SyMRI Branch¶

The SyMRI branch handles classification of Synthetic MRI outputs from SyMRI, MAGiC (GE), MDME, and QALAS sequences.

Overview¶

Synthetic MRI represents a paradigm shift from qualitative contrast-weighted imaging to quantitative tissue property measurement. A single acquisition produces multiple output types, each requiring specialized classification.

Aspect	Description
Provenance	`SyMRI`
Classification Branch	`symri`
Source Module	`backend/src/classification/branches/symri.py`
Output Types	17+ distinct types
Series per Acquisition	1-20+ (highly variable)

The Physics: Multi-Dynamic Multi-Echo (MDME)¶

How SyMRI Works¶

Unlike conventional MRI that produces a single contrast-weighted image, SyMRI uses the MDME (Multi-Dynamic Multi-Echo) sequence to sample tissue signal evolution:

flowchart TB
    subgraph acquisition["MDME Acquisition"]
        acq1["4 Saturation Delays × 2 Echo Times = 8 Volumes"]
        acq2["Dynamic 1 (TI ~100ms) → Echo 1 + Echo 2"]
        acq3["Dynamic 2 (TI ~600ms) → Echo 1 + Echo 2"]
        acq4["Dynamic 3 (TI ~1500ms) → Echo 1 + Echo 2"]
        acq5["Dynamic 4 (TI ~4000ms) → Echo 1 + Echo 2"]
        acq6["Each volume: Magnitude + Phase = 16 potential series"]
    end

    subgraph quant["Quantification Algorithm"]
        q1["Solves Bloch equations for T1, T2, PD at each voxel"]
    end

    subgraph outputs["Outputs"]
        o1["Quantitative Maps (T1map, T2map, PDmap)"]
        o2["Synthetic Weighted Images (any contrast!)"]
        o3["Derived Maps (Myelin, B1, R1, R2)"]
    end

    acquisition --> quant --> outputs

Why Series Count Varies¶

A single SyMRI acquisition can produce 1 to 20+ DICOM series:

Export Configuration	Series Count	Contents
Single synthetic contrast	1	Just T2w or FLAIR
Magnitude + Phase pairs	2	Complex data for each output
Quantitative maps only	3-5	T1map, T2map, PDmap, B1map
Full raw data	8	4 dynamics × 2 echoes
Full raw + Phase	16	All magnitude and phase
Clinical protocol	12+	Maps + multiple synthetic contrasts

This variability is deterministic, not random - it depends on the export configuration.

Output Types¶

Critical Distinction: base=NULL vs base=T1w/T2w¶

The SyMRI branch understands a crucial physics distinction:

Quantitative maps have NO tissue contrast → base=NULL
A T1 map pixel value of "800" means "800 milliseconds"
This is a measurement, not an image contrast
Cannot be meaningfully labeled as "T1-weighted"
Synthetic weighted images HAVE tissue contrast → base=T1w/T2w/PDw
Simulate what a real acquisition would look like
Get proper base contrast labels

Raw Source Data¶

The MDME/QALAS acquisition produces complex (magnitude + phase) data:

Output	Base	Construct	Description
Magnitude	NULL	`Magnitude`	Raw magnitude source (4×2 volumes)
Phase	NULL	`Phase`	Raw phase source (for T1 polarity detection)

Detection: - Magnitude: has_magnitude flag, M_SE in ImageType - Phase: has_phase flag (without has_magnitude)

Quantitative Maps¶

Parametric maps with physical units - NO tissue contrast:

Output	Base	Construct	Units	Description
T1 Map	NULL	`T1map`	ms	Longitudinal relaxation time
T2 Map	NULL	`T2map`	ms	Transverse relaxation time
PD Map	NULL	`PDmap`	%	Proton density
R1 Map	NULL	`R1map`	1/s	1/T1 rate
R2 Map	NULL	`R2map`	1/s	1/T2 rate
B1 Map	NULL	`B1map`	%	RF field inhomogeneity
MultiQmap	NULL	`MultiQmap`	-	GE MAGiC bundled T1/T2/PD
Myelin Map	NULL	`MyelinMap`	%	Myelin water fraction

