Functional MRI (fMRI) detects changes in blood oxygenation in the brain, not electrical activity directly. When a brain region becomes active, blood flow to that area increases, delivering more oxygen than the neurons actually consume. The fMRI scanner picks up this shift in oxygenation and uses it to create maps of which parts of the brain are working during a given moment or task.
The BOLD Signal: What the Scanner Actually Measures
The signal fMRI relies on is called BOLD, which stands for Blood Oxygen Level Dependent. It works because of a basic physical property of hemoglobin, the oxygen-carrying molecule in your blood. When hemoglobin is carrying oxygen, it’s essentially non-magnetic. When it releases that oxygen, it becomes weakly magnetic, with four unpaired electrons per iron atom. This difference in magnetic behavior between oxygenated and deoxygenated hemoglobin is what the MRI scanner can detect.
When neurons in a brain region ramp up their activity, the body responds by flooding that area with extra oxygenated blood, well beyond what the neurons need. This oversupply means the local concentration of deoxygenated hemoglobin actually drops, even though the neurons are consuming more oxygen than usual. The BOLD signal increases when deoxygenated hemoglobin decreases. So a “bright spot” on an fMRI image represents a region where blood is more oxygenated than its surroundings, a sign that the area recently became active.
A common misconception is that BOLD measures how much oxygen neurons are burning. It doesn’t. It reflects the overoxygenation that happens when blood flow surges to meet (and exceed) local demand. The active process linking neuronal activity to this blood flow increase is called neurovascular coupling, and it’s the biological bridge that makes fMRI possible.
How Close Is It to Measuring Brain Activity?
Because fMRI tracks a metabolic consequence of brain activity rather than the electrical signals themselves, there’s always the question of how well it matches what neurons are actually doing. Research combining fMRI with direct electrical recordings from the brain has shown that the BOLD signal correlates most closely with local field potentials, which reflect the collective electrical input arriving at a group of neurons. It does not reliably track the firing of individual neurons, which represents their output.
In studies of human temporal cortex, researchers found significant overlap between fMRI signal increases and increases in high-frequency local field potentials (in the 50 to 250 Hz range). But they found no independent relationship between fMRI signals and the firing rate of individual neurons. This means fMRI is better understood as a measure of how much processing a brain area is receiving, not a count of how many neurons are firing.
The Time Lag
One major limitation is speed. Neural events happen in milliseconds, but the blood flow response that fMRI detects is far slower. After a stimulus, there’s a brief initial delay of under two seconds, followed by a peak in the BOLD signal roughly 6 to 7 seconds after the stimulus begins. The signal then dips below baseline before returning to normal about 15 seconds after the event. This means fMRI has a temporal resolution measured in seconds, not milliseconds. It can tell you which brain region was active during a task, but it can’t capture the rapid, moment-to-moment sequence of neural processing the way EEG or MEG can.
How Sharp Is the Image?
Spatial resolution depends on the strength of the scanner’s magnet. A standard clinical or research scanner operating at 3 Tesla typically captures brain activity at a resolution of about 2 millimeters per voxel (the 3D equivalent of a pixel). More powerful 7 Tesla scanners, found in specialized research centers, can push that down to about 1.5 millimeters. Both typically take a full snapshot of the brain every 2 seconds.
Signal quality is affected by more than just magnet strength. Physiological noise from your heartbeat and breathing, even small head movements inside the scanner, can degrade the image. Scanner hardware, the choice of receiver coils, and how the scan is set up all influence the signal-to-noise ratio. Researchers and technicians spend considerable effort minimizing these sources of interference.
Task-Based vs. Resting-State Scans
There are two main ways fMRI is used, and they detect different things. In task-based fMRI, you’re asked to do something specific inside the scanner: tap your fingers, look at pictures, listen to words, solve a problem. The scanner captures which brain regions show increased blood oxygenation in response to that task. The resulting signal closely tracks the timing and frequency of the task design.
In resting-state fMRI, you lie still and do nothing in particular. The scanner picks up slow, spontaneous fluctuations in blood oxygenation across different brain regions. Regions that fluctuate in sync are considered functionally connected, forming what researchers call resting-state networks. The most well-known is the default mode network, a set of brain areas that tend to activate together when you’re not focused on any external task, often associated with mind-wandering and self-referential thought. These resting-state signals tend to be dominated by lower frequencies compared to the sharper, task-locked patterns seen during active tasks.
Clinical Uses
The most established clinical application of fMRI is presurgical brain mapping. Before removing a brain tumor or treating epilepsy surgically, neurosurgeons need to know exactly where critical functions like language, movement, and vision are located in that specific patient’s brain. Everyone’s functional anatomy is slightly different, and a tumor can push or reorganize nearby functional areas. fMRI is currently the most commonly used non-invasive tool for this kind of surgical planning, helping surgeons identify which tissue they can safely remove and which they need to preserve.
Beyond surgery, fMRI is widely used in research to study everything from how the brain processes pain and emotion to how neurological and psychiatric conditions alter brain function. It has contributed to mapping the brain networks involved in memory, decision-making, addiction, and dozens of other cognitive processes. While it doesn’t read thoughts or diagnose mental illness from a single scan, it provides a window into which brain regions and networks are engaged during specific mental states, something no other non-invasive tool does with the same combination of spatial detail and whole-brain coverage.