Fold-over Direction (Phase-Encoding Direction)

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

1. Introduction and General Purpose

Fold-over direction is the operator's choice of which of the two in-plane anatomical axes is used for phase encoding, with the other axis carrying frequency encoding (the continuous readout). The name "fold-over direction" — the term used on the Philips console reviewed for this page — describes its most historically well-known consequence: wrap-around (aliasing) artefact, when present, occurs specifically along this axis. But the choice governs three separate, clinically consequential behaviours, each developed fully below: (1) where wrap-around artefact will appear if it occurs at all; (2) along which axis motion-related ghosting and blurring propagate, since phase encoding — unlike frequency encoding — is built up incrementally across many TRs and is therefore far more exposed to inter-repetition physiological motion; and (3) in some circumstances, total acquisition time, since the number of phase-encoding steps required depends on the anatomical extent along whichever axis is chosen for phase encoding.

This is one parameter with many names — a pattern that recurs across MRI parameters and is worth stating explicitly here as a general principle for this platform: fold-over direction (Philips), phase encoding direction or "Phase enc. dir." (Siemens, confirmed directly from real protocol exports reviewed for this page — Section 3), phase direction or PE direction (GE and general usage), and simply PE axis or phase-encode direction in the technical/radiological literature, all refer to the same underlying acquisition decision. Readers moving between vendor consoles, published protocols, and the technical literature should treat these as synonyms, not as different parameters.

A necessary distinction from adjacent, easily-confused console fields, in the same spirit as this platform's treatment of "slice scan order" versus "profile order" (see the companion Slice Scan Order page): fold-over direction (which axis carries phase encoding) is not the same field as fold-over suppression (phase oversampling — a mitigation technique applied once a direction has already been chosen; see the companion Phase Oversampling page) nor the same as fat shift direction (a related, separately-set parameter governing the spatial direction of chemical-shift misregistration, addressed on its own dedicated page in this cluster's future roadmap).

2. Physical Foundations

2.1 Why Frequency and Phase Encoding Behave So Differently

Frequency encoding is applied as a continuous gradient during a single, brief readout window (typically a few to tens of milliseconds); every point along the frequency-encoded axis is therefore sampled essentially simultaneously, within one short time interval, making this axis intrinsically resistant to inter-repetition physiological motion. Phase encoding, by contrast, is built up one discrete step at a time, once per TR, across the entire acquisition — for a typical 2D sequence, this means the phase-encoded dimension of k-space is assembled over an interval that can span many seconds to several minutes. Any physiological motion — respiration, cardiac pulsation, CSF pulsation, swallowing, bulk patient movement — occurring during that extended interval introduces genuine inconsistency between different phase-encoding steps, and this inconsistency manifests specifically as blurring (for random, aperiodic motion) or discrete ghosting (for periodic, repeating motion) along the phase-encoding axis [2,6]. This asymmetry between the two encoding axes is the foundational physical fact from which essentially everything else on this page follows.

2.2 Aliasing (Wrap-Around) Along the Phase Axis

If the true anatomical extent along the phase-encoding axis exceeds the prescribed field of view in that direction, the resulting spatial-frequency under-sampling causes tissue outside the nominal FOV to be mismapped ("wrapped") into the opposite edge of the image. In principle, the same Nyquist-sampling logic applies to the frequency-encoded axis as well; in practice, however, virtually all modern scanners oversample the frequency (readout) direction automatically and at no time cost, because digitising additional points along an already-continuous readout adds negligible overhead — whereas oversampling the phase direction to the same degree would directly and proportionally increase scan time (see the companion Phase Oversampling / Fold-over Suppression page). This asymmetric, essentially free frequency-direction protection is why wrap-around artefact is, by long-standing scanner-design convention and everyday clinical experience, functionally a phase-direction-only concern.

3. Units, Terminology and Vendor Nomenclature

This section is deliberately exhaustive, consistent with the principle established in Section 1 that a single MRI parameter frequently carries several different names across vendors and literature — a pattern this platform will continue to document thoroughly for every parameter, not only this one.

Vendor terminology confirmed from an actual Philips console (Advanced Parameters, Geometry tab): the "Fold-over direction" field sits within the same Stack block used for slice count, slice gap, and slice scan order (see the companion Number of Slices, Slice Gap, and Slice Scan Order pages), immediately adjacent to the separately-set fat shift direction field — two console fields that are easy to conflate but govern different physical phenomena (Section 1).

