Lingo
← Back to Neuroscience

Magnetic Resonance Imaging (MRI)

Anatomy, tissue microstructure, and blood-oxygenation-level contrast — millimetre spatial resolution in exchange for the poorest temporal resolution of any method here.

The signal

MRI works by aligning hydrogen nuclei in a static field B₀ (typically 1.5 T, 3 T, or 7 T), perturbing them with radiofrequency pulses, and listening to how they relax. Two time constants define the contrast: T1 (spin lattice, tissue-specific) and T2 (spin-spin, always shorter than T1). Grey and white matter differ in the ratio of the two, which is why nearly every sequence is a variation on the same physics.

The practical consequence is that the signal is inferred: a voxel value is a fitted parameter, not a measurement. That is why sequence parameters, field homogeneity, and the statistical threshold belong in a methods section alongside the claims they support.

Structural contrasts

  • T1-weighted (MPRAGE): grey matter brighter than white; the workhorse for segmentation, cortical surface reconstruction, and volume measurement.
  • T2 and FLAIR: fluid-bright; FLAIR nulls CSF to reveal periventricular lesions and oedema.
  • Diffusion-weighted imaging (DWI/DTI): the contrast is Brownian motion of water, giving white-matter tractography, fractional anisotropy, and mean diffusivity. This is the modality that makes structural connectivity measurable.
  • T2*/SWI: sensitive to paramagnetic substances — iron, deoxyhaemoglobin, veins, microbleeds.
  • MR spectroscopy: metabolite concentrations (N-acetylaspartate, choline, creatine, GABA) as markers of integrity and metabolism.

BOLD fMRI

Functional MRI does not measure neural activity. It measures the blood-oxygenation-level-dependent signal: the relaxation time of water protons in and around capillary beds. Neural activity drives local blood flow that overshoots oxygen extraction, raising local oxyhaemoglobin and shifting T2*.

Every step of that chain is a compromise. The signal change is only 1–2% against a noisy baseline. Spatial resolution is excellent (1–3 mm), but that resolution is bought with the capillary blurring that makes any precise cytoarchitectonic claim implausible. Temporal resolution is around a 1 s repetition time, and the underlying haemodynamic response rises over ~4–6 s and lags the neural event by 4–8 s. BOLD is a slow, blurred shadow of a fast electrical event.

This does not weaken fMRI — it defines its claims. BOLD is well suited to sustained state differences and to timing relative to other BOLD signals, and poorly suited to single-neuron firing, millisecond latency, or differences in amplitude between two small regions.

What can go wrong

  • Head motion dominates variance and is correlated with task conditions by design (stimulus onsets move heads). It is the first thing to check when a result fails to replicate.
  • Reverse inference: activation in a region does not establish that region's function. A region lights up for a consequence of a process, not the process itself.
  • Vascular reactivity varies across regions and with age, drugs, anaesthetic, and vascular disease, which can change the BOLD signal without changing neural activity. A drug that alters the coupling between neural activity and blood flow can produce a pharmacologically induced BOLD change with no neural effect at all.
  • Multiplicity: a whole-brain analysis tests hundreds of thousands of voxels. Uncorrected thresholds report false positives by design; report the correction method alongside the map.

Why MRI anchors the other modalities

EEG and MEG both need a template head built from a structural MRI, defining each subject’s cortical folding, and MRI supplies the priors that make the ill-posed inverse problem tractable. Simultaneous EEG-fMRI and MEG-fMRI work for exactly this reason: each constrains the other’s weaknesses, trading a shared acquisition cost for a much stronger inference.

Structural MRI also provides the ground truth for lesion studies and the anatomy behind tractography, where diffusion constrains a macroscopic connectivity hypothesis and not a monosynaptic one.

References

  1. Ogawa, Honda, Kawai et al. (1990) — Proceedings of the National Academy of Sciences— The original demonstration of intrinsic BOLD contrast in the human brain during activation.
  2. Logothetis (2008) — Nature Reviews Neuroscience— What we can and cannot do with fMRI: the physiology of the BOLD signal and the limits of the inferences it supports.
  3. Poldrack (2006) — Trends in Cognitive Sciences— Why activation maps do not license the reverse inference, with a concrete predictive-code framing.
  4. Beaulieu (2002) — The Anatomical Record— Diffusion-weighted imaging as a neuroanatomical tool, including what tractography can and cannot establish.

Examples