Detection priority: 1. Unified flags: has_t1_map, has_t2_map, has_pd_map, has_myelin 2. QMAP tokens: has_qmap_t1, has_qmap_t2, has_qmap_pd 3. Text keywords: "t1 map", "t2map", "myelin"

Synthetic Weighted Images¶

Simulated contrast-weighted images - HAVE tissue contrast:

Output	Base	Construct	Modifiers	Description
Synthetic T1w	T1w	`SyntheticT1w`	-	Synthetic T1-weighted
Synthetic T2w	T2w	`SyntheticT2w`	-	Synthetic T2-weighted
Synthetic PDw	PDw	`SyntheticPDw`	-	Synthetic PD-weighted

Detection: - Unified flags: has_t1_synthetic, has_t2_synthetic, has_pd_synthetic - ImageType pattern: T1\SYNTHETIC or T1W_SYNTHETIC

Synthetic IR Contrasts¶

Synthetic versions of inversion recovery sequences:

Output	Base	Construct	Modifiers	Description
Synthetic FLAIR	T2w	`SyntheticFLAIR`	FLAIR	CSF-nulled T2w
Synthetic DIR	T2w	`SyntheticDIR`	DIR	Double inversion recovery
Synthetic PSIR	T1w	`SyntheticPSIR`	PSIR	Phase-sensitive IR
Synthetic STIR	T2w	`SyntheticSTIR`	STIR	Fat-suppressed

Detection: - Unified flags: has_flair_synthetic, has_dir_synthetic, has_psir_synthetic - Text keywords: "flair" + "synthetic"

Detection Strategy¶

Priority Order¶

The SyMRI branch uses a strict detection priority (first match wins):

flowchart TB
    start["SyMRI Branch Input"] --> q1{Myelin indicators?}
    q1 -->|Yes| myelin["1. Myelin Map"]
    q1 -->|No| q2{MultiQmap?}

    q2 -->|Yes| multi["2. MultiQmap"]
    q2 -->|No| q3{Quantitative map?}

    q3 -->|Yes| qmaps["3. T1map / T2map / PDmap<br/>R1map / R2map / B1map"]
    q3 -->|No| q4{Synthetic IR?}

    q4 -->|Yes| ir["4. FLAIR / DIR / PSIR / STIR"]
    q4 -->|No| q5{Synthetic weighted?}

    q5 -->|Yes| synth["5. SyntheticT1w / T2w / PDw"]
    q5 -->|No| q6{Phase only?}

    q6 -->|Yes| phase["6. Raw Phase"]
    q6 -->|No| q7{Magnitude?}

    q7 -->|Yes| mag["7. Raw Magnitude"]
    q7 -->|No| fallback["8. Fallback → Magnitude"]

Why Myelin is First¶

Myelin maps often contain the T1 token (e.g., DERIVED\PRIMARY\MYC\T1). If T1 map detection ran first, myelin maps would be misclassified as T1 maps.

Technique Detection¶

SyMRI determines the acquisition technique:

Technique	Detection	Description
QALAS	`is_qalas` flag OR `qalas` in sequence name	3D quantitative acquisition
MDME	Default	2D multi-dynamic multi-echo

ImageType Token Patterns¶

Quantitative Map Tokens¶

Pattern	Meaning	Output
`DERIVED\PRIMARY\T1`	T1 map (no SYNTHETIC)	T1map
`DERIVED\PRIMARY\T2`	T2 map	T2map
`DERIVED\PRIMARY\PD`	PD map	PDmap
`DERIVED\PRIMARY\QMAP\T1`	Explicit qmap T1	T1map
`DERIVED\PRIMARY\MYC`	Myelin fraction	MyelinMap
`DERIVED\PRIMARY\MULTI_QMAP\T1\T2\PD`	GE bundled maps	MultiQmap

