Slice Thickness

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

1. Introduction and General Purpose

Slice thickness defines the extent of tissue sampled in the through-plane direction — the dimension perpendicular to the image plane. Together with in-plane resolution (determined by FOV and matrix), it completes the three-dimensional description of the voxel: the fundamental resolved volume element of every MRI acquisition.

In 2D acquisitions, slice thickness is the only spatial parameter that is not Fourier-encoded. While in-plane resolution is determined by gradient encoding and k-space sampling, the slice direction is defined by the interaction of a selective RF pulse with a slice-selection gradient — a fundamentally different physical mechanism with distinct implications for image quality, artefacts, and optimisation.

Slice thickness is the third dimension of the voxel but behaves differently from the other two. It directly determines: how much tissue contributes to each pixel's signal (the partial volume problem); how many slices are needed to cover a given anatomical extent (the coverage problem); and how many slices can be acquired within a given TR (the throughput problem). Getting slice thickness wrong — too thick for the target structure, or too thin for the available SNR — is one of the most common sources of missed diagnoses and non-diagnostic images in clinical MRI.

Historical evolution: the earliest clinical MRI systems in 1981–1984 used slice thicknesses of 10–15 mm, driven by low SNR from primitive RF coils and low field strengths. As hardware improved through the late 1980s and 1990s, routine clinical brain and MSK imaging moved to 3–5 mm. The introduction of 3D volumetric acquisitions in the early 1990s (3D SE brain, later MPRAGE and 3D SPACE) enabled true isotropic sub-millimetre slices — a qualitative leap that replaced the concept of "slice thickness" for 3D protocols with "partition thickness" or "voxel dimension in z," producing images reformattable in any plane without resolution penalty.

2. Physical Foundations

2.1 Mechanism of Slice Selection

In a standard 2D MRI acquisition, a slice is defined by simultaneously applying a frequency-selective RF pulse and a linear magnetic field gradient in the slice-selection direction (z). The gradient creates a linear variation of resonance frequency along z:

ω(z) = ω₀ + γ × G_z × z

where γ = gyromagnetic ratio, G_z = slice-selection gradient amplitude, and z = position. The RF pulse excites only protons whose resonance frequency falls within the bandwidth of the pulse. The slice thickness (Δz) is therefore:

Δz = BW_RF / (γ × G_z)

where BW_RF is the bandwidth of the RF excitation pulse (in Hz).

Physical meaning: slice thickness is the quotient of the RF pulse bandwidth and the gradient strength. This is a direct analogy to the FOV in the frequency direction (Section 2.1 of the FOV page), where FOV_r = BW_receiver / (γ × G_r).

Optimisation implication: to produce a thinner slice at a given gradient amplitude, the RF pulse must have a narrower bandwidth — requiring a longer pulse duration (BW × duration ≈ constant for a given pulse shape). This increases the minimum TE for SE sequences (the slice-selection pulse contributes to the TE dead time). For very thin slices at 3T, the longer RF pulse also interacts with SAR management.

Alternative: increase the gradient amplitude G_z at constant RF bandwidth → thinner slice without longer pulse → shorter minimum TE. This is limited by gradient hardware (maximum G amplitude, typically 40–100 mT/m on clinical systems).

2.2 Mathematical Foundations

Voxel volume and SNR

Voxel_volume = Δx × Δy × Δz = (FOV_x/N_x) × (FOV_y/N_y) × Slice_thickness

SNR ∝ Voxel_volume × √(NSA) × B₀ × Coil_factor / √BW

Clinical meaning: slice thickness contributes linearly to SNR, exactly as do Δx and Δy. Halving the slice thickness halves the voxel volume → halves the SNR (all else equal). This is the fundamental constraint on minimum slice thickness: at some point, the signal from thinner slices is too low for diagnostic use even with surface coils, high field, and multiple averages.

SNR scaling with slice thickness:

• Slice 5 mm → Slice 3 mm: SNR reduces by 3/5 = 0.6 (40% SNR loss)
• Slice 5 mm → Slice 2 mm: SNR reduces by 2/5 = 0.4 (60% SNR loss)
• Slice 5 mm → Slice 1 mm: SNR reduces by 1/5 = 0.2 (80% SNR loss)

These are the most consequential numbers for protocol design: a 1 mm slice has only 20% the SNR of a 5 mm slice at identical in-plane parameters.

