Construct Axis¶

The Construct axis identifies computed/derived outputs from MRI acquisitions. It answers the question: "What type of calculated map or derived image is this?"

Overview¶

Constructs are computed outputs - not directly acquired, but calculated from raw acquisition data.

Category	Constructs	Source
Diffusion	ADC, eADC, FA, Trace, MD	DWI/DTI
Perfusion	CBF, CBV, MTT, Tmax, TTP	DSC/ASL
Dixon	Water, Fat, InPhase, OutPhase	Dixon acquisition
Quantitative	T1map, T2map, R1map, R2map, PDmap	Relaxometry
MP2RAGE	INV1, INV2, Uniform, Denoised	MP2RAGE sequence
Synthetic	SyntheticT1w, SyntheticT2w, SyntheticFLAIR	SyMRI/MAGiC
SWI	SWI, Phase, QSM	SWI processing
Projection	MIP, MinIP, MPR	Post-processing
Field Maps	B0map, B1map	Calibration
Components	Magnitude, Real, Imaginary	Complex data

Key Concepts¶

Construct vs DICOM DERIVED¶

These are not the same:

DICOM DERIVED: Generic flag meaning "pixels computed from other images"
Construct: The specific TYPE of computed output (ADC, FA, T1map, etc.)

A DERIVED image could be an ADC map, an MPR reformat, or a subtraction image - the construct tells you which.

Additive Logic¶

Like Modifiers, constructs are additive - multiple can apply:

DTI acquisition → construct_csv = "ADC,FA"
MDME/SyMRI     → construct_csv = "SyntheticT1w,T1map,T2map"

Output Format¶

Comma-separated
Alphabetically sorted
Empty string for original acquisitions (no constructs)

Detection Strategy¶

Same three-tier approach:

Tier 1: Exclusive Flag (95% confidence)¶

ADC:
  detection:
    exclusive: has_adc

Tier 2: Keywords Match (85% confidence)¶

ADC:
  keywords:
    - "adc"
    - "apparent diffusion coefficient"

Tier 3: Combination (75% confidence)¶

SWI:
  detection:
    combination:
      - has_swi
      - is_derived

Diffusion Constructs¶

Derived from diffusion-weighted imaging (DWI/DTI/HARDI).

ADC (Apparent Diffusion Coefficient)¶

Units: mm²/s

Detection: - Exclusive flag: has_adc - Keywords: adc, apparent diffusion coefficient

Computation: Fitted from multi-b-value DWI using: S(b) = S₀ × exp(-b × ADC)

Clinical interpretation: - Low ADC = restricted diffusion (acute stroke, tumor cellularity) - High ADC = free water diffusion

eADC (Exponential ADC)¶

Formula: eADC = exp(-b × ADC)

Detection: - Exclusive flag: has_eadc - Keywords: eadc, exponential adc

Use: Some clinical applications prefer exponential form.

FA (Fractional Anisotropy)¶

Range: 0 (isotropic) to 1 (perfectly anisotropic)

Detection: - Exclusive flag: has_fa - Keywords: fractional anisotropy, fa map

Clinical interpretation: - High FA in white matter tracts (organized fibers) - Low FA in gray matter, CSF - FA reduction indicates white matter damage

Trace¶

Description: Direction-independent diffusion-weighted image.

Detection: - Exclusive flag: has_trace - Keywords: trace, tracew, isodwi

Computation: Combined DWI from multiple directions (geometric mean).

MD (Mean Diffusivity)¶

Description: Average diffusion in all directions.

Detection: Keywords only: mean diffusivity, md map

Relationship: MD ≈ ADC for isotropic diffusion.

Perfusion Constructs¶

Derived from DSC-MRI or ASL perfusion imaging.

CBF (Cerebral Blood Flow)¶

Units: ml/100g/min

Detection: - Exclusive flag: has_cbf - Keywords: cbf, cerebral blood flow

Clinical use: Tissue perfusion assessment.

CBV (Cerebral Blood Volume)¶

Units: ml/100g

Detection: - Exclusive flag: has_cbv - Keywords: cbv, cerebral blood volume

Clinical use: Vascular density, tumor grading.

MTT (Mean Transit Time)¶

Units: seconds

Detection: - Exclusive flag: has_mtt - Keywords: mtt, mean transit time

Relationship: MTT = CBV / CBF (central volume theorem).