Vendor default divergence worth knowing: for axial (and axial-oblique) brain imaging, Siemens systems default to an anterior-posterior (A>>P) phase-encoding convention while GE systems default to posterior-anterior (P-A); the practical image-quality consequence — most apparent on EPI — is that the two conventions compress and stretch the frontal versus occipital poles in opposite directions, a genuine, documented cross-vendor difference rather than a purely arbitrary labelling quirk [5].

4. Typical Value Ranges (Typical Direction Choices by Application)

5. Parameter Interaction Ecosystem

5.1 Parameter Relationships Matrix

6. Effects on Image Appearance

Changing fold-over direction does not, by itself, alter image contrast, resolution, or SNR (Section 8). Its appearance consequences are entirely about where, not whether, certain effects occur: which axis carries wrap-around artefact if the anatomy exceeds the prescribed FOV in that direction; which axis carries motion-related ghosting/blurring; and, for EPI-family sequences specifically, which axis carries geometric distortion from B0 inhomogeneity (Section 10). A structure that appears artefact-free in one phase-encoding direction can show substantial ghosting, wrap, or distortion when the direction is swapped — the underlying tissue is unchanged, only the spatial location and axis of the artefact.

7. Effects on Acquisition Time

Fold-over direction has no inherent time cost by itself — swapping which axis carries phase encoding does not change total acquisition time when the number of phase-encoding steps and all other parameters are held constant. Its genuine, direct time relevance arises specifically when combined with a rectangular (asymmetric) field of view: because scan time scales with the number of phase-encoding steps (T_acq ∝ N_y, per the companion Number of Slices page's timing framework), choosing the phase-encoding axis to align with the anatomically shorter dimension allows a smaller phase FOV — and therefore fewer phase-encoding steps — to be used for the same in-plane resolution, directly shortening scan time [6]. Choosing the phase axis along the anatomically longer dimension forfeits this opportunity and, in 3D acquisitions specifically, can materially lengthen the scan, since one of the two phase-encoded dimensions in 3D governs slice-select/partition encoding directly [6].

8. Effects on SNR and CNR

Fold-over direction has no direct effect on the SNR of the acquired image — voxel volume, bandwidth, field strength, coil, and NSA are unaffected by which axis carries phase encoding. Its CNR-adjacent consequence is indirect but clinically important via parallel imaging: because acceleration is applied along the phase-encoding axis (Section 5.1), the g-factor-related SNR penalty of parallel imaging is imposed on whichever anatomical axis is chosen for phase encoding — meaning the fold-over direction decision partially determines where any parallel-imaging-related SNR loss is spatially concentrated [3].

9. Artefacts Associated with Fold-over Direction

10. Behaviour Across Sequence Families

Spin Echo (SE) / Turbo Spin Echo (TSE): standard motion-ghosting and wrap-around behaviour exactly as developed in Sections 2 and 9; distortion from B0 inhomogeneity is comparatively minor because the frequency-direction bandwidth used in conventional (non-EPI) sequences is typically far higher than EPI's phase-direction bandwidth, keeping non-EPI distortion negligible by comparison [7].

Gradient Echo (GRE/FLASH): standard behaviour as above; susceptibility-related signal dropout (a related but distinct phenomenon from geometric distortion) is governed more by TE and voxel geometry than by fold-over direction specifically.

EPI (DWI, fMRI, DSC) — the sequence family in which this parameter is most consequential: single-shot EPI has markedly lower effective bandwidth along the phase-encoding axis than along the frequency axis, making the phase-encoding axis the dominant direction of B0-inhomogeneity-driven geometric distortion, with reported displacements reaching the centimetre range in regions of substantial susceptibility gradient (frontal and temporal lobes, skull base, orbitofrontal cortex) [1,7]. This is the foundational reason EPI phase-encoding direction is treated as a first-order acquisition decision rather than a workflow convenience, and the reason reversed-phase-encoding-polarity acquisitions (opposite phase direction pairs) are widely used specifically to enable post-hoc geometric distortion correction [1].