Coverage and number of slices

For a given anatomical coverage extent (C, in mm) and slice thickness (Δz) with inter-slice gap (g):

N_slices = C / (Δz + g)

Example: to cover 120 mm of lumbar spine:

• Δz=5 mm, g=0.5 mm: N_slices = 120/5.5 = 22 slices
• Δz=3 mm, g=0 mm: N_slices = 120/3 = 40 slices
• Δz=1.5 mm (3D isotropic): N_slices = 120/1.5 = 80 partitions (3D)

More slices = more acquisition time per pass (in 2D) or more data to collect (in 3D).

Cross-talk and slice gap

In 2D multi-slice acquisitions, the RF pulse profile is not a perfect rectangle — it has finite transition bands that excite adjacent slices slightly. If adjacent slices are acquired sequentially without a gap, the partially-excited adjacent tissue has not fully recovered → T1 saturation of adjacent slices → signal reduction and T1 contrast distortion.

The signal loss from cross-talk:

S_crosstalk ≈ S₀ × [1 − e^(−TR/T1)] × f(transition_band_overlap)

Mitigation strategies: 1. Inter-slice gap (10–20% of slice thickness): the most common solution. Creates coverage gaps where pathology may be missed. 2. Interleaved acquisition: even-numbered slices first, then odd-numbered slices in a second pass; adjacent slices are no longer sequentially acquired → cross-talk reduced. Standard for most clinical 2D multi-slice protocols. 3. 3D acquisition: no slice-direction RF selectivity; no cross-talk; no need for gaps; full coverage; isotropic voxels. The definitive solution, at the cost of longer acquisition time per TR.

Slice profile non-ideality

The actual slice profile is not a perfect rectangle — it has transition zones (sloping edges) where signal is partially excited. This means the effective slice thickness is slightly larger than the nominal value, and the signal at the slice boundaries is less than at the centre. For a typical sinc-modulated RF pulse:

• The 50% signal level corresponds approximately to the nominal slice thickness
• The 10% level extends typically 20–30% beyond the nominal edges
• For adjacent slices with zero gap, the overlap of transition zones produces the cross-talk discussed above

For 3D acquisitions: the partition thickness (equivalent of slice thickness) is defined by the Fourier encoding in z, exactly as the in-plane resolution is defined by the in-plane Fourier encoding. The partition profile is a sinc function (the Fourier transform of the rectangular k_z window) — giving true rectangular slice profiles with Gibbs ringing in z analogous to in-plane Gibbs ringing. Unlike 2D slices, 3D partitions have no gap and no cross-talk.

3. Units, Terminology and Vendor Nomenclature

Slice thickness is universally expressed in millimetres (mm).

Siemens Distance Factor: expressed as a percentage of the slice thickness (not an absolute mm value). A Distance Factor of 10% with 5 mm slices = 0.5 mm gap. This convention is a common source of confusion when comparing protocols across vendors.

Zero gap vs contiguous vs overlapping:

• Zero gap (g=0): no empty coverage space, but cross-talk risk if non-interleaved
• Contiguous (g=0 with interleaved acquisition): standard for brain, avoids cross-talk
• Overlapping (g<0): slices overlap in z; redundant coverage; used occasionally for targeted small-structure protocols to ensure no gap
• Positive gap (g>0): standard for body multi-slice to avoid cross-talk; creates true coverage gaps

4. Typical Value Ranges

4.1 Slice Thickness by Anatomical Region and Clinical Purpose

4.2 Slice Thickness Context by Field Strength

5. Parameter Interaction Ecosystem

5.1 Parameter Relationships Matrix

The fundamental 2D multi-slice constraint — the relationship between slice thickness, coverage, and TR:

In 2D multi-slice acquisitions, all slices within one "package" must be excited and read out within one TR period. The maximum number of slices per package is:

N_slices_max = TR / T_per_slice

where T_per_slice ≈ TE_max + dead_time (approximately 100–200 ms per slice for standard TSE). At TR=5000 ms and T_per_slice=120 ms: N_slices_max ≈ 41. If 60 slices are needed (thin slices over a long coverage), either TR must be increased or a second package must be added (with a time penalty).

This constraint is the primary reason that very thin 2D slice protocols for large-coverage examinations are often impractical: thin slices → more slices needed → TR must increase → scan time increases.

6. Effects on Image Appearance

6.1 Decreasing Slice Thickness

Through-plane spatial resolution improves — the most direct benefit. A structure that spans 2 mm in z is resolved in 4 voxels at 0.5 mm partition thickness, but only 1 voxel at 2 mm slice thickness.

