice!)

This means: for equal slice coverage, 2D and 3D have identical acquisition time (in the idealised case). However, in practice, 3D acquires all partitions in a single, continuous k-space filling process with no gaps — which is its key efficiency advantage in scenarios where many thin slices are needed.

SNR advantage of 3D

The fundamental SNR advantage of 3D over 2D (at matched voxel dimensions) is:

SNR_3D / SNR_2D = √N_z

where N_z is the number of z-partitions. This advantage arises because in 3D, the signal from all N_z slices contributes to each partition reconstruction (via the 3D Fourier transform), whereas in 2D, each slice is independent. At N_z = 64: SNR_3D = 8 × SNR_2D at matched voxel size.

Practical implication: 3D enables thinner partitions at the same SNR as 2D, or equivalently: for a given voxel size, 3D produces substantially higher SNR than 2D. This is the primary physical reason why 3D is preferred for thin-slice high-resolution imaging.

3D isotropic resolution

When the three voxel dimensions are equal:

Δx = Δy = Δz = a (isotropic voxel)

The image can be reformatted in any plane — coronal, sagittal, axial, or any oblique — with identical spatial resolution in all directions. The reformatted image quality is limited only by the interpolation algorithm, not by reduced resolution in any direction.

Verification: Δx = FOV_x / N_x; Δy = FOV_y / N_y; Δz = FOV_z / N_z. Isotropic when all three are equal.

2D vs 3D SNR at matched coverage

For a given anatomical coverage (C mm), N_z = C/Δz partitions, and:

SNR_3D / SNR_2D = √(C/Δz)

This means: the thinner the partition relative to the total coverage, the greater the 3D SNR advantage. For brain coverage 160 mm at 1 mm partitions (N_z=160): SNR_3D / SNR_2D = √160 = 12.6 — a massive advantage that makes sub-millimetre 3D brain MRI feasible at clinical field strengths.

3. Units, Terminology and Vendor Nomenclature

2D vs 3D is a binary acquisition design choice, not a numerical parameter. It is expressed in protocol descriptions as "2D" or "3D" acquisition mode.

4. Typical Value Ranges

4.1 2D vs 3D by Application

4.2 3D Voxel Size by Field Strength

5. Parameter Interaction Ecosystem

5.1 Parameter Relationships Matrix

5.2 The 2D vs 3D SNR Advantage in Context

The √N_z SNR advantage of 3D is real but must be interpreted correctly:

However, the total acquisition time for 3D is also N_z × longer than for a single 2D slice. The advantage manifests only when the full 3D volume is needed and the entire scan time is invested in a single 3D acquisition rather than multiple 2D slices.

6. Effects on Image Appearance

6.1 3D vs 2D — Direct Comparison

Slice/partition profile: 2D produces imperfect RF-defined profiles with transition zones. 3D produces Fourier-defined rectangular profiles. At identical nominal thickness, 3D partitions are slightly sharper in z — contributing to better through-plane resolution.

Cross-talk and signal uniformity: 2D multi-slice acquisitions show subtle slice-to-slice signal variation from cross-talk even with interleaving and gaps. 3D acquisitions show no such variation — all partitions are acquired in the same steady-state magnetisation condition. For longitudinal signal comparisons (treatment response, serial examination), 3D is more reproducible.

MPR quality: a 2D acquisition with 4 mm slices reformatted into the sagittal plane produces a blocky, low-resolution image in the through-plane direction. A 3D isotropic acquisition at 1 mm reformatted into any plane produces a 1 mm resolution image in all directions — qualitatively superior for complex anatomy and oblique pathology.

Slab boundary effects in 3D: the edges of the 3D slab show transition zones from the slab-excitation RF pulse. The first and last 10–15% of partitions may show reduced signal. Slice oversampling (see Slice Thickness page) adds extra partitions beyond the clinical target to ensure that the usable partitions are all within the flat-top region of the slab profile.

T2 blurring in 3D TSE (VFA): 3D TSE (SPACE/CUBE) uses very long ETLs (50–200 echoes) with variable flip angle modulation to maintain signal throughout the echo train. Despite VFA optimisation, peripheral k-space lines (filled by late echoes) still show more T2 decay than early echoes. This produces a T2-dependent blurring that is more pronounced in 3D TSE than in 2D TSE at equivalent ETL. Short-T2 structures (cortical bone, dense fibrous tissue, calcification) appear more blurred in 3D TSE vs 2D TSE.

