How Does an MRI Scanner See Inside the Body? | How It Works

What is the short answer?

An MRI scanner makes pictures from the hydrogen nuclei in the body's water and fat. Each nucleus is a single proton that behaves like a tiny magnet. In a strong magnetic field, slightly more of them line up with it than against it.

A pulse of radio waves tips this alignment over. As the protons settle back, they give off a faint radio signal, which coils pick up and a computer turns into an image. Tissues settle at different rates, so they look different.

Definition: Magnetic resonance imaging (MRI) builds images from radio signals given off by atomic nuclei, mostly hydrogen, held in a strong magnetic field and excited by radio waves at their resonant frequency.

How do hydrogen nuclei act as tiny magnets?

A proton has a property called spin, which gives it a small magnetic moment, like a bar magnet. The US Food and Drug Administration (FDA) says the signal in an MR image comes mainly from protons in fat and water molecules.

Outside a scanner, the protons point every which way, so their magnetism cancels. In a strong field, called B₀ and measured in tesla (T), a small excess lines up with it, forming the net magnetisation.

Each proton also wobbles around the field like a top, a motion called precession. Its rate is the Larmor frequency, f = γB₀ ÷ 2π, where γ is the gyromagnetic ratio. For the proton, the US standards body NIST gives γ ÷ 2π as 42.577 MHz per tesla.

Magnetic resonance in liquids and solids was first reported in 1946, independently, by Felix Bloch's team, working with water, and Edward Purcell's, working with paraffin wax. Both shared the 1952 Nobel Prize in Physics.

How does a radio pulse make the body send back a signal?

The scanner sends a short radiofrequency (RF) pulse at the Larmor frequency. Only protons precessing at that frequency take up its energy; this matching is resonance. The pulse tips the net magnetisation away from B₀, so it sweeps around the field and induces a voltage in a receiver coil. Then the protons relax in two ways:

• T1 relaxation (longitudinal): the magnetisation along the field grows back. An Open University course defines T1 as the time to recover about 63% of the way.
• T2 relaxation (transverse): the protons fall out of step, so the signal fades. T2 is the time to fall to about 37%, says Questions and Answers in MRI, a site by the radiologist Allen D. Elster.

Table: Approximate relaxation times of tissue protons at 1 tesla

The table comes from the same course; Elster notes that T1 rises with field strength for most tissues, while T2 changes much less.

Why do tissues look different on a scan?

The timing of pulses and listening decides which relaxation time controls brightness, and so the contrast between tissues. On a standard T1-weighted image, fat, with its short T1, looks bright and fluid looks dark. On a standard T2-weighted image, fluid, with its long T2, looks bright. Elster notes these rules are not universal laws.

MRI contrast agents often contain gadolinium, a rare earth metal, and are usually given through a vein, the FDA says. It speeds the relaxation of nearby water, so tissues it reaches stand out. Regulators have studied traces left behind:

• In July 2017, the European Medicines Agency confirmed suspensions and restrictions of some "linear" agents, saying there was no evidence that brain deposits had harmed patients.
• In December 2017, the FDA required a new class warning. It said retention had not been directly linked to harmful effects in patients with normal kidney function, and that the agents' benefit outweighed potential risks.

Functional MRI (fMRI) tracks brain activity through the BOLD (blood oxygen level dependent) signal. Haemoglobin without oxygen weakens the signal, Elster explains, so active areas, flushed with oxygen-rich blood, grow brighter.

How does the scanner work out where each signal came from?

In a uniform field, all protons share one frequency, so a coil cannot tell where a signal started. Gradient coils add a small extra field that grows steadily across the body, so the Larmor frequency changes with position.

• Slice selection: a gradient during the RF pulse excites one thin slice.
• Frequency encoding: a gradient during read-out gives each column its own frequency.
• Phase encoding: a brief earlier gradient shifts each row's timing, or phase, over many repeats.

With an illustrative gradient of 10 millitesla per metre, two points 10 cm apart differ by 1 mT, so their frequencies differ by about 42.6 kHz. A Fourier transform splits the mixed signal into its frequencies, as the student's ear picks notes from a chord.

The Nobel Assembly credits Paul Lauterbur with finding that gradients could make a two-dimensional picture, and Peter Mansfield with showing how to analyse the signals and image very fast. They shared the 2003 Nobel Prize in Physiology or Medicine.

What is inside an MRI scanner?