Partial volume artefact decreases. When the voxel is thinner than the structure of interest, the signal in each voxel is more purely representative of that structure, rather than an average of multiple tissue types. This is clinically critical for: thin cartilage layers; cortical bone margins; small nerve structures; thin ligaments; focal cortical dysplasia in the brain.

SNR decreases linearly. The image appears noisier. At very thin slices without surface coils or high field, the noise level may exceed the tissue signal differences → image is non-diagnostic.

Slice profile effects: very thin 2D slices may show more pronounced profile non-ideality (the RF pulse profile is less rectangular for very narrow bandwidths) → signal variations at the slice edges → subtle z-direction signal inconsistencies between adjacent slices.

3D vs 2D appearance difference at equivalent thickness: 3D partitions at 1 mm isotropic have perfectly rectangular profiles (Fourier-encoded) and no inter-slice signal variation. 2D slices at 1 mm have non-ideal RF profiles, potential cross-talk, and may show slice-to-slice signal variability. For very thin slice applications, 3D is always preferred if feasible.

6.2 Increasing Slice Thickness

SNR increases linearly. The image appears smoother and less noisy.

Through-plane partial volume artefact worsens. Structures thinner than the slice appear as averaged signal with their surroundings. A 2 mm cartilage lesion at 5 mm slice thickness may be invisible — the signal change is diluted by the surrounding normal tissue within the 5 mm voxel.

Coverage efficiency improves: fewer slices needed for the same anatomical extent → shorter scan time in 2D, or the same scan time with higher quality (SNR/contrast).

Cross-section display quality: thick slices produce a block-like appearance of curved or fine anatomical structures (e.g., nerves appear as blocky segments rather than smooth tracts).

7. Effects on Acquisition Time

7.1 Direct Effects — 2D Acquisitions

In 2D multi-slice MRI, slice thickness has no direct effect on acquisition time for a fixed number of slices. The equation:

T_acq = TR × N_y × NSA / ETL

does not contain slice thickness. However, there is a critical indirect effect through the number of slices needed for coverage:

• Thinner slices → more slices for the same coverage → may require a longer TR to fit all slices within the TR period → acquisition time increases

Example: Spine sagittal T2 TSE, coverage = 24 cm:

• Δz = 4 mm, gap = 0.4 mm: N_slices = 240/4.4 = 54 slices, TR = 3500 ms → time = 3500 × 320/16 = 70 s → feasible
• Δz = 2 mm, gap = 0: N_slices = 240/2 = 120 slices. At T_per_slice = 120 ms: TR_required ≥ 120 × 120 = 14400 ms → scan time = 14400 × 320/16 = 288 s = 4.8 min → significantly longer

7.2 Direct Effects — 3D Acquisitions

In 3D acquisitions, slice thickness is directly equivalent to reducing N_z (the partition matrix). The acquisition time for 3D:

T_acq_3D = TR × N_y × N_z × NSA / ETL

Thinner partitions at fixed coverage → more N_z → proportionally longer scan time. Halving the 3D partition thickness doubles the acquisition time (if the coverage is held constant and no compensatory acceleration is applied).

7.3 2D vs 3D Time Efficiency for Different Slice Thicknesses

8. Effects on SNR and CNR

8.1 SNR Scaling with Slice Thickness

SNR ∝ Δz (linear relationship, all other parameters equal)

This table is the most important quantitative reference for slice thickness decision-making: going from 5 mm to 1 mm slices sacrifices 80% of the SNR. Recovering this loss requires: 25× NSA increase (4× SNR recovery per √NSA → 25× NSA for 5× SNR recovery) — clearly impractical. In practice, thin slice MRI requires higher field strength, dedicated surface coils, and/or DLR.

8.2 CNR and Slice Thickness

CNR depends on SNR and on whether the tissue contrast benefit of thin slices (reduced partial volume) outweighs the SNR reduction. For a small lesion (smaller than the slice thickness):

• Thick slice: lesion signal is averaged with surrounding tissue → low contrast from partial volume; but SNR is high
• Thin slice: lesion signal is less diluted → potentially higher intrinsic CNR; but SNR is low → noise may cancel the contrast gain

The optimal slice thickness for small lesion detection is therefore not "as thin as possible" but rather "thin enough to avoid significant partial volume, while maintaining adequate SNR." This optimal point depends on lesion size, tissue contrast, and available SNR — and is different for every clinical context.

