Basic Neurosciences · Paper A
Functional MRI (fmri)
Start a stopwatch the moment a patch of cortex fires harder, and the scanner sees nothing for four seconds. What it eventually records is not electrical activity but a change in how much of the haemoglobin in that patch is carrying oxygen. Firing burns oxygen, the local vessels then over-supply the patch, and the ratio of oxygenated to deoxygenated haemoglobin climbs; because those two forms behave differently in a magnetic field, an MRI scanner can read that climb. The reading is called the BOLD signal, and every strength and every weakness of functional MRI follows from the delay between the firing and the reading.
Any mention of blood oxygenation or of the BOLD signal while a subject performs a cognitive task puts you here, as does the mapping of eloquent cortex before neurosurgery, and as does a spontaneous low-frequency signal correlated between separated regions with no task at all — the first two task-based, the third resting-state. Stems that contrast excellent spatial with poor temporal resolution, or that note a several-second lag between stimulus and peak signal, are testing the haemodynamic response specifically.
fMRI vs EEG — fMRI resolves a few millimetres in space but seconds in time, lagged by the haemodynamic delay; EEG resolves milliseconds in time but localises poorly because the signal smears through the skull.,fMRI vs structural MRI — fMRI reads the BOLD signal to show which regions were active; structural MRI shows anatomical detail and carries no functional information.,Task-based vs resting-state fMRI — task-based compares an active period against a control condition; resting-state uses no task and measures spontaneous correlated activity between regions.,BOLD signal vs direct electrical recording — BOLD is a delayed vascular and metabolic consequence of firing, not a recording of the firing itself as EEG or single-unit recording would be.,fMRI vs PET — fMRI needs no radioactive tracer and resolves better in both space and time, while PET can quantify metabolism or receptor binding directly, which fMRI cannot do at all.,fMRI vs DTI — DTI maps the orientation and integrity of white-matter tracts, showing wiring rather than activity, and produces no activation map.
The error that costs most marks is reading the BOLD signal as neural firing itself. It is a vascular consequence of firing, several seconds behind it, which is precisely why a plain block design cannot sequence events happening close together in time — and that limitation is what stems asking about the main weakness of the technique are after. Watch out too for the assumption that resting-state scanning needs a task to compare against; it does not, and the absence of a task is the entire point of the design. When several modalities are listed together, read what each one measures rather than the name of the machine, because structural MRI, DTI and PET all share either hardware or vocabulary with functional MRI while measuring something else.
Work through two questions in order. Is a task being performed and compared against a control condition, which makes it task-based, or is a network of correlated regions described with no task, which makes it resting-state? Then, is the described signal electrical, pointing to EEG, or indirectly vascular through oxygenation, pointing here, or tracer-based with an absolute metabolic rate, pointing to PET? A third question comes up occasionally: if asked whose guidance governs the use of this scan, treat it as a question about method rather than treatment. Choosing to scan a patient before neurosurgery sits with the operating team's own protocol and the multidisciplinary meeting, and no national psychiatry guideline addresses it.
Imagine starting a stopwatch at the moment a small patch of your motor cortex begins firing harder because you have started tapping a finger. Nothing about that firing is visible to a scanner. What follows over the next several seconds is visible, and functional MRI is built entirely out of it. Before the technique existed, the only way to learn which part of a living brain did what was to wait for a stroke or a tumour to take it away and see what stopped working. Following the clock from that first burst of firing through to a coloured blob on a brain slice is the whole of this page.
Say what the machine measures, and nothing more. It records how much of the haemoglobin in a small cube of your brain tissue is carrying oxygen, relative to how much is not. The name for that measurement is the BOLD signal, short for blood-oxygen-level-dependent. Firing is never measured. The oxygen bookkeeping that follows firing is.
By the time the recorded signal peaks, the neurons that caused it have fired and fallen quiet again. The peak lands about four to six seconds after the neural event, and the curve can take twenty seconds or more to return to baseline. Everything you will hear praised about this technique, and everything you will hear held against it, follows from that single gap: millimetres of spatial detail, seconds of temporal blur.
