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