8.3 Field-Strength Dependency

The higher SNR at 3T directly enables thinner slices at equivalent image quality:

• 3T vs 1.5T: approximately 1.7–2× SNR gain
• Applied to slice thickness: the SNR saved by 3T can support slices approximately 0.5–0.6× as thick as at 1.5T with equivalent SNR
• Example: 4 mm at 1.5T ≈ 2–2.5 mm at 3T for equivalent SNR

This is the primary clinical benefit of 3T for MSK, neuro, and prostate MRI: the ability to halve slice thickness (doubling through-plane resolution) at maintained SNR.

9. Artefacts Associated with Slice Thickness

10. Behaviour Across Sequence Families

Spin Echo and Turbo Spin Echo

Standard 2D SE/TSE: slice selection by RF pulse; all the cross-talk, profile, and gap considerations described above apply. The minimum TE is slightly longer for very thin slices (longer RF pulse for narrow bandwidth). Interleaved acquisition is standard; contiguous (zero-gap) 2D TSE is acceptable for the brain and spinal cord where the interleaved mode prevents cross-talk. For very thin requirements (< 2 mm), 3D TSE (SPACE/CUBE/VISTA) is preferred.

Gradient Echo (GRE/FLASH/SPGR)

GRE uses shorter RF pulses than SE (lower flip angles → more time-efficient RF deposition) → minimum TE is shorter → thin slices are more easily achievable. Spoiled GRE (VIBE, LAVA) for 3D abdominal acquisitions routinely achieves 1.5–2 mm partitions at 3T within a breath-hold. Cross-talk is less severe in GRE than in SE (lower flip angles → less saturation per pulse).

Inversion Recovery (STIR, FLAIR, MPRAGE)

For 2D STIR and FLAIR, the same 2D constraints apply. 3D FLAIR (standard for ARIA monitoring, AD protocol) uses 3D encoding → isotropic 1–1.5 mm partitions → no cross-talk → complete brain coverage with no inter-slice gaps. MPRAGE: 3D encoding; 1 mm isotropic partition is the ADNI standard (see AD child page).

EPI (DWI, DSC, fMRI)

EPI slice thickness is 2D-selected (not Fourier-encoded in z, unlike 3D). Typical DWI: 5 mm slices with 0–0.5 mm gap (brain); 4–5 mm (body). At 3T, the EPI geometric distortion does not depend on slice thickness directly — it depends on in-plane phase-encoding parameters. Thin EPI slices (< 3 mm) at standard in-plane FOV suffer from SNR loss without distortion benefit. Multi-band (SMS) EPI enables simultaneous thin slices (2 mm) across the whole brain in fMRI and DTI without the per-slice SNR penalty of sequential thin-slice acquisition.

Dixon

Underlying sequence family determines slice behaviour. Dixon GRE at thin slices: same as GRE above. The specific TE requirements of Dixon (IP/OP values) may constrain minimum TE, which interacts slightly with RF pulse duration for very thin slices.

DWI (dedicated)

Body DWI (liver, kidney, prostate) uses 5–7 mm slices (2D EPI) to maintain adequate SNR despite the intrinsically low DWI signal at b=800–1000. Reducing to 3 mm body DWI slices severely compromises SNR and is generally not acceptable without 3T + surface coil + NSA compensation. Brain DWI for acute stroke uses 5 mm slices (same rationale). Research thin-slice DWI uses multi-band acquisition (see fMRI above).

DSC Perfusion

DSC EPI: typically 5 mm slices, 20–30 slices per brain volume at temporal resolution ≈ 1.5 seconds. Reducing slice thickness would require fewer slices per volume (worse coverage) or faster readout (higher parallel imaging, lower SNR). The temporal resolution constraint limits slice thickness options.

DCE (Dynamic Contrast Enhanced)

3D GRE-based DCE: isotropic 1–1.5 mm for breast; 3–4 mm for liver at 1.5T; 2–3 mm for liver at 3T. The temporal resolution constraint (arterial phase ≤ 20–25 s) is the primary limiting factor — reducing slice thickness at fixed coverage increases N_z → longer scan per phase → temporal resolution degrades.

ASL

3D pCASL: typically 3–5 mm partitions (GRASE or spiral 3D readout). The intrinsically low ASL SNR (signal ≈ 1% of total) forces relatively thick slices. Reducing to 2 mm ASL partitions requires many NSA (typically 40–80 averages per measurement).

bSSFP

Cardiac bSSFP (TrueFISP/FIESTA): 6–8 mm 2D slices for cardiac cine; 1–1.5 mm for 3D coronary MRA. For 3D coronary MRA, the partition thickness is the primary determinant of coronary vessel detection: vessels < 2 × partition thickness may not be detected.