Time-of-flight and other flow-sensitive sequences: fold-over/phase-encoding direction is a separate decision from the spatial acquisition direction discussed on the companion Slice Scan Order page (which concerns flow-related saturation, not phase-encoding wrap/ghosting); the two should not be conflated even though both involve a "direction" choice on the same anatomy.

3D sequences generally: two phase-encoded dimensions exist rather than one (standard in-plane phase encoding plus partition/slab encoding), and the choice of which physical axis serves which encoding role has a correspondingly larger time impact, since the more anatomically extensive of the two phase-encoded dimensions should generally not be assigned to the dimension requiring the most encoding steps (Section 7) [6].

11. Field Strength Behaviour

Fold-over direction itself is not intrinsically field-strength-dependent, but its most consequential downstream effect — EPI geometric distortion — is strongly field-strength-dependent: B0 inhomogeneity-related distortion scales approximately linearly with field strength, meaning the same phase-encoding-direction choice produces materially more severe distortion at 3T than at 1.5T, and still more at 7T [7]. This makes deliberate, informed fold-over/phase-direction selection progressively more clinically important as field strength increases for any EPI-family sequence.

12. Vendor-Specific Implementation

Siemens: "Phase enc. dir." field, set via directional-arrow notation (A>>P, P>>A, R>>L, L>>R, H>>F, F>>H); default axial convention is A>>P [4,5]. Automated slice/scan-plane alignment tools (e.g. AutoAlign-type features) can occasionally introduce an unintended in-plane rotation that mixes the intended frequency/phase axis assignment away from the primary gradient axes, which is a recognised, specifically documented pitfall for EPI ghosting — operators are advised to verify the phase-encoding-direction field remains at its intended value after any automated planning step.

GE: phase direction field, with a default axial convention of P-A, differing from Siemens' A>>P default [5]; the frontal-versus-occipital-lobe compression/stretch pattern on EPI differs correspondingly between the two vendors for nominally equivalent axial protocols.

Philips: "Fold-over direction" field within the Stack block (confirmed directly from console review, Section 3), set independently from — but adjacent to — the "fat shift direction" field, requiring the explicit disambiguation established in Section 1.

Canon: equivalent phase-encode-direction control on the protocol card, following the same conceptual convention as the other vendors.

Cross-vendor protocol translation note: because Siemens and GE differ in their default axial phase-encoding convention (Section 3), a protocol translated between these two platforms without deliberately re-verifying the phase direction can produce a subtly different distortion/ghosting pattern on EPI even when every other parameter is matched — a translation risk worth checking explicitly, in the same spirit as the percentage-versus-absolute-mm gap conversion trap described on the companion Slice Gap page.

13. Practical Optimisation Strategies

13.1 Clinical Optimisation Recipes

Choose the phase-encoding direction so the line connecting the dominant motion source and the structure of clinical interest is not aligned with that axis: since ghosting propagates along the phase axis, the classical, widely-taught rule is to set phase encoding perpendicular to the vector between a periodic motion source (eyes, aorta, bowel) and the diagnostically critical structure — the textbook example being left-right phase encoding for axial brain imaging, specifically to keep eye-movement ghosting from spilling into brain parenchyma [6].

Exploit rectangular FOV and the shorter anatomical axis for scan-time savings wherever resolution requirements allow: aligning phase encoding with the anatomically shorter dimension directly reduces the number of phase-encoding steps needed (Section 7) [6].

For any EPI-family sequence, treat phase-direction selection as a distortion-management decision, not a default: consider which anatomical region's distortion is most clinically tolerable, and — where available — acquire reversed-phase-encoding-polarity pairs to enable post-hoc geometric distortion correction, particularly at 3T and above [1,7].

When empirical troubleshooting is warranted, simply swapping phase and frequency encoding directions is a legitimate, pragmatic strategy for reducing or relocating severe artefact away from a critical structure, even without a full first-principles explanation for why a particular swap works best in an individual case [6].

14. Parameter Extremes

14.1 Phase Direction Chosen Purely for Time-Saving (Rectangular FOV)

In elongated anatomy (extremity joints, spine segments) where the artefact-avoidance considerations of Section 13.1 are less pressing, phase direction can be selected primarily to minimise the number of phase-encoding steps via rectangular FOV, prioritising scan-time efficiency over artefact-location optimisation.