6.2 Motion Sensitivity

2D robustness: motion between slices affects only the specific slice acquired during the motion event. Adjacent slices are acquired independently and are not directly affected. A single corrupted slice in a 40-slice 2D dataset is a minor artefact (report the limitation; adjacent slices are diagnostic).

3D vulnerability: motion during a 3D acquisition affects the entire 3D k-space. Motion corrupts specific k_y or k_z lines that contaminate all partitions when the 3D Fourier transform is applied. A brief motion event during a 10-minute 3D SPACE acquisition can produce ghosting artefacts throughout the entire 3D volume.

Practical rule: 3D acquisitions require better patient cooperation than 2D for equivalent image quality. For uncooperative patients (paediatric, claustrophobic, pain-limited), 2D provides more robust partial diagnostics.

7. Effects on Acquisition Time

7.1 Direct Comparison at Matched Coverage and Resolution

For a given anatomical coverage (C mm) at a given voxel size (Δz mm in z):

N_z = C/Δz

3D: T_3D = TR × N_y × N_z × NSA / ETL = TR × N_y × (C/Δz) × NSA / ETL

2D: T_2D = TR × N_y × NSA / ETL (per slice) × N_slices = TR × N_y × (C/Δz) × NSA / ETL

The total acquisition times are identical (for the idealised case of equal ETL, TR, N_y, NSA, and no gaps). The difference is:

• 2D: the scan time is divided into N_z independent acquisitions; each slice has its own excitation
• 3D: the scan time is one continuous acquisition

7.2 Where 3D Is More Efficient

Thin partitions: at very small Δz (< 2 mm), 3D is more time-efficient because:

• 2D at 1 mm requires very long TR (many slices × T_per_slice) → TR may be forced to > 10 seconds → long scan
• 3D at 1 mm has no TR-slice constraint; TR is set purely by contrast requirements; the N_z partitions are encoded within each TR

Higher ETL in 3D: VFA TSE enables ETL 50–200 in 3D vs typically ETL 12–30 in 2D TSE. The time savings from very long ETL in 3D are substantial. A 3D SPACE brain with ETL=100 is 5–10× faster per covered volume than an equivalent 2D TSE at ETL=16.

2D vs 3D time examples (brain T2, 160 mm coverage, 1 mm partition/slice):

This is the fundamental reason why 1 mm isotropic brain imaging is exclusively done with 3D protocols: 2D is physically impractical.

8. Effects on SNR and CNR

8.1 The √N_z SNR Advantage — Practical Consequences

The 3D SNR advantage over 2D (at matched voxel size) is √N_z. This advantage has two major clinical applications:

Application 1 — thinner partitions at same SNR: if 3D provides 8× more SNR (N_z=64) than 2D at the same voxel, the 3D can afford voxels 8× smaller by volume while maintaining the same SNR. At fixed in-plane dimensions: Δz_3D = Δz_2D / 8. In practice, the usable gain is partially consumed by other factors (VFA blurring, slab edge effects), but 3D consistently enables 3–5× thinner effective partitions at equivalent SNR.

Application 2 — faster acquisition at same SNR and voxel size: the √N_z SNR advantage can instead be converted to scan time reduction. A 3D acquisition at N_z=64 has the same SNR as a 2D acquisition at NSA=64 (since SNR_3D = √N_z × SNR_2D-1NSA, and NSA=N_z → SNR_2D-NSA64 = √64 × SNR_2D-1NSA). The 3D acquires this SNR in T_3D = TR × N_y × N_z / ETL; the equivalent 2D would require T_2D = TR × N_y × 64 / ETL — identical total time but distributed differently.

8.2 3D at Different Field Strengths

9. Artefacts Associated with 2D and 3D

10. Behaviour Across Sequence Families

Spin Echo (SE)

Standard SE is almost exclusively 2D. 3D SE is possible but extremely time-consuming (no ETL; N_y × N_z TR periods per volume). Impractical for clinical use. The 3D SE was historically used for brain T1 in early MRI (1987–1992) before faster alternatives became available.