At second zero the neurons fire and burn oxygen, so the local supply dips for an instant. Within a second or two the vessels feeding that patch dilate and flow rises — and here is the part that will surprise you, the increase overshoots what the tissue is using. More oxygenated blood arrives than gets consumed, so the ratio of oxygenated to deoxygenated haemoglobin in the working patch climbs, even though consumption has climbed too.
That ratio is what the magnet notices. Deoxyhaemoglobin is paramagnetic (it distorts the local magnetic field and drags MRI signal down), while oxyhaemoglobin is diamagnetic (a much weaker, opposite behaviour that leaves the field close to undisturbed). Fewer deoxygenated molecules per cube means less distortion, so signal from that cube rises. The whole inference runs like judging which office in a building was busiest by checking whose supply cupboard was restocked that evening: informative, and always late.
Second 0, firing. Seconds 1 to 2, flow rises and overshoots. Seconds 4 to 6, the signal peaks — and throughout, the scanner has been photographing the whole head every two seconds or so, one volume of voxels (small cubes of tissue, each treated as a single measurement) at a time. Seconds 10 to 20 and beyond, the curve falls away, often dipping a little below where it started before settling.
You design your study around that curve. A block design has the participant tap a finger or name pictures for twenty or thirty seconds, rest for a similar stretch, and compares the two; with a curve this slow, blocks spaced that far apart separate easily. An event-related design goes finer, using single brief trials jittered deliberately in time with the response shape modelled rather than assumed, which pulls apart neural onsets around a second apart. Poor temporal resolution is real, but it is not a fixed wall.
Take the task away entirely and the curve carries on wandering. Resting-state scanning records those spontaneous swings, which are slow: conventionally below about 0.1 hertz, meaning roughly six cycles a minute or fewer. Regions whose swings rise and fall together count as functionally connected. The best-studied set is the default mode network — medial , posterior cingulate and precuneus, and angular gyrus — busy when attention drifts and suppressed the moment an external task takes hold.
| Technique | What it measures | Where it wins | Where it loses |
|---|---|---|---|
| Functional MRI | Blood oxygenation after activity | Millimetres of localisation | Seconds of lag |
| EEG | Electrical activity at the scalp | Millisecond timing | Localises poorly; signal smears through the skull |
| PET | Tracer uptake: metabolism or receptor binding | Direct quantification of metabolism | Radioactive tracer; coarser in space and time |
| Structural MRI and DTI | Anatomy; white-matter tract orientation | Fine structural detail | No activity shown at all |
Read what each machine measures rather than what it is called, and the four separate cleanly for you whatever order a stem lists them in.
Nothing here diagnoses a disease. The job done before a real operation is localisation: with a tumour or an epileptic focus sitting near language or motor territory, your patient performs standard naming and movement tasks in the scanner so the team can see where eloquent cortex (tissue whose removal would leave a disabling deficit) lies relative to the lesion.
What the scan changes for your patient is the resection plan, not the prescription. The surgical team draws its margin around the mapped language and motor areas, and where the map is ambiguous or the lesion sits directly on eloquent tissue, the plan may shift towards awake craniotomy with intraoperative direct cortical stimulation, which is still the reference standard the scan is validated against. The map is read alongside the rest of the workup rather than on its own, because head movement, a patient who cannot perform the task, and vascular pathology close to a lesion all distort it.
One word in the stem usually settles your answer: oxygenation or BOLD means this technique; millisecond or electrical means EEG; radioactive tracer or absolute metabolic rate means PET; correlated regions with no task means the resting-state design specifically. Quick check before moving on — can you explain why the technique with the finest spatial detail has the worst timing?
Functional MRI maps brain activity indirectly, through the delayed rise in blood oxygenation that follows neural firing: millimetre localisation, a lag of several seconds, and one routine clinical job, which is mapping eloquent cortex before neurosurgery.