Spectroscopy (SVS/MRSI)

In SVS, the "slice thickness" is the voxel dimension in z — equivalent to any other voxel dimension. For typical clinical SVS: 15–20 mm voxels in all three dimensions. Reducing to 10 mm requires 2× NSA compensation. In MRSI, the "slice thickness" of the MRSI slab determines the integration volume for each spectral element.

11. Field Strength Behaviour

Key 3T advantage for slice thickness: the ~2× SNR at 3T can be invested in halving slice thickness. 2 mm slices at 3T ≈ SNR equivalent to 4 mm slices at 1.5T. This is the reason that 3 mm slice protocols are clinically standard at 3T for applications where 4–5 mm was standard at 1.5T.

0.55T constraint: the limited SNR at 0.55T severely constrains thin slice application. 5–7 mm slices are the practical minimum for routine imaging without DLR. With DLR, 3–4 mm becomes feasible for brain; body typically remains ≥ 5 mm.

12. Vendor-Specific Implementation

Siemens

The "Distance Factor" (expressed as % of slice thickness) controls the gap. A Distance Factor of 0% = contiguous (zero gap); the default for most brain protocols is 10–20%. The Siemens system automatically interleaves multi-slice acquisitions by default; the technologist must verify that the interleave is applied (visible in the sequence detail). For 3D sequences, "Slice oversampling" (%) adds extra partitions at the slab edges to prevent wrap-around in z — default is typically 15–25%.

Siemens offers "Averaged slice" as an option in some protocols — adjacent thin slices can be retrospectively averaged to produce thicker effective slices with higher SNR. This is useful for post-acquisition review but creates thicker apparent resolution in clinical reading.

GE

Gap is expressed in mm (absolute), not as a percentage. The "Slice gap" field accepts negative values for overlapping slices. GE protocols often use 0 mm gap for brain (interleaved acquisition standard). For body MRI, a 1–2 mm gap is typical. The "3D phase" parameter in GE controls the number of z-partitions; slice thickness = slab thickness / N_slices (3D).

Philips

Gap is expressed in mm. Philips uses "Oversampling" in 3D to avoid z-aliasing. The Philips system displays the actual voxel dimensions in the three directions directly in the protocol card, including the reconstructed partition thickness for 3D acquisitions — very useful for rapid resolution verification.

Canon

Standard absolute gap convention. Canon's implementation of multi-slice interleaving is automatic for most sequence types. Canon documentation explicitly states the inter-slice gap as a percentage in their protocol nomenclature, but the UI accepts absolute mm values.

United Imaging

Standard conventions consistent with Siemens. UIH scanners include an automated check for cross-talk risk that warns the technologist if the slice gap is insufficient for the selected TR and flip angle combination — a useful safety feature.

Hidden coupling — all vendors: when the requested slice number × slice period exceeds the TR, the scanner automatically increases TR to accommodate all slices. This TR increase may change T1 contrast and increase scan time without explicit technologist awareness. Always verify TR after entering the slice thickness and number.

13. Practical Optimisation Strategies

13.1 Clinical Optimisation Recipes

14. Parameter Extremes

14.1 Extremely Thin Slices (< 1 mm, 2D)

At sub-millimetre 2D slice thickness, three fundamental problems converge:

• SNR is critically low: < 20% of the 5 mm reference SNR. Without a closely coupled surface coil and high field (3T), the image is typically too noisy for diagnosis.
• RF pulse duration is long: a 0.5 mm slice at typical gradient amplitudes requires a narrow-bandwidth RF pulse with long duration (> 5 ms) → long minimum TE → significant T2 weighting even for "T1" sequences.
• Cross-talk is severe: the very thin slice has a wide relative transition zone → adjacent slices overlap heavily unless the gap is large (> 100% of slice thickness) → large coverage gaps.

The solution for sub-millimetre resolution is invariably 3D acquisition — which encodes z with Fourier gradients rather than RF selection, avoiding all three problems above.