Turbo Spin Echo (TSE/FSE)

Both 2D and 3D TSE are routine. 2D TSE (ETL 8–20): brain T2, spine T2, joint PD/T2 — the clinical workhorse. 3D TSE (SPACE/CUBE/VISTA, ETL 50–200 with VFA): brain volumetric T2, joint cartilage, MRCP, inner ear, brachial plexus root sleeves. The VFA flip angle scheme is specific to 3D TSE and has no 2D equivalent — it is the enabling technology for practical 3D TSE.

Gradient Echo (GRE/FLASH/SPGR)

Both 2D and 3D GRE are widely used. 2D GRE: single-slice cardiac, fast survey. 3D GRE (VIBE/LAVA/THRIVE): abdominal DCE, liver fat quantification, DCE breast, head and neck post-contrast. 3D GRE is the standard for any volumetric post-contrast T1 body imaging.

Inversion Recovery (STIR, FLAIR, MPRAGE)

2D STIR: whole-body, extremities, neck — the standard for fat-suppressed T2. 3D STIR is used for WB-MRI research but is less common due to SAR constraints. 3D FLAIR: brain ARIA monitoring, white matter disease quantification. 3D MPRAGE/BRAVO/TFE: brain volumetry — exclusively 3D. MPRAGE cannot be performed in 2D (it requires the 3D inversion-prepared GRE design to achieve its T1 contrast and resolution).

EPI (DWI, fMRI, DSC)

EPI is almost exclusively 2D. The extremely short TR of single-shot EPI per slice (< 100 ms) makes 2D the natural approach — a full 2D EPI slice is acquired in one shot. 3D EPI exists (for ASL, some fMRI) but requires multi-shot approaches (readout-segmented EPI, rs-EPI), which are used specifically for DWI quality improvement (rs-EPI / RESOLVE / MUSE) — not for whole-brain volumetric EPI in the standard sense. Standard DWI: 2D. Advanced DWI (rs-EPI): technically a multi-shot 2D with partial 3D k-space concepts.

DCE (Dynamic Contrast Enhanced)

3D GRE is the universal standard for DCE in all body regions (liver, breast, prostate, head and neck). The temporal series of 3D volumes provides both kinetic information and volumetric coverage simultaneously. 2D DCE (single-slice or few-slice dynamic) was used before 3D became practical and remains appropriate for some targeted applications (small tumour at known location).

ASL (Arterial Spin Labelling)

3D readout (GRASE or stack-of-spirals): the standard for modern pCASL. The 3D readout provides whole-brain perfusion coverage in 3–5 minutes per measurement. 2D EPI-based ASL is possible but less efficient. The labelling plane (in the neck, for pCASL) is always a 2D-selective RF pulse; only the readout is 3D.

bSSFP (TrueFISP/FIESTA/CISS)

Both 2D and 3D bSSFP are used. 2D bSSFP: cardiac cine (gold standard for ventricular function). 3D bSSFP (CISS/FIESTA-C/DRIVE): inner ear, brachial plexus root sleeves, CSF cisterns, small joints (3D DESS). The 3D CISS for inner ear and brachial plexus root sleeves is one of the most technically demanding 3D applications — sub-millimetre isotropic at 3T with bSSFP banding management.

Spectroscopy (SVS/MRSI)

SVS is a "pseudo-2D" application (slice-selected in all three dimensions → a 3D rectangular voxel). MRSI can be 2D (one-slice chemical shift imaging) or 3D (full-volume). 3D MRSI provides volumetric metabolite mapping but is time-intensive and technically demanding. Most clinical MRSI is 2D.

11. Field Strength Behaviour

At 3T, the 3D case is strongest: the combined advantage of √N_z SNR × 1.7 (intrinsic 3T SNR) makes thin isotropic 3D imaging at 3T substantially superior to any 2D equivalent in most applications. SAR management (particularly for 3D TSE) is the primary limitation at 3T.

At 0.55T, 2D generally prevails: SNR at 0.55T is too limited for thin-partition 3D imaging without DLR. The √N_z advantage exists but the base SNR is so low that even √N_z × SNR_0.55T may be below diagnostic threshold for thin partitions. Thick 3D (2–3 mm, N_z = 50–80) with DLR is achievable; sub-millimetre 3D is not.

