MRI and multimodality foundations — reference
Reference · The sequence-family table, modality orientation, and further reading. Teaching versions live in the lessons.
Sequence families
| Family | Idea | Typical use |
|---|---|---|
| Spin echo (SE) | 90° excitation + 180° refocus | clean T1/T2 contrast; the reference |
| Turbo / fast spin echo (TSE/FSE) | many 180° refoci per excitation → faster | routine T1/T2/PD |
| Gradient echo (GRE) | no 180° refocus; flip angle < 90°; faster, T2*-sensitive | dynamic, 3D, susceptibility |
| Inversion recovery (IR) | 180° inversion pulse before excitation; TI nulls a tissue | fat suppression, FLAIR |
| STIR | IR that nulls fat | fat-suppressed, edema-sensitive |
| Fat suppression | reduces bright fat (spectral sat, DIXON, STIR) | T2 fat-sat, STIR |
T1-weighted: short TR/TE, fat bright, fluid dark. T2-weighted: long TR/TE, fluid bright. PD-weighted: long TR/short TE. “Weighted” ≠ measuring T1/T2.
The signal model (lesson 2)
B0 polarises along z; RF pulse tips into transverse (flip angle; 90° for SE); two simultaneous relaxations: T1 (longitudinal recovery, spin–lattice) and T2 (transverse decay, spin–spin). T2* adds macroscopic inhomogeneity (GRE/BOLD). TR drives T1 weighting; TE drives T2 weighting.
Diffusion (lesson 5)
DWI sensitises to microscopic water motion via the b-value. ADC from ≥2 b-values:
S(b)=S0·exp(−b·ADC), units mm²/s. Restricted (low ADC) → cellularity (tumour,
stroke); free (high ADC) → fluid. “Quantitative-ish”: parameter/distortion-sensitive.
k-space (lesson 6)
MRI acquires spatial frequencies (k-space) → inverse Fourier → image. Centre = contrast/gross shape; edges = detail/resolution. Extent → resolution; density → FOV. Sampling is a choice (acceleration, parallel imaging, compressed sensing, DL recon) — reconstruction changes quantitative values.
Domain shift axes (lesson 7)
field strength · vendor · coil · sequence/params (TR/TE/flip/TI) · protocol · reconstruction · scanner/site. MRI is the worst case for shift (non-physical intensity × many axes). Document, test, harmonise only where defensible, scope claims.
Multimodality orientation (where CT/MRI sit)
| Modality | Measured | Intensity ~means | Dominant quant/geometry issue | Typical AI tasks |
|---|---|---|---|---|
| CT | x-ray attenuation | HU (physical scale) | reconstruction/kernel dependence (Ch. 1) | seg, detection, quantification |
| MRI | RF from protons | relative, sequence-dependent | intensity normalisation, harmonisation (this ch.) | seg, synthesis, quantification |
| PET/CT | radiotracer uptake (+ CT AC) | SUV, semi-quantitative | attenuation correction, SUV reproducibility (QIBA) | uptake quantification, seg |
| PET/MRI | tracer uptake + MRI anatomy | SUV + MRI contrast | MR-based attenuation correction | quantification, seg |
| Ultrasound | acoustic echoes | relative, probe/angle-dependent | speckle, no fixed geometry, operator-dependent | seg, measurement, guidance |
| Radiography (XR) | x-ray projection (2D) | relative attenuation, projected | 2D projection, magnification | detection (CXR), classification |
| Digital pathology (adjacent) | stained tissue, transmitted light | stain-dependent colour | stain variation, very large WSI tiles | detection, grading |
Orientation only — do not dilute the CT/MRI focus. PET/CT matters because QIBA SUV profiles (Ch. 3) are the canonical quantitative-imaging example, and hybrid imaging shares the geometry lessons of Chapter 2.
Geometry (reuses Ch. 2)
MRI still lives in physical patient (LPS) space with spacing, orientation, origin, direction, FoR — all of Chapter 2 applies. MRI-specific: more motion/distortion (breathing, peristalsis, EPI/DWI); multi-sequence (T1/T2/DWI/post-contrast) are different series that must be co-registered before voxel-wise comparison.
Read / watch
- Stanford Medicine BMR — MRI Physics Education Materials —
https://med.stanford.edu/bmrgroup/education/mri-physics.html (
VERIFIED; university source, free for education). Sections: Overview, Image Formation (k-space), Resolution/FOV, Contrast Mechanisms (T1/T2, fat sat/IR, diffusion), Pulse Sequences (SE/GRE/magnetisation-prepared). - A recognised MRI physics text for technical readers (Hashemi/Bradley MRI: The Basics, or the MRI chapters of Bushberg Essential Physics of Medical Imaging).
- mriquestions.com — practical MRI education reference (verify pages at point of use).
- 3D Slicer Training Compendium — https://training.slicer.org/ — DICOM/MRI loading with sample data.
- TotalSegmentator MRI — Akinci D’Antonoli et al., Radiology 2025;314(2):e241613, PMID 39964271.
- OmniMRI — He et al., arXiv:2508.17524 (preprint) — unified MRI foundation model.
What to retain (chapter summary)
- MRI intensity is relative and sequence-dependent, not a physical scale — this drives MRI radiomics fragility and harmonisation difficulty.
- TR/TE set T1/T2 weighting; “weighted” ≠ a direct measurement of T1/T2.
- ADC (from DWI/b-values) is one of the more directly quantitative routine MRI maps, but still parameter-sensitive.
- Pre→post-contrast synthesis asks the model to invent enhancement — different problem from denoising/translation.
- MRI is particularly vulnerable to domain shift (field/vendor/coil/sequence/recon); geometry still uses all of Chapter 2.
- Orient across modalities: each has a different “what intensity means” and a different dominant quant/geometry issue.