Acceleration Axis¶

The Acceleration axis identifies k-space acceleration methods used to reduce scan time. It answers the question: "How was the acquisition sped up?"

Overview¶

Accelerations are additive - multiple methods can be used simultaneously in a single acquisition.

Acceleration	Abbreviation	Mechanism
ParallelImaging	PI	Coil sensitivity encoding
SimultaneousMultiSlice	SMS	Multiband excitation
PartialFourier	PF	Incomplete k-space with reconstruction
CompressedSensing	CS	Sparse reconstruction
ViewSharing	VS	Temporal k-space reuse

Key Concepts¶

Additive Logic¶

Multiple accelerations can coexist:

fMRI acquisition → acceleration_csv = "ParallelImaging,PartialFourier,SimultaneousMultiSlice"
Standard T1w    → acceleration_csv = "ParallelImaging"
Dynamic MRA     → acceleration_csv = "ParallelImaging,ViewSharing"

Database Coverage¶

From 457K+ validated fingerprints:

Acceleration	Stack Count	Common Use
Partial Fourier (Phase)	98,118	TSE, EPI
Parallel Imaging	80,746	Nearly universal
Multiband/SMS	37,982	fMRI, DWI
Partial Fourier (Freq)	35,041	Less common
HyperSense	12,969	GE specific
Compressed Sensing	~200	Newer sequences

Detection Strategy¶

Acceleration uses a five-tier detection priority:

Unified flags (95%) - Pre-computed from multiple sources
Scan options flags (90%) - Vendor-specific DICOM tags
DICOM tag (90%) - MR Parallel Acquisition Technique
Keywords (80%) - Text matching
Sequence name patterns (75%) - Regex matching

ParallelImaging (GRAPPA/SENSE/ARC)¶

Physics: Uses spatial sensitivity profiles of multiple receiver coils to reduce phase-encoding steps. Missing k-space data is reconstructed using coil sensitivity information.

Vendor Names¶

Vendor	Names
Siemens	GRAPPA, mSENSE, iPAT
GE	ARC, ASSET
Philips	SENSE, SPEEDER
Canon	SPEEDER

Detection¶

Unified flags: - has_parallel_imaging

Scan options: - has_parallel_gems (GE ACC_GEMS) - has_hypersense (GE HyperSense)

DICOM tag: MR Parallel Acquisition Technique (0018,9078) - Values: SENSE, GRAPPA, SPEEDER, CSENSE

Keywords: - grappa, sense, asset, ipat, msense, accelerat - Bounded patterns: \barc\b, arc[, _arc_

Exclusions: hypersense (separate category)

Characteristics¶

Typical R factor: 2-4x
Requirement: Multi-channel coil arrays
Trade-off: SNR reduction (√R penalty)
Nearly universal in modern clinical MRI

SimultaneousMultiSlice (SMS/Multiband)¶

Physics: Uses composite RF pulses to excite multiple slices at once. Slice separation uses coil sensitivity encoding (similar to parallel imaging).

Aliases¶

Multiband (MB)
SMS
SMS-EPI
HyperBand

Detection¶

Keywords: - multiband, multib, hyperband - Bounded patterns: \bmb\d, mb[, \bsms\b, _sms_

Sequence patterns: - cmrr (CMRR multiband sequences) - hyperband

Exclusions: combat, ambig, membrane, chamber, number

Characteristics¶

Typical MB factor: 2-8x
Primary use: EPI sequences (fMRI, DWI)
Trade-off: SNR reduction, slice cross-talk
Requires post-processing to separate slices

PartialFourier (Half-Fourier)¶

Physics: Exploits Hermitian symmetry of k-space to acquire only slightly more than half of k-space. Missing data is estimated using homodyne/POCS reconstruction.

Subtypes¶

Subtype	Direction	Scan Option
PFP	Phase-encoding	More common
PFF	Frequency-encoding	Less common

Aliases¶

Half-Fourier
5/8, 6/8, 7/8 (fraction of k-space)

Detection¶

Unified flags: - has_partial_fourier

Scan options: - has_partial_fourier_phase (PFP) - has_partial_fourier_freq (PFF)

Keywords: - partial fourier, half fourier, half-fourier - Bounded patterns: \bpf\b, 5/8, 6/8, 7/8

Characteristics¶

Time reduction: ~20-40%
SNR penalty: Some due to reconstruction
Essential for: HASTE/SS-FSE sequences
Can specify both phase and frequency directions

CompressedSensing (CS)¶

Physics: Exploits sparsity of MR images in transform domains (wavelets, etc.) to reconstruct from highly undersampled, incoherently sampled k-space. Uses iterative nonlinear reconstruction.