12. Vendor-Specific Implementation

Siemens

3D TSE is implemented as SPACE (Sampling Perfection with Application-optimised Contrasts using different flip angle Evolution). The VFA scheme is T2-contrast weighted by default but can be modified for T1-weighting (T1-SPACE) or FLAIR-like weighting (FLAIR-SPACE). SAR management at 3T is handled by automatic TR extension if the prescribed ETL and flip angle schedule exceed the SAR limit — the technologist must monitor for auto-TR increase, which changes T1 saturation and scan time unexpectedly. MPRAGE is the standard 3D T1 sequence; the ADNI-protocol MPRAGE parameters are supported as a Siemens product sequence.

GE

3D TSE is CUBE (Controlled Aliasing in Parallel Imaging Results in Higher Acceleration). The VFA scheme is vendor-specific and adapted for CUBE's parallel imaging integration. The LAVA (Liver Acquisition with Volume Acceleration) and LAVA-Flex (with Dixon fat-water separation) are the primary 3D GRE implementations for body imaging. BRAVO is the 3D T1 brain sequence equivalent to MPRAGE.

Philips

VISTA (Volume Isotropic Turbo spin echo Acquisition) is the Philips 3D TSE. THRIVE (T1 High-Resolution Isotropic Volume Excitation) is the 3D T1 GRE body sequence. The Philips 3D FLAIR implementation provides the direct FLAIR display with isotropic voxels. DRIVE is an alternative to CISS for inner ear and CSF imaging, using a RARE-based readout instead of bSSFP — avoids the banding artefact problem of CISS at 3T.

Canon

isoFSE is the Canon 3D TSE with VFA. Canon's 3D GRE for body imaging follows standard 3D FLASH principles. AiCE deep learning reconstruction is integrated into Canon's 3D protocols to enable thinner partitions at maintained SNR.

United Imaging

UIH 3D protocols follow Siemens-equivalent nomenclature. The uCS (Compressed Sensing) integration in UIH scanners provides significant acceleration for 3D acquisitions, enabling body 3D protocols within breath-hold duration.

Hidden coupling — all vendors: when 3D is selected, the scanner always adds slice oversampling automatically (typically 15–25%). The total FOV_z displayed to the technologist includes this oversampling; the diagnostic FOV_z is smaller. Verify that the diagnostic anatomy is within the central (non-oversampled) portion of the slab.

13. Practical Optimisation Strategies

13.1 Clinical Optimisation Recipes

14. Parameter Extremes

14.1 Extremely Thin 3D Partitions (< 0.5 mm)

Sub-half-millimetre 3D partitions are currently feasible only at 7T for targeted brain imaging. Applications: cortical layer imaging (4–6 cortical layers, each ~0.5–0.8 mm); individual hippocampal subfields; inner ear structures (scala tympani, scala vestibuli, organ of Corti).

At 1.5T and 3T, sub-0.5 mm isotropic 3D requires surface coils, very long scan times (10–20+ min), or DLR — and may produce images where the VFA blurring exceeds the nominal partition thickness (making the actual z-resolution worse than the nominal value suggests).

14.2 Very Large 3D Slab (Whole-Body 3D)

The WB-MRI 3D concept (a single 3D slab covering the entire body, 150 cm) is technically possible but practically limited by: gradient non-linearity over 150 cm; B0 inhomogeneity requiring per-station shimming; SAR accumulation; and coil sensitivity fall-off. Current WB-MRI is achieved with multiple overlapping stations (4–5 stations of 35–45 cm each), each acquired as a separate 2D or 3D protocol. True single-slab whole-body 3D remains a research concept.

15. Common Optimisation Errors

16. MRI Technologist Pearls

The 2D/3D decision is made before the protocol starts — not recoverable afterwards: unlike most parameter adjustments (matrix, NSA, TR), the 2D/3D architecture is a commitment. A 2D acquisition cannot be retrospectively converted to isotropic MPR data. Verify the intended acquisition mode before starting.

3D always needs oversampling: without exception, enable 15–20% slice oversampling for every 3D protocol. The 30–60 seconds of extra scan time prevents z-wrap artefact that can simulate or obscure pathology at the slab edges.

VFA-TSE (SPACE/CUBE) is not equivalent to 2D TSE for internal joint signal: the long VFA ETL that enables 3D TSE also produces T2 blurring of internal joint signals. The internal meniscal signal, the tendon internal architecture, and cortical bone detail are all blurred in 3D TSE compared with optimised 2D PD-weighted TSE. Use 3D TSE for cartilage; use 2D TSE for meniscus and ligament internal signal.

