How an MR scanner is built: four subsystems

Split any MRI complaint into one of four subsystems - magnet, gradients, RF chain, or computer and reconstruction - and choose the first thing to check on site before touching a cabinet.

The scanner as four subsystems

An MR scanner looks like one big white ring with a table, but for the service engineer it is four separate systems that share a room and a timing controller. Sort a complaint into one of them first and the search narrows quickly. Treat the scanner as a black box and you end up chasing symptoms from cabinet to cabinet and replacing boards that were never faulty.

The four are: the magnet, which makes the strong, uniform static field called B0; the gradient system, which adds small, fast-switched field slopes along X, Y and Z so that the signal carries position information; the RF chain, which transmits a radio pulse to excite hydrogen nuclei and then receives their very weak echo; and the computer and reconstruction chain, which runs the sequence, collects the raw data and turns it into images, stores them and sends them to the hospital network.

The physics connects them. B0 makes hydrogen nuclei precess at the Larmor frequency, about 42.58 MHz per tesla, so roughly 64 MHz at 1.5 T and 128 MHz at 3 T. The RF transmitter must hit exactly that frequency to excite them. The gradients shift the local frequency and phase depending on position. The receive chain picks up the signal, and the reconstructor uses a Fourier transform to sort it back into pixels. A fault at any of these steps shows up in the image, and each leaves a different pattern.

Underneath all four is infrastructure, which is not part of the scanner but breaks it most often: mains power and the UPS, the chiller and the water circuits, room temperature and humidity, and the RF-shielded room itself.

The magnet: the part that never switches off

A clinical superconducting magnet is a coil of superconducting wire held in liquid helium at about 4.2 K inside a cryostat. The cryostat is a vessel inside a vessel: a helium bath, one or more thermal shields, and a vacuum space that stops heat from reaching the helium by conduction and convection. Once the coil is energised at installation, current flows in it without loss, and the field stays on day and night, with the scanner software shut down and with the building power off. Treat every magnet as live, always.

Heat still leaks in. A cryocooler - the cold head mounted on top of the magnet - removes it and, on modern near-zero-boil-off magnets, re-condenses evaporating helium back into the bath. The cold head is driven by a helium compressor in the technical room, which in turn needs electrical power and cooling water, usually from the chiller. An adsorber cartridge in the compressor keeps the helium gas clean. That makes the magnet a chain: power and water feed the compressor, the compressor drives the cold head, the cold head keeps the helium level stable. A falling helium level is therefore a symptom of a weak link in that chain, not normal consumption.

Homogeneity is the second job of the magnet. Passive shims (small steel pieces placed in the bore at installation) and active shim coils make the field uniform over the imaging volume. When homogeneity drifts, you see poor fat suppression, geometric distortion and shading. A common cause is external: a new steel cabinet, a heavy door frame or building work near the magnet. Ask what changed around the room before you suspect the magnet itself.

A hair grip is enough to spoil fat suppression. Fat and water protons precess at slightly different frequencies; the textbook difference is about 3.5 ppm, roughly 220 Hz at 1.5 T. Fat-saturation and water-excitation pulses are tuned to that exact offset in a narrow band. A ferrous object in or near the bore - a hair grip behind the bore liner, a forgotten tool, a steel part on a trolley - bends the local field by a comparable amount or more. Near it the fat and water peaks move, the saturation pulse hits water instead of fat or misses fat altogether, and you see patchy, failed fat suppression, worst off-centre and close to the object. Remove the object, then confirm with a homogeneity check or phantom scan before you suspect the shims or the magnet.

The magnet also carries the most serious safety hazards of the whole system: projectiles, the quench, cryogenic gas and oxygen depletion. They get their own lessons. The one point to take from this lesson is that the magnet is always on.

• Typical magnet-related signs: falling helium level or rising pressure in the cryogenics log, changed sound or rhythm of the cold head, frost or condensation on the cold head or lines, compressor alarms, centre-frequency drift on the daily or weekly QA, poor fat saturation across all coils.

Shutting down the system

Before a long holiday, a building power job or simply to save energy, someone will ask to switch the scanner off. The answer comes from the manufacturer's shutdown procedure for that model, not from common sense. It lists which parts may be powered down, in what order, and how they are brought back. Follow it step by step and record what you did.

The magnet keeps its field without power, but its cooling does not. The helium compressor and the water that cools it must keep running continuously: with the compressor stopped, the cold head no longer recondenses helium and the bath starts to boil off. Switching the compressor off at night or over a weekend to save electricity does not save money: the lost helium and the wear from repeated warm-up and cool-down of the cold head cost far more than the power. Any planned compressor stop is an exception approved by the manufacturer, with a time limit, not something the hospital decides on its own. Label the compressor and chiller breakers so nobody opens them by mistake.

Gradient coils and amplifier

Three gradient coils (X, Y and Z) are cast into a cylinder inside the bore, between the magnet and the patient. Each is driven by its own channel of a gradient amplifier in the technical room - a high-power switching amplifier that delivers large, precisely shaped current pulses many times per second. The gradients are what make the scanner loud: the coil conductors sit in a strong field and are pushed by every current pulse.

Both the coil and the amplifier get hot and are water-cooled. On most systems this cooling depends on the chiller and on a secondary water circuit with pumps, filters, flow switches and temperature sensors. Poor flow or a warm chiller is one of the most frequent reasons for scans that abort halfway, especially on demanding sequences such as diffusion or fast imaging. Before you suspect the amplifier, read the water temperatures and flow.

