MRI and fMRI

How magnets, radio waves and hydrogen atoms produce detailed brain images; how functional MRI tracks activity through blood oxygen; diffusion imaging of white matter; and the famous dead salmon that taught a statistical lesson.

How MRI works

Magnetic resonance imaging (MRI) exploits the fact that the nuclei of hydrogen atoms — abundant in the water and fat of the body — behave like tiny magnets. In a strong magnetic field (typically 1.5 or 3 tesla for clinical scanners, tens of thousands of times Earth's field) they align. Radio-frequency pulses knock them out of alignment; as they relax back, they emit signals whose timing depends on the surrounding tissue. Magnetic field gradients encode position, letting a computer reconstruct detailed images. Grey matter, white matter and fluid look different, so structural MRI reveals anatomy, tumours, strokes and MS lesions. Paul Lauterbur and Peter Mansfield shared the 2003 Nobel Prize for developing MRI. It involves no ionising radiation.

fMRI and the BOLD signal

In 1990 Seiji Ogawa and colleagues at Bell Labs showed that MRI signals change with the amount of oxygen in the blood: oxygenated and deoxygenated haemoglobin have different magnetic properties. When a brain region becomes active, local blood flow increases more than oxygen use, so the proportion of oxygenated blood rises and the MRI signal increases slightly. This blood-oxygen-level-dependent (BOLD) signal is the basis of functional MRI, first demonstrated in humans in the early 1990s.

fMRI typically has a spatial resolution of a few millimetres and, because blood flow responds sluggishly, a temporal resolution of seconds. It measures neural activity indirectly, through blood flow, and shows relative changes between conditions rather than absolute activity.

Diffusion imaging and connectivity

Diffusion MRI measures the movement of water molecules, which travel more easily along axons than across them. This allows tractography, reconstructing white matter pathways in living people (with known pitfalls). Resting-state fMRI studies correlations between regions' spontaneous fluctuations to map functional networks.

The dead salmon

In 2009 Craig Bennett and colleagues scanned a dead Atlantic salmon while "showing" it photos of people in social situations. Without proper correction for testing tens of thousands of brain locations at once, a few voxels in the fish's brain appeared "active". The tongue-in-cheek poster made a serious point about multiple comparisons, and it helped push the field toward stricter statistics. Later concerns about small sample sizes and reproducibility have led to larger, pre-registered and shared datasets.