Verify 3D slab coverage on localiser before starting: check that the full anatomical target is within the central 80% of the slab (not within the oversampled edge zones). For inner ear 3D, verify both cochleae are within the slab. For brachial plexus 3D CISS, verify that C4–C5 through T1 foramina are all within the slab.

Monitor scan time after entering 3D parameters: in 3D, the time impact of N_z changes is immediately visible on the console. A planned 5-minute 3D SPACE that becomes 8 minutes because the technologist added 10 more partitions for coverage will not fit within a standard slot. Check time after every parameter change.

3D CISS banding artefact at 3T — adjust frequency before the diagnostic acquisition: at 3T, the bSSFP banding artefact in 3D CISS can be shifted by adjusting the centre frequency (±100–200 Hz steps). Do this on a test acquisition before running the full 3D dataset. The goal is to position the banding node outside the foraminal level where root sleeves are assessed.

17. Real Clinical Examples

Example 1: Brain Volumetry for AD Follow-Up — 3D MPRAGE Essential

Clinical scenario: a 72-year-old with confirmed AD on lecanemab therapy. Baseline MPRAGE acquired; 6-month follow-up planned for ARIA monitoring and atrophy progression.

2D option (hypothetical): T1 sagittal 2D, 1 mm slices, N=160. T_per_slice = 150 ms; TR required = 160 × 150 = 24000 ms → scan = 24000 × 256/16 = 384 s = 6.4 min. Slice profiles imperfect; cross-talk at zero gap; not reformattable; hippocampal volume not calculable by automated tools.

3D MPRAGE: TR=2300/TI=900/TE=3 ms; 1 mm isotropic; N_z=192; R=2; ETL=1 (single echo per TR in MPRAGE). Scan = 2300 × 256 × 192 / (2 × 1) = ~56 s? No — MPRAGE is structured differently (TR × number of segments per inversion block × inversion blocks). In practice: ≈ 4.5–6 min. Provides: isotropic 1 mm voxels; rectangular z-profiles; FreeSurfer/Clinica-compatible; ADNI-compliant hippocampal volume; direct comparison with baseline.

Verdict: 2D is theoretically feasible but produces inferior data. 3D MPRAGE is the only appropriate choice for AD longitudinal monitoring.

Example 2: Knee Cartilage Mapping — 3D SPACE vs 2D PD-FS

Clinical scenario: 45-year-old with knee pain; surgeon requests cartilage map for osteochondral allograft planning. The cartilage surface assessment in all compartments is required.

2D PD-FS protocol (3-plane, 3 mm slices): axial + sagittal + coronal acquisitions, each 3–4 min. Total: ~12 min. Covers: meniscal signal; ligament internal signal; bone marrow oedema; global cartilage survey at 3 mm.

3D SPACE T2 (0.6 mm isotropic, 3T): single 3D volume; MPR in any plane; thinner partitions. Time: 8 min.

Trade-off: 3D SPACE provides better cartilage surface mapping and quantification. 2D PD-FS provides better meniscal signal assessment and ligament internal signal.

Optimal protocol for surgical planning: both. 3D SPACE for cartilage; 2D PD-FS sagittal for meniscus + ACL/PCL signal; 2D PD-FS coronal for collateral ligaments. Total: ~20 min.

Lesson: 3D does not replace 2D for joint MRI — they answer different diagnostic questions.

Example 3: Motion Disaster — 3D in Uncooperative Patient

Clinical scenario: 78-year-old with severe osteoarthritis and hip pain; unable to lie still for > 5 minutes. Protocol includes 3D T1 MPRAGE (7 min) for brain atrophy assessment.

Problem: the patient moved at minute 4 of the 7-minute MPRAGE. Motion artefact affects all 192 partitions. The entire 3D dataset is corrupted. The 2D T2 sequences (acquired first) are diagnostic.

Lesson: in uncooperative patients: 1. Acquire the most critical sequences first (2D is more motion-robust) 2. For cognitive patients who cannot cooperate with 7-minute 3D MPRAGE, use a 4-minute accelerated version (CS or DLR enabled); or accept 2D T1 for qualitative MTA assessment 3. Document the acquisition mode failure in the report

Example 4: MRCP — Both 2D and 3D Are Required

Clinical scenario: jaundice; suspected bile duct stone or stricture. MRCP required.