Tmax (Time to Maximum)¶

Description: Time to maximum of residue function.

Detection: - Exclusive flag: has_tmax - Keywords: tmax, time to max

Clinical use: Stroke - penumbra identification (Tmax > 6s threshold).

TTP (Time to Peak)¶

Description: Time to peak of contrast bolus.

Detection: - Exclusive flag: has_ttp - Keywords: ttp, time to peak

Dixon Constructs¶

Fat-water separation outputs from Dixon acquisition.

Water¶

Description: Water-only image (fat suppressed).

Detection: - Exclusive flag: has_water - Keywords: water only, dixon water

Advantage: Better fat suppression near metal than spectral methods.

Fat¶

Description: Fat-only image (water suppressed).

Detection: - Exclusive flag: has_fat - Keywords: fat only, dixon fat

Use: Fat quantification, lipid assessment.

InPhase¶

Description: Echo where water and fat signals are in phase.

Detection: - Exclusive flag: has_in_phase - Keywords: in phase, inphase

Note: Can be original acquisition, not necessarily derived.

OutPhase¶

Description: Echo where water and fat signals are opposed.

Detection: - Exclusive flag: has_out_phase - Keywords: opposed phase, out of phase

Use: Signal cancellation at fat-water interfaces reveals fat content.

Quantitative Map Constructs¶

Relaxometry and parameter maps.

T1map¶

Units: milliseconds

Detection: - Exclusive flag: has_t1_map - Keywords: t1 map, t1 relaxation - Combination: is_qmap

Acquisition methods: VFA-GRE, IR sequences, MP2RAGE.

T2map¶

Units: milliseconds

Detection: - Exclusive flag: has_t2_map - Keywords: t2 map, t2 relaxation

Acquisition method: Multi-echo SE.

R1map / R2map¶

Description: Relaxation rate maps (1/T1 and 1/T2).

Detection: - Exclusive flags: has_r1, has_r2 - Keywords: r1 map, r2 map

PDmap¶

Description: Quantitative proton density map.

Detection: Keywords: pd map, proton density map

MP2RAGE Constructs¶

Outputs from MP2RAGE acquisition. All are fundamentally T1-weighted.

INV1 (First Inversion)¶

TI: ~700-1000ms (short)

Detection: - Exclusive flag: is_mp2rage_inv1 - Keywords: inv1, inv 1

Characteristics: - Strong T1 weighting - Lower SNR than INV2 - Not derived (original acquisition)

INV2 (Second Inversion)¶

TI: ~2500-3200ms (long)

Detection: - Exclusive flag: is_mp2rage_inv2 - Keywords: inv2, inv 2

Characteristics: - Higher SNR - Weaker T1 contrast - Not derived (original acquisition)

Uniform¶

Description: Bias-corrected T1w image.

Detection: - Exclusive flag: has_uniform - Keywords: uni image, uniform image

Computation:

S_UNI = Re(INV1* · INV2) / (|INV1|² + |INV2|²)

Caveat: Has "salt-and-pepper" noise in background (division of small numbers in air).

Denoised (UniformDenoised)¶

Description: Bias-corrected with clean background.

Detection: - Exclusive flag: is_uniform_denoised - Keywords: uni-den, uniform denoised

Computation:

S_UNI-DEN = Re(INV1* · INV2) / (|INV1|² + |INV2|² + β)

The regularization term β suppresses background noise.

Use: Primary input for segmentation (FreeSurfer, FSL, SPM) - has clean dark background.

Synthetic MRI Constructs¶

Generated from quantitative maps (SyMRI/MAGiC).

SyntheticT1w¶

Detection: - Exclusive flag: has_t1_synthetic - Keywords: synthetic t1

Source: Computed from T1, T2, PD maps using Bloch equations.

SyntheticT2w¶

Detection: - Exclusive flag: has_t2_synthetic - Keywords: synthetic t2

SyntheticFLAIR¶

Detection: - Exclusive flag: has_flair_synthetic - Keywords: synthetic flair

Advantage: FLAIR contrast without additional scan time.

SyntheticPDw¶

Detection: - Exclusive flag: has_pd_synthetic - Keywords: synthetic pd

MyelinMap¶

Description: Myelin content / myelin water fraction.

Detection: - Exclusive flag: has_myelin - Keywords: myelin map, mwf

Use: White matter integrity assessment, MS research.

SWI Constructs¶

Susceptibility-weighted imaging outputs.