14.2 Reversed Phase-Encoding-Polarity Pairs (EPI Distortion Correction)

At the opposite extreme, some EPI-family research and increasingly clinical protocols deliberately acquire the same anatomy twice with opposite phase-encoding polarity specifically to enable computational geometric-distortion correction, accepting the time cost of a duplicate acquisition in exchange for substantially improved spatial accuracy in regions of high susceptibility gradient [1].

15. Common Optimisation Errors

16. MRI Technologist Pearls

Before scanning, ask "what is the dominant source of periodic motion here, and where does the diagnostically critical structure sit relative to it" — then set phase direction perpendicular to that line (Section 13.1).

For extremity and elongated-anatomy protocols, actively consider whether aligning phase encoding with the shorter dimension would meaningfully shorten the scan via rectangular FOV, rather than defaulting to a fixed convention regardless of anatomy shape.

For any EPI-family sequence, especially at 3T and above, treat phase direction as a genuine image-quality decision — know which anatomical region will bear the brunt of distortion for a given direction choice, and discuss with the radiologist when the region of greatest diagnostic interest coincides with the region of greatest expected distortion.

After using an automated planning/alignment tool, glance at the phase-encoding-direction field before acquiring, particularly for EPI — an unintended small rotation is a recognised, easily-missed source of avoidable ghosting (Section 12).

17. Real Clinical Examples

Example 1: Axial Brain — Left-Right Phase Encoding to Protect Parenchyma from Eye Ghosting

Clinical scenario: routine axial T2/FLAIR brain imaging.

Protocol logic: phase encoding set left-right rather than anterior-posterior; eye movement during the (comparatively long) phase-encoding acquisition interval produces ghosting that propagates left-right, confined largely to a band lateral to the orbits, rather than propagating anteriorly-posteriorly directly through frontal and temporal brain parenchyma [6].

Lesson: this is the single most widely taught example of deliberate fold-over-direction selection in clinical MRI, and a direct application of the perpendicular-to-motion-source principle developed in Section 13.1.

Example 2: Axial Brain EPI — Vendor Default Divergence and Distortion Direction

Clinical scenario: whole-brain DWI or BOLD fMRI acquired on two different vendor platforms as part of a multi-site study or equipment transition.

Problem: the Siemens platform's default A>>P phase-encoding convention and the GE platform's default P-A convention compress and stretch the frontal versus occipital poles in opposite directions on EPI, producing a systematic, vendor-dependent difference in apparent brain shape/coverage even when every other parameter is nominally matched [5].

Resolution: explicit, deliberate phase-direction verification and, where cross-site comparability matters, harmonisation of the phase-encoding convention (or acquisition of reversed-polarity pairs for distortion correction, Section 14.2) rather than assuming vendor defaults are interchangeable.

Lesson: fold-over direction is not a purely local, single-scanner concern — it is a genuine cross-vendor protocol-harmonisation variable for any multi-site EPI study.

Example 3: Sagittal Lumbar Spine — Superior-Inferior Phase to Protect the Cord and Discs

Clinical scenario: routine sagittal T2 lumbar spine.

Protocol logic: phase encoding set superior-inferior (head-foot); pulsatile aortic and CSF-related ghosting propagates along the long axis of the sagittal image rather than across the narrower anterior-posterior extent where the cord, discs, and vertebral bodies are concentrated.

Lesson: the same perpendicular-to-motion-source logic as Example 1, applied to a different anatomical plane and a different dominant motion source.

Example 4: Extremity Joint Imaging — Rectangular FOV Time-Saving

Clinical scenario: routine knee MRI, coronal plane.

Protocol logic: phase encoding aligned with the anatomically narrower medial-lateral dimension of the knee rather than the longer superior-inferior dimension, combined with a rectangular FOV, directly reducing the number of phase-encoding steps required for the same in-plane resolution and shortening scan time without any resolution compromise [6].

Lesson: for elongated extremity anatomy where artefact-avoidance considerations are less pressing than for brain or spine imaging, fold-over direction is primarily a scan-time optimisation lever (Section 7).

Example 5: High-Field Diffusion MRI — Distortion in the Frontal and Temporal Lobes

Clinical scenario: 3T whole-brain DWI for a patient with prior orbitofrontal or anterior temporal lobe surgery, where accurate anatomical localisation of diffusion abnormality in that specific region is the clinical priority.