Aliases¶

CS-SENSE
Sparse MRI
Wave-CAIPI

Detection¶

Scan options: - has_cs_gems (GE CS_GEMS)

Keywords: - compressed sensing, compressedsense, sparse - wave-caipi, caipi - Bounded patterns: \bcs\[, _cs_

Exclusions: csf (cerebrospinal fluid), csa

Characteristics¶

Clinically available: ~2015+
Typical R factor: 4-10x+ possible
Reconstruction: Computationally intensive
Often combined with parallel imaging

ViewSharing (TWIST/TRICKS/Keyhole)¶

Physics: Updates only central (low-frequency) k-space frequently while reusing peripheral (high-frequency) data from adjacent time frames. Enables high temporal resolution for dynamic imaging.

Aliases¶

Vendor	Name
Siemens	TWIST
GE	TRICKS, DISCO, 4D-TRAK
Generic	Keyhole

Detection¶

Keywords: - twist, tricks, keyhole - view sharing, disco, 4d-trak - time-resolved, differential subsampling

Sequence patterns: - fldyn (dynamic with view sharing)

Characteristics¶

Primary use: Dynamic/DCE imaging, time-resolved MRA
Trade-off: Spatial resolution for temporal resolution
Often combined with parallel imaging

Output Format¶

Acceleration output is a list (can be empty or contain multiple):

{
    "acceleration_csv": "ParallelImaging,PartialFourier,SimultaneousMultiSlice"
}

Or as list:

["ParallelImaging", "PartialFourier", "SimultaneousMultiSlice"]

Common Combinations¶

Sequence Type	Typical Accelerations
Standard T1 MPRAGE	ParallelImaging
T2 TSE	ParallelImaging, PartialFourier
DWI-EPI	ParallelImaging, PartialFourier
fMRI (multiband)	ParallelImaging, PartialFourier, SimultaneousMultiSlice
Time-resolved MRA	ParallelImaging, ViewSharing
3D with CS	ParallelImaging, CompressedSensing
HASTE	PartialFourier

Confidence Levels¶

Detection Method	Confidence
Unified flag	95%
Scan options	90%
DICOM tag	90%
Keywords	80%
Sequence pattern	75%

Examples¶

Series Description	Detected Accelerations
`t1_mprage_sag_p2_iso`	ParallelImaging
`ep2d_diff_mddw_30_mb3_p2`	ParallelImaging, SimultaneousMultiSlice
`tse_tra_fs_pf`	PartialFourier
`t1_space_sag_grappa2`	ParallelImaging
`ep_bold_mb4`	SimultaneousMultiSlice
`twist_mra_tra`	ViewSharing
`t1_mprage_cs`	CompressedSensing

Detection Notes¶

False Positive Prevention¶

Several keywords have built-in exclusion patterns to prevent false positives:

Acceleration	Exclusions
ParallelImaging	`hypersense`
SMS	`combat`, `ambig`, `membrane`, `chamber`, `number`
CompressedSensing	`csf`, `csa`

Word Boundary Patterns¶

Some keywords use word boundaries to avoid false matches: - \barc\b - Matches "ARC" but not "search" or "march" - \bmb\d - Matches "mb2", "mb3" but not "symbol" - \bsms\b - Matches standalone "SMS"

YAML Configuration¶

Acceleration is configured in backend/src/classification/detection_yaml/acceleration-detection.yaml:

accelerations:
  ParallelImaging:
    name: "ParallelImaging"
    abbreviation: "PI"
    description: "Coil-sensitivity-based acceleration"

    detection:
      unified_flags:
        - has_parallel_imaging
      scan_options_flags:
        - has_parallel_gems
        - has_hypersense
      dicom_tag:
        tag: "(0018,9078)"
        values: ["SENSE", "GRAPPA"]
      keywords:
        - "grappa"
        - "sense"
      exclude_keywords:
        - "hypersense"

  SimultaneousMultiSlice:
    name: "SimultaneousMultiSlice"
    abbreviation: "SMS"
    detection:
      keywords:
        - "multiband"
        - "hyperband"
      sequence_name_patterns:
        - "cmrr"
      exclude_keywords:
        - "combat"

detection_priority:
  - unified_flags
  - scan_options_flags
  - dicom_tag
  - keywords
  - sequence_name_patterns

confidence_thresholds:
  unified_flag: 0.95
  scan_options: 0.90
  keywords: 0.80

Convenience Methods¶

# Check if any acceleration detected
output.has_acceleration  # True/False

# Check specific acceleration
output.has("ParallelImaging")  # True/False

# Get all detected acceleration names
output.values  # ["ParallelImaging", "PartialFourier"]

# Get specific acceleration result with details
result = output.get("PartialFourier")
result.subtype  # "phase" or "frequency" or "phase+frequency"
result.confidence  # 0.95
result.detection_method  # "unified_flag"