2D thick-slab MRCP (40–70 mm, 5–7 projections, 3–5 s breath-hold each): total 1–2 min. Provides: the overview "ERCP-equivalent" projection; rapid whole-system survey; completes in < 5 s per projection → useful for claustrophobic patients or those with limited breath-hold.

3D navigator-triggered MRCP (SPACE/CUBE/VISTA, 1.0–1.5 mm isotropic, free-breathing, navigator): 5–8 min. Provides: thin isotropic partitions; MPR in any plane; measurement of stricture length; detection of small stones (< 5 mm); normal variation vs pathology at branch points.

Complementarity: the 2D thick-slab provides immediate overview; the 3D provides detail. Neither alone is sufficient for complete MRCP assessment. The 2D projection missed a 3 mm common bile duct stone clearly visible on 3D MPR in 50% of cases in one landmark study.

Acquisition order: always acquire the 2D thick-slab projections before the 3D navigator. If the 3D fails (poor navigator efficiency, patient motion), the 2D projections provide the clinical overview.

18. Visual Educational Material

18.1 2D vs 3D Encoding Architecture

2D ACQUISITION:
  Z-direction: SELECTED by RF pulse (not encoded)
  ┌─────────────────────────────────────────┐
  │ Slice 1: RF excitation → read/phase → ADC │
  │ Slice 2: RF excitation → read/phase → ADC │  (interleaved)
  │ ...                                       │
  │ Slice N: RF excitation → read/phase → ADC │
  └─────────────────────────────────────────┘
  Z-profile: imperfect (RF pulse shape)
  Cross-talk: managed by interleaving + gap
  MPR: not isotropic

3D ACQUISITION:
  Z-direction: ENCODED by phase-encoding gradient k_z
  ┌─────────────────────────────────────────┐
  │ TR 1: k_y=1, k_z=1 → read → ADC        │
  │ TR 2: k_y=1, k_z=2 → read → ADC        │
  │ ...                                     │
  │ TR N_y×N_z: k_y=N_y, k_z=N_z → ADC     │
  └─────────────────────────────────────────┘
  Z-profile: rectangular (Fourier)
  Cross-talk: none
  MPR: isotropic (Δx = Δy = Δz)

18.2 The 2D/3D Decision Tree

CLINICAL QUESTION REQUIRES:

Thin slices (< 2 mm)?
  → YES → 3D (2D impractical: TR constraint, cross-talk, SNR)

Isotropic MPR capability?
  → YES → 3D

Motion-robust acquisition for uncooperative patient?
  → YES → 2D (individual slices salvageable; 3D entire volume lost)

Fast per-slice acquisition (< 2 min per plane)?
  → YES → 2D

Reproducible serial volumetry (brain, tumour)?
  → YES → 3D (consistent profiles; automated segmentation compatible)

Standard clinical joint (meniscus/ligament internal signal)?
  → YES → 2D PD-weighted (VFA blurring in 3D TSE degrades internal signal)

Dynamic series with temporal resolution < 30 s?
  → Depends: 3D GRE for volume coverage; 2D EPI for fast brain perfusion

18.3 SNR Comparison Diagram

SNR_3D / SNR_2D = √N_z

N_z = 10:   √10  = 3.2×  advantage
N_z = 64:   √64  = 8×    advantage
N_z = 160:  √160 = 12.6× advantage (MPRAGE)
N_z = 256:  √256 = 16×   advantage (thin partition protocols)

PRACTICAL USE OF THE √N_z ADVANTAGE:
Option A: Keep SNR constant → reduce partition thickness by √N_z
  (make thinner partitions at same noise level as thicker 2D slices)

Option B: Keep partition thickness constant → faster acquisition
  (3D at N_z partitions has same SNR as 2D with N_z averages)

Option C: Split advantage between A and B
  (moderate thickness reduction + moderate time reduction)

19. Evidence Gaps and Ongoing Debate

3D TSE vs 2D TSE for knee MRI diagnostic accuracy: multiple studies have compared 3D SPACE/CUBE with 2D PD-FS for meniscal tear detection, finding overall comparable sensitivity and specificity. However, individual tear pattern characterisation (horizontal vs complex tear) may differ between protocols because of VFA blurring effects on internal meniscal signal. The "best" knee MRI protocol for a comprehensive diagnostic evaluation remains centre-dependent and surgeon-preference dependent.