Fast-switched gradients also induce eddy currents in the surrounding metal of the magnet. The system compensates for them with calibrated pre-emphasis. If the compensation drifts or a calibration is wrong, you see distortion or ghosting, often worse along one axis. A fault confined to one axis - errors from one amplifier channel, distortion only in one direction - is a strong hint that the problem lives in the gradient chain.

Listen as well as look. Every MR scanner is noisy, and the noise is part of normal operation. A new noise is not: a rattle, buzz or resonance that was not there last month often means a loose cover, panel or mount, and occasionally something worse. Note it in the report with the sequence that provokes it.

The RF chain and the Faraday cage

The RF chain has two halves. The transmit half is an RF amplifier that feeds a transmit coil, usually the body coil built into the bore, to tip the magnetisation. The receive half starts at the local receive coils placed on the patient - head, knee, spine, surface arrays - and runs through preamplifiers, the table connectors and the receivers to the digitiser. Depending on the architecture, digitisation happens in a cabinet or inside the coil itself; in the second case a receive coil is an electronic instrument with its own supply and data link, not a passive loop of copper, and dropping it can damage it.

The received MR signal is extremely weak, so the whole examination room is a Faraday cage: shielded walls, floor and ceiling, a shielded window, waveguides for pipes and gas lines, filters on every electrical line that enters, and a door that seals the shield through spring contact fingers along its edge. If any of these is broken, outside radio signals leak in and appear as a bright line or row of dots across the image at a fixed frequency position - the classic 'zipper'.

In the field, the cage is broken far more often than the electronics. In rough order of frequency: the door is not fully closed or its contact fingers are bent, dirty or worn; someone brought an emitting device into the room - a phone, a charger, a cheap LED lamp driver, a wireless device; a cable was routed through the wall bypassing the filter panel; the shield was damaged during building work. Checking the door contacts takes two minutes and should be on every preventive maintenance visit.

• Signs pointing to transmit: no signal at all, or uniformly low signal everywhere; SAR or reflected-power messages; failed RF calibration at the start of the exam.
• Signs pointing to receive: signal drop or noise in one region or with one coil only; a coil not recognised; problems that follow the coil when you move it to another connector, or follow the connector when you swap coils.
• Signs pointing to the cage: a straight line of interference across the image in the phase direction, present on different coils and sequences, often appearing after something changed in the room.

Computer and reconstruction

The fourth system is the computer chain. A host computer runs the user interface, the sequences and the database; one or more reconstruction computers turn raw data into images; disks hold raw data and images; a network link sends studies to PACS and receives worklists. On top of this sit licences and options that switch features on and off. Some systems also keep service logs, configuration and calibration files on the same disks.

Faults here look like software rather than physics: the console freezes, reconstruction is slow or fails with a message, images are missing or arrive late in PACS, the worklist is empty, a sequence option disappears after an update. They rarely change the look of an image that does reconstruct. Disk health, free space, the network path and the time since the last configuration backup are the first things to check.

Two rules apply on every visit. First, make sure there is a recent backup of configuration and calibration data, done the way the manufacturer's documentation describes, before you change anything. Second, never touch licences or options beyond what the owner has bought. Bypassing a licence is outside the scope of this course and outside professional practice.

Where to look first for any complaint

When a user says 'the images are bad' or 'the scanner keeps stopping', resist the urge to open cabinets. Ask questions and read data first. The answers usually point to one subsystem within ten minutes.

Start with scope. Does the problem appear with every coil or one coil? Every sequence or only heavy ones? Every patient or one body region? Since when, exactly, and what changed at that time - a power cut, a chiller alarm, building work, new furniture, a software update, a new staff member? Then read the event and error log for the same time window; the first error in a cascade is usually the cause, the following ones are consequences. Then check infrastructure: room temperature, chiller and water temperatures, mains and UPS events, the cryogenics readings. Only then run the manufacturer's image quality test with the phantom and compare it with the baseline and the trend.

Use the pattern of the symptom to route the complaint. The same image can have more than one cause, so treat this as a first guess to confirm, not a verdict.

• Fat suppression poor everywhere, distortion, centre frequency drifting: magnet homogeneity or field stability - ask what changed near the room; check cryogenics log.
• Scans abort on demanding sequences, amplifier temperature or current errors: gradients and their cooling - check chiller, water flow and temperatures first.
• Distortion or ghosting along one axis only: gradient channel or eddy-current compensation.
• Zipper line across the image on all coils: Faraday cage - door, contacts, filters, devices in the room.
• Signal loss or noise with one coil or one connector: receive chain - coil, cable, table connector.
• No signal at all, RF calibration fails: transmit side.
• Freezes, slow or failed reconstruction, missing images in PACS: computer, disks, network.

On-site checklist

• Ask: since when, on which coils, which sequences, which patients - and what changed in or near the room at that time.
• Read the event and error log for the same time window; find the first error, not the loudest.
• Check infrastructure: room temperature and humidity, chiller and water temperatures and flow, mains and UPS events.
• Read the cryogenics values (helium level, pressure, compressor status) and compare with the last visits.
• Inspect the RF door: fully closing, contact fingers clean and undamaged; look for emitting devices in the room.
• Run the manufacturer's image quality test with the phantom and compare with baseline and trend.
• Confirm a recent configuration backup exists before changing anything.

Check yourself