SWI (Processed)¶

Description: Magnitude × phase mask^n for maximum susceptibility sensitivity.

Detection: - Keywords: swi processed, swi combined - Combination: has_swi + is_derived

Note: SWI branch (branches/swi.py) handles this directly for SWI provenance.

Phase¶

Description: Phase image (wrapped or unwrapped).

Detection: - Exclusive flag: has_phase - Keywords: phase, phase map

Use: - QSM derivation - Iron vs calcium differentiation - Phase-contrast flow

Note: Can be original acquisition, not necessarily derived.

QSM (Quantitative Susceptibility Mapping)¶

Units: ppm (parts per million)

Detection: Keywords: qsm, susceptibility map, chi map

Computation: Derived from phase via dipole inversion.

Clinical use: Iron quantification, calcification detection.

Projection Constructs¶

Post-processing reformats.

MIP (Maximum Intensity Projection)¶

Detection: - Exclusive flag: is_mip - Keywords: mip, maximum intensity projection

Use: Angiography visualization, vessel overview.

MinIP (Minimum Intensity Projection)¶

Detection: - Exclusive flag: is_minip - Keywords: minip, minimum intensity projection

Use: Airway visualization, SWI vein display.

MPR (Multiplanar Reformation)¶

Detection: - Exclusive flag: is_mpr - Keywords: multiplanar reformat, reformatted

Note: Keyword "mpr" excluded to avoid false matches with "mprage".

Field Map Constructs¶

Calibration maps.

B0map¶

Description: Static field off-resonance map.

Detection: Keywords: b0 map, field map, fieldmap

Use: Distortion correction, shimming.

B1map¶

Description: RF transmit/receive field map.

Detection: Keywords: b1 map, b1+, b1-

Use: RF inhomogeneity correction.

Component Constructs¶

Complex data components.

Magnitude¶

Description: Magnitude from complex data.

Detection: Keywords only: magnitude image, magnitude only

Note: has_magnitude flag not used for detection - most images are magnitude.

Real¶

Detection: - Exclusive flag: has_real - Keywords: real image, real component

Imaginary¶

Detection: - Exclusive flag: has_imaginary - Keywords: imaginary image, imaginary component

Original vs Derived¶

Most constructs require is_derived=True in DICOM ImageType:

Requires Derived	Construct Examples
Yes	ADC, FA, CBF, T1map, SyntheticT1w, MIP
No	INV1, INV2, InPhase, OutPhase, Phase, Magnitude

Some outputs (like MP2RAGE inversions, Dixon echoes, phase images) are original acquisitions, not derived.

Common Output Examples¶

Acquisition	construct_csv
Standard T1w MPRAGE	`""` (empty - no constructs)
DTI with maps	`"ADC,FA"`
DSC perfusion	`"CBF,CBV,MTT"`
Dixon	`"Fat,Water"`
SyMRI/MAGiC	`"SyntheticFLAIR,SyntheticT1w,T1map,T2map"`
MP2RAGE	`"Denoised,T1map"` or `"INV1"`
TOF-MRA MIP	`"MIP"`
SWI	`"Phase,SWI"`

Confidence Levels¶

Detection Method	Confidence
Exclusive flag	95%
Keywords match	85%
Combination (AND)	75%

YAML Configuration¶

Constructs are configured in backend/src/classification/detection_yaml/construct-detection.yaml:

constructs:
  ADC:
    name: "ADC"
    category: "diffusion"
    description: "Apparent Diffusion Coefficient map"
    keywords:
      - "adc"
      - "apparent diffusion coefficient"
    detection:
      exclusive: has_adc
    requires_derived: true

  INV1:
    name: "INV1"
    category: "mp2rage"
    detection:
      exclusive: is_mp2rage_inv1
    requires_derived: false  # Original acquisition

rules:
  allow_multiple: true
  priority_order:
    - "ADC"
    - "FA"
    # ... more specific first

Convenience Methods¶

The detector provides helper methods:

# Get all constructs by category
detector.get_constructs_by_category("diffusion")  # ["ADC", "FA", "MD", "Trace", "eADC"]

# Get all categories
detector.get_categories()  # ["component", "diffusion", "dixon", "mp2rage", ...]

# Check if specific construct detected
output.has("ADC")  # True/False

# Get constructs by category from output
output.by_category("perfusion")  # List of ConstructMatch objects