14.2 Extremely Thick Slices (> 10 mm)

At very thick slices:

• SNR is high: excellent for noise-limited applications.
• Partial volume is severe: any structure thinner than the slice is averaged with adjacent tissue.
• Clinical applications: thick-slab MRCP (40–80 mm single thick slab in 3–5 seconds breath-hold); MIP projections for MRA; dynamic liver surveys where temporal resolution > spatial resolution.
• Thick-slab MRCP is the most important clinical extreme: a single 40–80 mm coronal thick slab covers the biliary tree and pancreatic duct as a projection — analogous to the radiographic ERCP image. Each of the multiple angulations (5–7 projections in different planes) is acquired in < 5 seconds. This is not a slice but a projection MIP — a qualitatively different use of slice thickness than routine imaging.

15. Common Optimisation Errors

16. MRI Technologist Pearls

The 2× rule for structure visibility: to reliably detect a structure or lesion in the through-plane direction, the slice thickness must be ≤ half the structure's z-extent. For a 4 mm disc herniation extending into the canal: slice ≤ 2 mm (or 3D isotropic ≤ 2 mm) for reliable detection. Structures at the slice thickness limit appear inconsistently — present on some examinations, absent on others, depending on alignment.

Check the TR after changing slice count: the most common hidden time and contrast change in 2D MRI protocols is the TR auto-increase triggered by adding slices. After changing slice thickness or slice count, immediately verify the displayed TR — if it has changed, assess whether the T1 contrast is still appropriate, and whether the scan time is still acceptable.

3D for thin slices, always: if the required slice thickness is < 2–3 mm, the 3D acquisition path is almost always superior to 2D for clinical MRI. 3D provides: no cross-talk; no gaps; rectangular profiles; isotropic MPR; and better SNR efficiency per acquired voxel (√N_z advantage). The only exception is when motion makes 3D impractical and the short 2D acquisition time per slice is preferred.

Oversampling in 3D is cheap insurance: always leave 3D slice oversampling at ≥ 15%. The extra partitions at the slab edges prevent wrap-around artefacts that can simulate or obscure pathology at the brain vertex or at the inferior organ pole. The time cost of 15% oversampling is minimal.

Thick-slab MRCP projections before 3D navigator: always acquire the 2D thick-slab MRCP projections before the 3D navigator-triggered MRCP. The thick-slab projections take 20–30 seconds and provide the clinical overview image equivalent. If the navigator fails or the patient cannot cooperate, the thick-slab projections provide the essential diagnostic information even if the 3D dataset is incomplete.

SNR check before thin-slice protocols: before starting any protocol with slice < 2 mm in 2D, or < 0.8 mm in 3D, run a rapid test acquisition (1 slice, 1 NSA, short TR) and review SNR. If the image is already marginal at the survey level, reduce the matrix or increase NSA before committing to the full protocol.

17. Real Clinical Examples

Example 1: Missed Focal Cartilage Defect — Slice Thickness Error

Clinical scenario: post-meniscectomy knee; surgeon queries focal chondral lesion at the medial tibial plateau. MRI performed at an outside institution with 5 mm coronal T2 slices.

Problem: the tibial cartilage is 2–2.5 mm thick. At 5 mm slice thickness, even a full-thickness (grade 4) cartilage defect spanning the full 2.5 mm occupies only 50% of the voxel z-extent → its signal is averaged with adjacent normal cartilage and subchondral bone → the apparent defect signal is intermediate → read as "no focal cartilage defect."

Re-examination: 3T; 3D SPACE T2; 0.7 mm isotropic; matrix 320×256; TR=1500/TE_eff=35/ETL=70; R=2.

• The 2.5 mm cartilage is resolved in 3.6 voxels (2.5 / 0.7)
• A 1 mm focal defect occupies 1.4 voxels → clearly visible as a discrete z-direction signal void

Diagnosis: full-thickness focal cartilage defect at the central medial tibial plateau — missed on the original study.

Trade-off: 3D acquisition adds 6–8 min to the protocol; requires patient cooperation.

Example 2: Cross-Talk Artefact — T1 Contrast Error

Clinical scenario: post-gadolinium T1 TSE axial brain at 5 mm slices, zero gap, sequential (not interleaved) acquisition.

Problem: in sequential acquisition (slice 1 → slice 2 → slice 3...), slice 2 is acquired while the transition zone of slice 1 RF pulse is still partially saturating slice 2 tissue. At TR=600 ms, T1 ≈ 1000 ms (white matter): recovery is 45% between slices. The adjacent slice sees tissue with only 45% of the full longitudinal magnetisation → apparent T1 shortening of white matter → white matter appears relatively brighter → less grey-white contrast → some enhancing lesions appear less bright.