Prostate 3D T2 vs 2D T2 PI-RADS compliance: PI-RADS v2.1 was developed and validated primarily using 2D T2 protocols. Several centres have adopted 3D T2 (0.8–1 mm isotropic) as an alternative, with some studies showing equivalent or superior lesion detection. However, PI-RADS formally recommends axial 2D T2 with ≤ 3 mm slices as the primary standard, and 3D protocols remain "acceptable alternatives" without the same level of validation.

Motion-robust 3D: radial k-space sampling and PROPELLER-equivalent 3D acquisitions (XD-GRASP, Koosh-ball trajectories) offer some motion correction for 3D acquisitions. Their clinical utility for applications where patient motion is a limiting factor (paediatric, pain, dementia) has been demonstrated in research settings but is not yet standardised in clinical protocols.

7T and the 3D advantage ceiling: at 7T, B1+ inhomogeneity severely limits the usable FOV for uniform 3D imaging. Parallel transmit (pTX) technology can partially address this, but the practical clinical deployment of 7T 3D imaging beyond the brain remains limited. Whether the theoretical SNR/resolution advantages of 7T 3D will translate to routine clinical diagnostic benefit has not been demonstrated in multi-centre prospective trials.

AI denoising and minimum viable partition thickness in 3D: DLR (Deep Resolve, AIR Recon DL) applied to 3D acquisitions enables thinner partitions at maintained image quality. The minimum clinically viable partition thickness with DLR-enabled 3D acquisitions has been characterised for specific applications (MPRAGE, prostate) but not comprehensively validated across all body regions and sequence types.

20. Miscellaneous and Future Directions

Historical milestone — MPRAGE as the 3D template: the MPRAGE sequence (Mugler and Brookeman, 1990 [1]) was not the first 3D MRI protocol, but it became the universal template for understanding the practical advantages of 3D over 2D for brain imaging. Its combination of T1-weighted inversion preparation with a spoiled GRE 3D readout at 1 mm isotropic — enabling cortical surface reconstruction, hippocampal volumetry, and multiplane reformatting — established the 3D acquisition as the preferred approach for brain structural MRI.

Variable flip angle TSE — enabling 3D T2: the introduction of VFA schemes for 3D TSE (SPACE by Siemens 2006; CUBE by GE; VISTA by Philips) was the enabling technology for practical 3D T2-weighted imaging of joints, the spine, and the abdomen. Before VFA, 3D TSE was impractical because the long echo trains required produced severe T2 signal collapse. VFA compensated for this by modulating flip angles to maintain signal throughout the 200-echo train.

XD-GRASP and motion-resolved 3D: the Extended Dynamic GRASP (XD-GRASP) reconstruction algorithm treats respiratory motion as an additional encoding dimension, simultaneously resolving multiple respiratory phases from a continuously acquired 3D dataset without gating. This enables free-breathing 3D liver MRI at near breath-hold quality — eliminating the largest practical limitation of 3D body imaging.

Synthetic MRI — one 3D acquisition replacing multiple 2D: quantitative MRI (T1, T2, PD maps from a single acquisition) combined with synthetic image generation in principle eliminates the need for multiple separate 2D and 3D acquisitions. From one 3D quantitative dataset, any contrast-weighted image (T1w, T2w, PD, FLAIR, DIR) can be synthesised in any plane. This may fundamentally change the 2D/3D decision: if a single isotropic 3D quantitative dataset provides all contrasts, the distinction between "2D T2" and "3D T2" becomes irrelevant.

Related parameter deep dive: Parallel Imaging explains acceleration factor, g-factor, SENSE/GRAPPA reconstruction and SMS/Multiband trade-offs relevant to this parameter.

End of document — 2D vs 3D Acquisition — MRIninja v1.0 — May 2026 Parent page: MRI Parameters — Overview and Classification (9501) Related child pages: FOV — Field of View · Acquisition Matrix · Slice Thickness · ETL / Turbo Factor · VFA Echo Train Design · Parallel Imaging (R) · Motion Artefact Reduction.