SWI Branch¶

The SWI branch handles classification of Susceptibility-Weighted Imaging outputs, a post-processed modality that produces multiple distinct series from a single acquisition.

Overview¶

Unlike conventional T1 or T2-weighted imaging that produces a single output, SWI is a computational reconstruction that synthesizes magnitude (anatomy) and phase (susceptibility) data into multiple diagnostic outputs.

Aspect	Description
Provenance	`SWIRecon`
Classification Branch	`swi`
Source Module	`backend/src/classification/branches/swi.py`
Output Types	6 distinct types
Series per Acquisition	2-6 typically

The Physics: Susceptibility Contrast¶

Why SWI is Different¶

Traditional MRI contrast (T1, T2) arises from proton relaxation properties. SWI contrast arises from magnetic susceptibility—how tissues distort the local magnetic field.

flowchart LR
    subgraph diamagnetic["Diamagnetic (χ < 0)"]
        d1["Calcium, Water, Soft tissue"]
        d2["→ Lower local field"]
        d3["→ Negative phase shift"]
    end

    subgraph paramagnetic["Paramagnetic (χ > 0)"]
        p1["Deoxyhemoglobin (blood)"]
        p2["Hemosiderin (old blood)"]
        p3["Ferritin (iron storage)"]
        p4["→ Higher local field"]
        p5["→ Positive phase shift"]
    end

Key Insight: Phase Differentiates Iron from Calcium¶

On magnitude images, both iron (paramagnetic) and calcium (diamagnetic) appear dark due to signal dephasing. Only phase images can distinguish them:

Substance	Magnitude	Phase (Right-Handed)	Phase (Left-Handed)
Iron/Blood	Dark (blooming)	Bright (+)	Dark (-)
Calcium	Dark (blooming)	Dark (-)	Bright (+)

Why GRE, Not Spin Echo¶

SWI must use Gradient Echo (GRE) acquisition:

Spin Echo: The 180° refocusing pulse nullifies susceptibility-induced phase shifts
Gradient Echo: No refocusing pulse, allowing T2* effects to accumulate

This is why all SWI outputs are inherently T2*-weighted.

The Four-Series Output¶

A single SWI acquisition produces multiple series, each representing a distinct processing stage:

Processing Pipeline¶

flowchart TB
    raw["Raw k-space Data (Complex)"]

    raw --> mag["Series 1: Magnitude<br/>|S| = √(R² + I²)<br/>T2*-weighted anatomy"]
    raw --> phase["Series 2: Phase<br/>High-pass filtered<br/>Removes background field"]

    mag --> mask
    phase --> mask["Phase Mask Creation<br/>Negative phase → 0-1 range<br/>Positive phase → 1"]

    mask --> swi["Series 3: SWI (Processed)<br/>I_SWI = I_Mag × [Mask]^n (n=4)<br/>Extreme vein/microbleed sensitivity"]

    swi --> minip["Series 4: MinIP (Venogram)<br/>Minimum across slab of slices<br/>Shows 3D venous continuity"]

Why Each Series Matters¶

Series	Purpose	Clinical Use
Magnitude	Anatomical reference	Localization, size estimation
Phase	Iron vs calcium differentiation	Definitive lesion characterization
SWI	Maximum sensitivity	Microbleed detection, tumor characterization
MinIP	3D vascular visualization	DVA evaluation, venography

Output Types¶

All SWI outputs share base=SWI to indicate the contrast type. The construct specifies the output:

Output	Base	Construct	Description
Magnitude	SWI	`Magnitude`	Source T2*-weighted image
Phase	SWI	`Phase`	Filtered phase map
SWI	SWI	`SWI`	Processed (Mag × Phase mask^4)
MinIP	SWI	`MinIP`	Minimum intensity projection
MIP	SWI	`MIP`	Maximum intensity projection
QSM	SWI	`QSM`	Quantitative susceptibility map (ppm)

Detection Strategy¶

Priority Order¶

The SWI branch uses strict priority (first match wins):

flowchart TB
    start["SWI Branch Input"] --> q1{has_qsm OR<br/>'qsm' in text?}
    q1 -->|Yes| qsm["1. QSM<br/>(Quantitative)"]
    q1 -->|No| q2{is_minip OR<br/>'minip' in text?}

    q2 -->|Yes| minip["2. MinIP<br/>(Venogram)"]
    q2 -->|No| q3{is_mip OR<br/>'mip' in text?}

    q3 -->|Yes| mip["3. MIP"]
    q3 -->|No| q4{has_phase<br/>without magnitude?}

    q4 -->|Yes| phase["4. Phase"]
    q4 -->|No| q5{has_swi token<br/>in ImageType?}

    q5 -->|Yes| swi["5. SWI Processed"]
    q5 -->|No| q6{is_projection?}

    q6 -->|Yes| swi2["5.5. SWI (projection)"]
    q6 -->|No| q7{has_magnitude?}

    q7 -->|Yes| mag["6. Magnitude"]
    q7 -->|No| fallback["7. Fallback → SWI"]