Correction: interleaved mode activated (even slices first: 2,4,6...; then odd: 1,3,5...); adjacent slices are now separated by the full TR → complete T1 recovery between adjacent slices.

Result: correct T1 contrast restored; enhancing lesions appear brighter against appropriate grey-white background.

Lesson: never use zero-gap 2D slices without interleaved acquisition. The scanner default is interleaved on most modern platforms; verify it has not been manually disabled.

Example 3: Prostate T2 PI-RADS — 3 mm Axial Slice Requirement

Clinical scenario: prostate MRI for suspicious PSA rise. PI-RADS v2.1 specifies T2 axial slice thickness ≤ 3 mm.

Outside MRI: T2 axial with 4 mm slices, 0 mm gap. A 5 mm lesion at the apex is visible but poorly characterised: with 4 mm slices, the 5 mm lesion spans 1.25 slices (partial volume with adjacent apex fibromuscular stroma on one slice).

Compliant protocol: 3 mm slices, 0 mm gap, contiguous, interleaved. Same lesion now spans 1.67 slices → better characterisation of its internal architecture (transition zone pattern vs peripheral zone pattern) and extraprostatic extension at the apex.

Trade-off: at 3T, going from 4 mm to 3 mm → SNR reduces by 25%. With TR=4500/TE=90 at 3T with R=2: scan time = 5 min — acceptable. At 1.5T with 3 mm slices, SNR may be marginal; PI-RADS v2.1 ≤ 3 mm is achievable but at the edge of diagnostic quality.

Example 4: Brachial Plexus — 3D vs 2D at 3 mm

Clinical scenario: post-traumatic brachial plexopathy; suspected preganglionic injury; 3D CISS for root sleeve assessment.

Option A: 2D T2 STIR coronal, 3 mm, 0 mm gap, interleaved. Coverage 25 cm → 83 slices required. At TR=4000 ms, T_per_slice = 120 ms: TR must be ≥ 83 × 120 ms = 9960 ms → actual TR used = 10000 ms. STIR TI recalculates to maintain fat null at this TR → all correct. Scan time = 10000 × 256/20 (ETL) = 128000 ms ≈ 2.1 min.

Option B: 3D CISS, 0.8 mm isotropic, coverage 25 cm, N_z = 310 partitions. True rectangular profile; no cross-talk; no gap; full MPR; root sleeves resolved individually. Scan time ≈ 6–8 min.

Clinical decision: 2D STIR at 3 mm for the overall nerve T2 assessment (fast, reliable fat suppression at off-isocentre); 3D CISS at 0.8 mm specifically for root sleeve and pseudomeningocele assessment (focused, higher resolution). Combining both provides the best clinical information.

Lesson: the choice is not "2D or 3D" but "which task requires which approach." 2D STIR (3 mm) provides the pathological signal sensitivity (oedema, T2 change); 3D CISS (0.8 mm) provides the anatomical detail (root sleeves, pseudomeningoceles). Both are needed.

18. Visual Educational Material

18.1 Slice Thickness and Partial Volume Decision Flow

TARGET STRUCTURE z-EXTENT = D mm

Is Δz ≤ D/2?
│
├── YES → Structure is resolved in ≥ 2 voxels → partial volume is ACCEPTABLE
│
└── NO → Structure is unresolved → partial volume is DIAGNOSTIC RISK
         │
         ├── Can I reduce Δz?
         │   ├── YES (SNR permits): reduce slice thickness
         │   └── NO (SNR-limited): accept limitation; document in report;
         │       consider 3D protocol; supplement with orthogonal plane
         │
         └── Can I switch to 3D?
             ├── YES: 3D isotropic at Δz ≤ D/2
             └── NO: document limitation; plan repeat with dedicated small-FOV protocol

18.2 2D Slice Coverage and TR Constraint

COVERAGE REQUIRED: C mm
SLICE THICKNESS: Δz mm
GAP: g mm
NUMBER OF SLICES: N = C / (Δz + g)

MINIMUM TR REQUIRED: TR_min = N × T_per_slice

T_per_slice ≈ TE_max + recovery_dead_time ≈ 100–200 ms (sequence-dependent)

EXAMPLE: C=240mm, Δz=3mm, g=0mm, T_per_slice=120ms:
  N = 240/3 = 80 slices
  TR_min = 80 × 120 = 9600 ms

CONSEQUENCE: The scanner will auto-set TR ≥ 9600 ms
  → Long TR → longer scan time
  → Check if this is acceptable before proceeding