• The main magnet. Standard hospital scanners use a superconducting magnet: coils of wire with no electrical resistance when chilled by liquid helium. A 2026 study in NMR in Biomedicine puts the helium in a standard 1.5 T or 3 T scanner at 700 to 1,700 litres.
• Field strength. France's atomic energy commission, the CEA, puts hospital machines at 1.5 and 3 T. In October 2017, the FDA cleared the first 7 T scanner for clinical use in the United States, for the head, arms and legs. In April 2024, the CEA released the first images of living human brains from Iseult, an 11.7 T research scanner cooled to −271.35 °C by 7,500 litres of helium.
• RF coils send the pulses and pick up the signal.
• Gradient coils switch rapidly; the FDA says these changing fields cause loud knocking noises.
• The computer times the pulses and gradients and rebuilds the image.

How is an MRI scan kept safe?

MRI uses no ionising radiation, unlike X-rays and CT, as the FDA and the 2003 Nobel announcement note.

• Projectiles. The FDA warns that the static field attracts magnetic objects, which can become projectiles.
• Implants. The FDA says patients with implanted devices should not be scanned unless the device is identified as MR Safe or MR Conditional; unknown devices count as MR Unsafe. NIBIB, a US research institute, names pacemakers as one example.
• Heating. The specific absorption rate (SAR) is the RF power absorbed per kilogram. Elster gives the whole-body limits of the standard IEC 60601-2-33 (2022 edition): 2 watts per kilogram in normal mode and 4 in "first level" mode. The FDA says second-degree burns are the most commonly reported patient problem.
• Noise. NIBIB says some scanners reach 120 decibels; the NHS gives patients earplugs or headphones.

Where are scanners found, and what is debated?

The OECD's Health at a Glance 2025 finds that Japan has by far the most MRI units per person among its members, and the United States the second most. The OECD and WHO's Asia/Pacific 2024 report gives Japan 57 units per million people, while six countries, among them Pakistan and Bangladesh, report under one. In India, the Press Information Bureau said in March 2025 that the magnet of an indigenous scanner by SAMEER, a government lab, had reached 1.5 T, and that AIIMS New Delhi would clinically validate the system.

• Helium. The 2026 NMR in Biomedicine study calls supply highly concentrated, citing US Geological Survey figures of about 42% from the United States and 33% from Qatar. A 2025 Cureus review says sealed magnets need only single-digit litres, and rates cost-effectiveness reports as trending positive but insufficient for firm conclusions.
• Waiting times. The Royal College of Radiologists (RCR) reported that in September 2025 over 74,000 people in England had waited too long for a CT or MRI scan; the standard is at most 1% waiting six weeks or longer. It cited a 30% shortfall of radiologists.
• Cost and use. The OECD says there is no benchmark for the ideal number of scanners: too few may mean long distances or waits, and too many may mean overuse of costly procedures with little if any benefit for patients.
• Incidental findings. A 2007 study scanned 2,000 people aged 45 and over from the Rotterdam Study in the Netherlands, finding silent brain infarcts in 7.2%, aneurysms in 1.8% and benign tumours in 1.6%. Its authors called for research on their natural course.

How does this work with real numbers?

Both examples are illustrative, with constants from NIST.

What frequency does a scanner use, and why is it radio?

Table: Larmor frequencies and free-space wavelengths (λ = c ÷ f)

Waves metres long are radio waves; an illustrative 500 nanometre green light wave is about 9.4 million times shorter than the 1.5 T one.

How many protons line up with the field?

A proton's two states differ in energy by ΔE = γħB₀, with γ ≈ 2.675 × 10⁸ radians per second per tesla. At body temperature, 310 K, the excess fraction lined up is approximately ΔE ÷ 2kT, where k is the Boltzmann constant.

• 1.5 T: ΔE ≈ 4.23 × 10⁻²⁶ J and 2kT ≈ 8.56 × 10⁻²¹ J, so the excess is about 4.9 × 10⁻⁶, or 5 per million.
• 3 T: double, about 9.9 × 10⁻⁶, or 10 per million.

Elster gives about 3 per million at 1.0 T, which agrees. The signal works because the numbers are vast: one gram of water holds about 6.7 × 10²² hydrogen nuclei (two per molecule), so the excess at 1.5 T is still about 3 × 10¹⁷. Doubling the field doubles the excess, one reason stronger magnets give more signal.

Where can the student go deeper in World 101?