Key Detection Insight¶

The SWI token in ImageType is the critical differentiator between source magnitude and processed SWI:

ImageType	Detection	Output
`ORIGINAL\PRIMARY\M`	has_magnitude, no has_swi	Magnitude
`ORIGINAL\PRIMARY\M\SWI`	has_magnitude AND has_swi	SWI

Both contain the M (magnitude) token, but only processed SWI has the SWI token.

Technique Detection¶

SWI is a processing method, not a technique. The actual acquisition is either GRE or EPI:

Technique Priority¶

if has_epi:
    technique = "EPI"       # Fast SWI (1-2 min)
elif has_gre:
    technique = "GRE"       # Standard SWI (4-6 min)
elif "epi" in text_blob or "3depi" in text_blob:
    technique = "EPI"       # Text fallback for GE
else:
    technique = "GRE"       # Default

Acquisition Variants¶

Variant	Physics	Time	Trade-offs
Standard GRE	3D Flow-compensated GRE	4-6 min	Highest resolution, motion sensitive
EPI-SWI	Echo Planar Imaging	1-2 min	Fast, geometric distortion
SWAN	Multi-echo GRE	4-5 min	Higher SNR, better skull base coverage

Quantitative Susceptibility Mapping (QSM)¶

From Qualitative to Quantitative¶

SWI produces qualitative contrast (dark = susceptibility source). QSM produces quantitative measurements in parts per million (ppm).

Aspect	SWI	QSM
Output	Contrast image	Susceptibility map (ppm)
Blooming	Artifacts appear larger	True geometry
Orientation	Angle-dependent	Angle-independent
Iron/Calcium	Both appear dark	Iron=bright, Calcium=dark

QSM Processing Pipeline¶

QSM uses the same raw data as SWI but with additional processing:

Phase Unwrapping - Remove 2π aliasing
Background Field Removal - Remove air-tissue interface effects
Dipole Inversion - Solve inverse problem: ΔB = D ⊗ χ

Detection¶

QSM is detected with highest priority because it represents the most specific output type:

if uf.get("has_qsm") or "qsm" in text_blob:
    return BranchResult(
        base="SWI",
        construct="QSM",
        confidence=0.95,
    )

ImageType Token Patterns¶

Standard SWI Tokens¶

Pattern	Meaning	Output
`ORIGINAL\PRIMARY\M`	Magnitude only	Magnitude
`ORIGINAL\PRIMARY\M\SWI`	SWI token present	SWI
`ORIGINAL\PRIMARY\P`	Phase	Phase
`ORIGINAL\PRIMARY\PHASE MAP`	Phase map	Phase
`DERIVED\PRIMARY\MINIP`	Minimum IP	MinIP
`DERIVED\PRIMARY\MIP`	Maximum IP	MIP
`DERIVED\PRIMARY\QSM`	Quantitative susceptibility	QSM

Vendor Variations¶

Vendor	Sequence Names	Notes
Siemens	SWI, fl3d_swi	Standard implementation
GE	SWAN, 3DEPIks	Multi-echo (SWAN), EPI variants
Philips	SWI, PRESTO	Various implementations

Clinical Significance¶

Why SWI Matters¶

Application	SWI Advantage
Microbleeds	6x more sensitive than T2*-GRE
Trauma	Detects diffuse axonal injury
Tumors	Visualizes neovascularization, calcification
Stroke	Susceptibility vessel sign, penumbra
Parkinson's	"Swallow tail" sign in substantia nigra
MS	Central vein sign in lesions

Correct Classification Importance¶

Without SWI Branch	With SWI Branch
Magnitude → "T2*-weighted"	Magnitude → base=SWI, construct=Magnitude
Phase → unknown	Phase → base=SWI, construct=Phase
Processed SWI → confused with magnitude	SWI → base=SWI, construct=SWI
MinIP → "MIP"	MinIP → base=SWI, construct=MinIP
QSM → unknown	QSM → base=SWI, construct=QSM

Configuration Reference¶

Unified Flags Used¶

Flag	Description
`has_qsm`	QSM token in ImageType
`is_minip`	MinIP/MNIP token
`is_mip`	MIP token (excluding MinIP)
`has_phase`	Phase (P, PHASE, PHASE MAP)
`has_swi`	SWI token in ImageType
`has_magnitude`	Magnitude (M, M_FFE, etc.)
`is_projection`	PROJECTION IMAGE token
`has_epi`	EPI readout
`has_gre`	GRE readout