SOLUTION OPTIONS:
  1. Increase Δz → reduce N → reduce TR_min → shorter scan
  2. Accept long TR (T2-weighted; long scan)
  3. Switch to 3D (no TR-slice constraint)
  4. Split into two packages (two separate scans)

18.3 SNR vs Slice Thickness Chart (Relative to 5 mm Reference)

Slice:  10mm   7mm   5mm   4mm   3mm   2mm   1.5mm  1mm
SNR×:   2.0    1.4   1.0   0.8   0.6   0.4   0.3    0.2

To recover SNR to 1.0× at each thin slice by NSA alone:
  4mm: NSA = 1.6× (minor)
  3mm: NSA = 2.8× (moderate; 3× longer)
  2mm: NSA = 6.3× (severe; 6× longer)
  1mm: NSA = 25×  (impractical)

→ Thin slices require field strength increase, coil improvement, or DLR
   — not NSA compensation

19. Evidence Gaps and Ongoing Debate

Optimal slice thickness for specific diagnostic tasks: while PI-RADS v2.1 specifies ≤ 3 mm for prostate T2 axial [1], few other guidelines provide explicit slice thickness specifications based on formal validation studies. The ESHNR HNSCC staging recommendations [2] provide some guidance for head and neck protocols, but the specific slice thickness thresholds for each anatomical target are based on expert consensus rather than prospective accuracy studies. For most clinical applications, the "minimum slice for the available SNR" principle is applied empirically.

2D vs 3D diagnostic equivalence at matched voxel size: several studies have compared 3D TSE (SPACE/CUBE) with 2D TSE for specific applications (knee, spine, brain) at matched voxel size, generally showing equivalent or superior performance for 3D. However, the comparison is confounded by the different SNR efficiency of 3D (√N_z advantage) and the different T2 blurring characteristics of long VFA ETLs. No systematic cross-anatomy, cross-vendor comparative study has determined the precise threshold at which the 3D advantage over matched-voxel-2D becomes clinically significant.

DLR impact on minimum viable slice thickness: with deep learning reconstruction, images acquired at 2 mm partition thickness can approach the diagnostic quality of 1.5 mm partitions without DLR (for some applications). The minimum clinically viable 3D partition thickness with DLR has been characterised for some applications (MPRAGE, body DCE) but not systematically validated for all protocols. The risk of over-smoothing small structures when DLR is applied at aggressively thin slices requires prospective assessment.

Slice oversampling in 3D — optimal percentage: the standard recommendation of 15–20% 3D slice oversampling is based on gradient engineering considerations (the slab excitation profile transition zone) rather than formal image quality studies. Whether a specific oversampling percentage is optimal for every 3D slab size and sequence type, or whether institution-specific optimisation is required, has not been addressed in the literature.

20. Miscellaneous and Future Directions

Historical milestone: the transition from 10 mm to 5 mm clinical brain MRI slices between 1984 and 1990 approximately doubled the detection rate of small cortical lesions in multiple sclerosis — one of the earliest documented clinical impacts of slice thickness optimisation. The introduction of 3 mm slices for brain MRI in the early 1990s (enabled by high-performance gradient systems and surface coils) further improved lesion detection.

Super-resolution reconstruction across slice direction: multiple thin-slice 2D acquisitions with slightly different slice positions can be combined using super-resolution algorithms to produce an effectively thinner composite partition, analogous to the in-plane super-resolution approaches described in the Matrix child page. This is standard in fetal MRI (combining three orthogonal 2D stacks) and is being explored for neonatal brain, cardiac, and small-animal imaging.

MR fingerprinting and quantitative parameter maps independent of slice thickness: MR fingerprinting (MRF) simultaneously acquires T1 and T2 maps from a single acquisition. Because the quantitative maps are in physical units (ms), they are theoretically less sensitive to slice thickness effects than signal-intensity-based qualitative imaging. A T2 measurement in a 3 mm slice and a 1.5 mm slice should agree (accounting for partial volume) — potentially reducing the slice-thickness dependence of quantitative diagnoses. This is an active area of validation.

AI-planned slice prescriptions: automated anatomy recognition tools increasingly propose slice thickness recommendations based on body landmark detection. The clinical safety of AI-proposed thin-slice protocols (ensuring adequate SNR for the patient's body habitus) requires prospective validation before full automation.

End of document — Slice Thickness — MRIninja v1.0 — May 2026

Parent page: MRI Parameters — Overview and Classification (9501)