This page follows a scan from a single spinning proton to a finished picture. The World 101 subject Biomedical Engineering places MRI among the many machines, devices and materials that engineers design for medicine, and looks at how such tools are developed, tested and approved.

Three related subjects take these ideas further. Quantum Physics explains spin and why a proton in a magnetic field has only two energy states. Neuroscience shows what researchers learn about the working brain from methods such as functional MRI. Healthcare Systems & Access asks who can reach scarce, costly equipment, and how long they wait for it.

Sources

• NIST, CODATA 2022 values of the constants used.
• A. D. Elster, Questions and Answers in MRI, pages on NMR's discovery (Physical Review, 1946; Nobel Prize, 1952), relaxation, Boltzmann distribution, BOLD and SAR (IEC 60601-2-33, 2022), read 9 November 2026.
• The Open University, OpenLearn, "Metals in medicine", read 9 November 2026.
• Nobel Assembly at Karolinska Institutet, press release, 6 October 2003.
• US FDA: MRI pages, 29 August 2018 and 9 December 2017; 7 T clearance, 12 October 2017; gadolinium communication, 19 December 2017.
• EMA, linear gadolinium agents opinion, 21 July 2017.
• NIBIB, "Magnetic Resonance Imaging (MRI)", read 9 November 2026.
• NHS, "MRI scan", reviewed 11 September 2025.
• CEA, Iseult release, 2 April 2024.
• H. Ayyildiz and S. M. Erturk, NMR in Biomedicine, August 2026 (citing USGS, 2026).
• M. M. Alenezi and others, Cureus, 11 November 2025.
• OECD, Health at a Glance 2025, November 2025; OECD and WHO, Health at a Glance: Asia/Pacific 2024, November 2024.
• Press Information Bureau, India, SAMEER and AIIMS release, 25 March 2025.
• Royal College of Radiologists, 1 December 2025 (September 2025 data).
• M. W. Vernooij and others, New England Journal of Medicine, 2007.

Common errors and misconceptions

• Misconception: MRI takes pictures with X-rays. Correct: It uses a magnetic field and radio waves, with no ionising radiation, the FDA notes.
• Misconception: In the scanner, all the body's protons line up with the field. Correct: Only a tiny excess does, about 5 in a million at 1.5 T.
• Misconception: The loud knocking comes from the main magnet. Correct: It comes from the gradient coils, whose fields the FDA says change with time.
• Misconception: A bright area always means the same tissue. Correct: Brightness depends on the weighting; on standard sequences, fluid is dark on T1-weighted images but bright on T2-weighted ones.
• Misconception: The receiver coil can tell where each signal came from. Correct: Gradient coils encode position in the frequency and phase of the signal.

Key takeaways

• Hydrogen nuclei in water and fat act as tiny magnets; in a strong field a small excess lines up with it.
• Protons precess at the Larmor frequency, f = γB₀ ÷ 2π, about 42.58 MHz per tesla: 63.9 MHz at 1.5 T and 127.7 MHz at 3 T.
• An RF pulse at that frequency tips the magnetisation; tissues then differ in T1 and T2 relaxation, which gives contrast.
• Gradient coils make the frequency and phase of the signal depend on position, and a Fourier transform turns the signals into an image.
• A scanner combines a helium-cooled superconducting magnet, RF coils, gradient coils and a computer; hospital machines usually run at 1.5 or 3 T.
• MRI uses no ionising radiation; helium, waiting times, cost and incidental findings are debated, with the OECD, the RCR and researchers each stating findings.

Test yourself

Why does an MRI scanner use radio waves rather than light?

The pulse must match the protons' Larmor frequency, about 63.9 MHz at 1.5 T, which is a radio wave about 4.7 m long, not light.

How do T1 and T2 relaxation differ?

T1 is the regrowth of magnetisation along the field; T2 is the signal fading as protons fall out of step.

Why does cerebrospinal fluid look bright on a T2-weighted image?

Its T2 is long, about 150 ms at 1 tesla, so its signal fades slowly.

How do gradient coils locate a signal?

They make the frequency and phase of precession vary with position, which a Fourier transform decodes.

Why is the excess of aligned protons twice as large at 3 T?

The fraction is about γħB₀ ÷ 2kT, so it grows in step with B₀.

When does the FDA say a patient with an implant can be scanned?

Only when the device is identified as MR Safe or MR Conditional; a device of unknown status is treated as MR Unsafe.