Problem: the orbitofrontal and anterior temporal regions are classic sites of severe susceptibility-gradient-related B0 inhomogeneity, and EPI geometric distortion along the phase-encoding axis is correspondingly most severe exactly where diagnostic accuracy is most needed [7].

Mitigation: phase-direction selection informed by which specific region is the diagnostic priority, increased phase-direction bandwidth (parallel imaging, shorter echo spacing) to reduce distortion magnitude, and/or acquisition of reversed-polarity phase-encoding pairs for computational correction [1,7].

Lesson: for high-stakes EPI in susceptibility-prone regions, fold-over direction selection and distortion-mitigation strategy should be a deliberate, case-specific decision rather than an inherited protocol default.

18. Visual Educational Material

18.1 Perpendicular-to-Motion-Source Decision Logic

MOTION SOURCE ---- (line of concern) ---- CRITICAL STRUCTURE

If phase-encoding axis IS this line:
  ghost band propagates ALONG the line -> falls directly
  on the critical structure

If phase-encoding axis is PERPENDICULAR to this line:
  ghost band propagates AWAY from the critical structure
  -> structure remains artefact-free

Example (axial brain): eyes ---- (A-P line) ---- brain parenchyma
  Phase = A-P  -> ghost crosses parenchyma (AVOID)
  Phase = L-R  -> ghost stays lateral to orbits (PREFERRED)

18.2 Rectangular FOV Time-Saving Logic

Anatomy: 300 mm (long axis) x 150 mm (short axis)

Phase encoding along LONG axis (300 mm):
  more phase-encoding steps needed for matched resolution
  -> LONGER scan time

Phase encoding along SHORT axis (150 mm), rectangular FOV:
  fewer phase-encoding steps needed for matched resolution
  -> SHORTER scan time, same in-plane resolution

18.3 EPI Distortion Concentrated on the Phase Axis

Conventional (non-EPI) sequence:
  Frequency-direction bandwidth ~ 400-600+ Hz/pixel -> distortion negligible
  Phase-direction: stepped, but no readout-bandwidth-related distortion mechanism

Single-shot EPI:
  Frequency-direction bandwidth: high (readout largely unchanged)
  Phase-direction EFFECTIVE bandwidth: very low (~20-40 Hz/pixel typical)
    -> B0 inhomogeneity produces LARGE pixel shifts specifically
       along the phase-encoding axis
    -> reported displacement: up to several mm, occasionally
       reaching the centimetre range in high-susceptibility-
       gradient regions [1,7]

19. Evidence Gaps and Ongoing Debate

No formally validated, universally-adopted decision algorithm exists for optimal phase-encoding direction selection across the full range of clinical scenarios: the "perpendicular to the dominant motion source" principle (Section 13.1) is well-established, widely taught practical wisdom rather than the product of a systematic, prospective comparative trial across anatomical regions and clinical indications.

The relative clinical impact of vendor-default phase-direction divergence (Siemens A>>P vs GE P-A) on multi-site study comparability is acknowledged in the technical/methods community but has not been comprehensively, quantitatively characterised across the full range of clinical and research EPI applications.

Optimal, generalisable strategies for balancing reversed-phase-encoding-polarity distortion correction (extra acquisition time) against single-direction acquisition with computational correction alone remain an active area of technical development rather than settled, guideline-level clinical practice.

20. Miscellaneous and Future Directions

Continued improvement in gradient hardware performance (higher amplitude, faster slew rate) progressively shortens EPI echo spacing and therefore increases effective phase-direction bandwidth, gradually reducing (though not eliminating) the clinical significance of phase-direction-dependent EPI distortion on the most modern scanner platforms.

Deep-learning-based EPI distortion correction, building on classical field-map and reversed-polarity correction methods, is an active area of technical development that may progressively reduce the acquisition-time cost currently associated with robust distortion correction (Section 14.2).

AI-assisted automated phase-direction recommendation, informed by the specific anatomical region and clinical question rather than a fixed protocol default, is a conceptually natural extension of the case-by-case reasoning developed throughout this page, though not yet established clinical practice.

End of document — Fold-over Direction (Phase-Encoding Direction) — MRIninja v1.0 — August 2026

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