Output Type Mapping¶

From backend/src/classification/branches/common.py:

SWI_OUTPUT_TYPES = {
    "magnitude": {
        "base": "SWI",
        "construct": "Magnitude",
        "description": "SWI magnitude source image (T2*-weighted)",
    },
    "phase": {
        "base": "SWI",
        "construct": "Phase",
        "description": "SWI phase map (iron/calcium differentiation)",
    },
    "swi": {
        "base": "SWI",
        "construct": "SWI",
        "description": "Processed SWI image (magnitude × phase mask)",
    },
    "minip": {
        "base": "SWI",
        "construct": "MinIP",
        "description": "Minimum intensity projection (venogram)",
    },
    "mip": {
        "base": "SWI",
        "construct": "MIP",
        "description": "Maximum intensity projection",
    },
    "qsm": {
        "base": "SWI",
        "construct": "QSM",
        "description": "Quantitative susceptibility map (ppm)",
    },
}

Examples¶

Example 1: Source Magnitude¶

DICOM Fields:

ImageType: ORIGINAL\PRIMARY\M\ND
SeriesDescription: SWI Magnitude
ScanningSequence: GR

Classification:

base = "SWI"
construct = "Magnitude"
technique = "GRE"
directory_type = "anat"

Example 2: Processed SWI¶

DICOM Fields:

ImageType: ORIGINAL\PRIMARY\M\SWI\ND
SeriesDescription: SWI
ScanningSequence: GR

Classification:

base = "SWI"
construct = "SWI"
technique = "GRE"
directory_type = "anat"

Note: Both examples have magnitude (M) token, but the SWI token differentiates them.

Example 3: Phase Map¶

DICOM Fields:

ImageType: ORIGINAL\PRIMARY\P\ND
SeriesDescription: SWI Phase

Classification:

base = "SWI"
construct = "Phase"
technique = "GRE"
directory_type = "anat"

Example 4: MinIP (Venogram)¶

DICOM Fields:

ImageType: DERIVED\SECONDARY\MINIP
SeriesDescription: SWI mIP

Classification:

base = "SWI"
construct = "MinIP"
technique = "GRE"
directory_type = "anat"

Example 5: QSM¶

DICOM Fields:

ImageType: DERIVED\PRIMARY\QSM
SeriesDescription: QSM

Classification:

base = "SWI"
construct = "QSM"
technique = "GRE"
directory_type = "anat"

Example 6: EPI-SWI¶

DICOM Fields:

ImageType: ORIGINAL\PRIMARY\M\SWI
SeriesDescription: 3DEPIks SWI
ScanningSequence: EP

Classification:

base = "SWI"
construct = "SWI"
technique = "EPI"      # Detected from has_epi flag
directory_type = "anat"

Troubleshooting¶

Common Issues¶

Issue: Magnitude classified as SWI - Cause: Both have M token; branch checking SWI token last - Solution: Detection priority ensures has_swi is checked before fallback to Magnitude

Issue: MinIP classified as MIP - Cause: Both contain "mip" substring - Solution: MinIP is checked first with explicit "minip" pattern

Issue: Phase not detected - Cause: has_phase AND has_magnitude both true - Solution: Check has_phase AND NOT has_magnitude

Issue: EPI-SWI detected as GRE - Cause: GE uses "RM" in ScanningSequence, not "EP" - Solution: Text fallback checks for "epi" or "3depi" in text_search_blob

Issue: Unknown SWI output - Cause: No matching tokens or keywords - Solution: Falls back to SWI (construct="SWI") with lower confidence

SWI Branch¶

Overview¶

The Physics: Susceptibility Contrast¶

Why SWI is Different¶

Key Insight: Phase Differentiates Iron from Calcium¶

Why GRE, Not Spin Echo¶

The Four-Series Output¶

Processing Pipeline¶

Why Each Series Matters¶

Output Types¶

Detection Strategy¶

Priority Order¶

Key Detection Insight¶

Technique Detection¶

Technique Priority¶

Acquisition Variants¶

Quantitative Susceptibility Mapping (QSM)¶

From Qualitative to Quantitative¶

QSM Processing Pipeline¶

Detection¶

ImageType Token Patterns¶

Standard SWI Tokens¶

Vendor Variations¶

Clinical Significance¶

Why SWI Matters¶

Correct Classification Importance¶

Configuration Reference¶

Unified Flags Used¶

Output Type Mapping¶

Examples¶

Example 1: Source Magnitude¶

Example 2: Processed SWI¶

Example 3: Phase Map¶

Example 4: MinIP (Venogram)¶

Example 5: QSM¶

Example 6: EPI-SWI¶

Troubleshooting¶

Common Issues¶

See Also¶