Synthetic Weighted Tokens¶

Pattern	Meaning	Output
`DERIVED\PRIMARY\T1\SYNTHETIC`	Synthetic T1w	SyntheticT1w
`DERIVED\PRIMARY\T1W_SYNTHETIC`	Synthetic T1w (alt)	SyntheticT1w
`DERIVED\PRIMARY\T2\SYNTHETIC`	Synthetic T2w	SyntheticT2w
`DERIVED\PRIMARY\FLAIR\SYNTHETIC`	Synthetic FLAIR	SyntheticFLAIR
`DERIVED\PRIMARY\DIR_SYNTHETIC`	Synthetic DIR	SyntheticDIR

Raw Source Tokens¶

Pattern	Meaning	Output
`ORIGINAL\PRIMARY\M`	Magnitude	Magnitude
`ORIGINAL\PRIMARY\M_SE`	SE magnitude (Philips)	Magnitude
`ORIGINAL\PRIMARY\P`	Phase	Phase
`ORIGINAL\PRIMARY\PHASE MAP`	Phase map	Phase

Vendor Implementations¶

GE: MAGiC¶

Sequence name: Often contains "MAGIC" or proprietary identifiers
Private tags: GEHC_SyMR group (3005) for re-quantification
MultiQmap: Bundles T1/T2/PD in single series
Report Card: May include brain parenchymal fraction series

Siemens: SyMRI OpenApps¶

Sequence name: MDME patterns
Explicit saves: Users manually export maps vs synthetic images
Series description: Often includes "SyMRI" or "Synthetic"

Philips: SyntAc¶

Integration: With SmartSpeed pipeline
Complex data: May bundle magnitude/phase in legacy formats

Clinical Significance¶

Why SyMRI Matters¶

Time efficiency: Single 5-min acquisition replaces multiple sequences
Quantitative biomarkers: T1/T2 values for tissue characterization
Retrospective contrast: Generate any weighting post-acquisition
Myelin mapping: Direct visualization of demyelination (MS research)

Correct Classification Importance¶

Without SyMRI Branch	With SyMRI Branch
T1 map → "T1-weighted" ❌	T1 map → base=NULL, construct=T1map ✅
Synthetic T1w → unknown ❌	Synthetic T1w → base=T1w, construct=SyntheticT1w ✅
Myelin map → T1 map ❌	Myelin map → base=NULL, construct=MyelinMap ✅
8 raw volumes → confusion ❌	Raw Magnitude/Phase correctly identified ✅

Configuration Reference¶

Unified Flags Used¶

Flag	Description
`has_myelin`	Myelin (MYC) token detected
`has_multi_qmap`	MULTI_QMAP token (GE MAGiC)
`has_qmap_t1`, `has_qmap_t2`, `has_qmap_pd`	QMAP\Tx tokens
`has_t1_map`, `has_t2_map`, `has_pd_map`	TxMAP tokens
`has_r1`, `has_r2`	R1/R2 rate maps
`has_b1_map`	B1 field map
`has_t1_synthetic`, `has_t2_synthetic`, `has_pd_synthetic`	Synthetic weighted
`has_flair_synthetic`, `has_dir_synthetic`, `has_psir_synthetic`	Synthetic IR
`is_qalas`	3D-QALAS acquisition
`is_magic_raw`	GE MAGiC raw source
`has_magnitude`, `has_phase`	Complex components

Output Type Mapping¶

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

SYMRI_OUTPUT_TYPES = {
    # Raw source
    "magnitude": {"base": None, "construct": "Magnitude"},
    "phase": {"base": None, "construct": "Phase"},

    # Quantitative maps
    "t1_map": {"base": None, "construct": "T1map"},
    "t2_map": {"base": None, "construct": "T2map"},
    "pd_map": {"base": None, "construct": "PDmap"},
    "r1_map": {"base": None, "construct": "R1map"},
    "r2_map": {"base": None, "construct": "R2map"},
    "b1_map": {"base": None, "construct": "B1map"},
    "multi_qmap": {"base": None, "construct": "MultiQmap"},
    "myelin_map": {"base": None, "construct": "MyelinMap"},

    # Synthetic weighted
    "synthetic_t1w": {"base": "T1w", "construct": "SyntheticT1w"},
    "synthetic_t2w": {"base": "T2w", "construct": "SyntheticT2w"},
    "synthetic_pdw": {"base": "PDw", "construct": "SyntheticPDw"},

    # Synthetic IR
    "synthetic_flair": {"base": "T2w", "construct": "SyntheticFLAIR", "modifiers": ["FLAIR"]},
    "synthetic_dir": {"base": "T2w", "construct": "SyntheticDIR", "modifiers": ["DIR"]},
    "synthetic_psir": {"base": "T1w", "construct": "SyntheticPSIR", "modifiers": ["PSIR"]},
    "synthetic_stir": {"base": "T2w", "construct": "SyntheticSTIR", "modifiers": ["STIR"]},
}

Examples¶

Example 1: T1 Map¶

DICOM Fields:

ImageType: DERIVED\PRIMARY\T1
SeriesDescription: SyMRI T1 Map

Classification:

base = None           # No tissue contrast
construct = "T1map"   # Quantitative T1 map
technique = "MDME"    # Source sequence
directory_type = "anat"

Example 2: Synthetic FLAIR¶

DICOM Fields:

ImageType: DERIVED\PRIMARY\FLAIR\SYNTHETIC
SeriesDescription: SyMRI Synthetic FLAIR

Classification:

base = "T2w"                    # FLAIR is T2-weighted
construct = "SyntheticFLAIR"    # Synthetic version
modifiers = ["FLAIR"]           # Add FLAIR modifier
technique = "MDME"
directory_type = "anat"

Example 3: Raw MDME Magnitude¶

DICOM Fields:

ImageType: ORIGINAL\PRIMARY\M
SeriesDescription: MDME Dyn1 Echo1
ScanningSequence: SE

Classification:

base = None              # Raw data, no contrast
construct = "Magnitude"  # Magnitude component
technique = "MDME"
directory_type = "anat"

Example 4: Myelin Map¶

DICOM Fields:

ImageType: DERIVED\PRIMARY\MYC\T1
SeriesDescription: SyMRI Myelin

Classification:

base = None              # Quantitative, no contrast
construct = "MyelinMap"  # Myelin water fraction
technique = "MDME"
directory_type = "anat"

Note: The T1 token is present but myelin detection takes priority.

Troubleshooting¶

Common Issues¶

Issue: T1 map classified as T1-weighted - Cause: Branch not triggered (provenance not detected as SyMRI) - Solution: Check ProvenanceDetector, add keyword if needed

Issue: Myelin map classified as T1 map - Cause: Detection order issue - Solution: Myelin must be checked before T1 map in priority

Issue: Synthetic images missing construct - Cause: SYNTHETIC token not detected - Solution: Check ImageType parsing, add token pattern

Issue: Unknown SyMRI output defaulting to Magnitude - Cause: No matching detection rules - Solution: Add new unified flag or keyword pattern

SyMRI Branch¶

Overview¶

The Physics: Multi-Dynamic Multi-Echo (MDME)¶

How SyMRI Works¶

Why Series Count Varies¶

Output Types¶

Critical Distinction: base=NULL vs base=T1w/T2w¶

Raw Source Data¶

Quantitative Maps¶

Synthetic Weighted Images¶

Synthetic IR Contrasts¶

Detection Strategy¶

Priority Order¶

Why Myelin is First¶

Technique Detection¶

ImageType Token Patterns¶

Quantitative Map Tokens¶

Synthetic Weighted Tokens¶

Raw Source Tokens¶

Vendor Implementations¶

GE: MAGiC¶

Siemens: SyMRI OpenApps¶

Philips: SyntAc¶

Clinical Significance¶

Why SyMRI Matters¶

Correct Classification Importance¶

Configuration Reference¶

Unified Flags Used¶

Output Type Mapping¶

Examples¶

Example 1: T1 Map¶

Example 2: Synthetic FLAIR¶

Example 3: Raw MDME Magnitude¶

Example 4: Myelin Map¶

Troubleshooting¶

Common Issues¶

See Also¶