MRI has reached a high level of maturity in the past four decades. Although many of the system components can be considered to be “black boxes” that most users will never have to modify, it is beneficial to have an understanding of the various operational blocks. These can be categorized based on the functionality necessary to acquire and reconstruct an MR image. To produce an image, three basic groups of electromagnetic fields have to be generated: the static magnetic field, the gradient magnetic fields, and the radiofrequency (RF) fields; in addition, RF fields need to be detected and processed by the system. The subsystems necessary to generate and manipulate these fields are briefly described below as well as some further subsystems that are part of most clinical MRI scanners.

The most obvious subsystem of any MRI system is the magnet used to generate the static magnetic field B0. This field is required to orient the nuclear spins and separate them into discrete energy levels. The stronger the magnet, the larger the overall spin polarization and thus the larger the achievable signal-to-noise ratio (SNR). As this magnetic field is static with no time variation, there is no associated electric field. Typical field strengths of modern clinical MRI systems are 1.5 and 3 Tesla. To achieve higher SNR, there has been a continuing trend to higher magnetic fields throughout the history of MRI. Currently, 7 Tesla is being considered for introduction into routine clinical use, but there are also human research systems operating at 9.4 and 10.5 Tesla, with 11.7 Tesla in planning.

Lower magnetic fields can be generated by electromagnets or by permanent magnets. Above about 1 Tesla, it is more efficient to generate the field using superconducting technology, even though this approach requires cooling the magnet down to temperatures typically around 4 K and immersing the superconducting coils in a bath of liquid helium. Niobium titanium is the superconductor of choice for most MRI magnets, but this technology reaches its technical limits at 11.7 Tesla. For higher magnetic fields, new superconductors such as niobium tin will have to be utilized.

Two major criteria when designing magnets for MRI are spatial uniformity and stability over time. Minor departures from ideal uniformity are corrected with various shims, which can be generated using superconducting coils, passive shim material, or resistive coils.

The gradient amplifiers and coils are responsible for generating magnetic fields in addition to the main static magnetic field B0 provided by the magnet. These fields are fairly weak, i.e. on the order of a few mT, and ideally they are designed to vary linearly in the three orthogonal spatial directions (Gx, Gy, Gz). Thus, there are usually three gradient amplifiers driving three separate gradient coils. Due to switching of the magnetic fields in pulse sequences, fields are produced with a broad spectrum of frequency components in the audio band.

The application of these fields to body tissue leads to a minor change in the Larmor resonant frequency of the precessing nuclear spins, as the precession frequency is linearly proportional to the external magnetic field experienced by the spins. Without these fields, all of the spins would precess at the same nominal frequency, and it would not be possible to determine from which position in the body a particular signal is originating. By applying a Fourier transform to the received time-signal, it is possible to assign signal strengths along each of the spatial axes and reconstruct spatially resolved images. Note that by spectroscopically assigning the spatial origin of the signal, the resolution of MR images is not limited by the wavelength of the RF radiation emanated by the excited spins. This makes MR quite different than optical imaging.

If the gradient amplitude is too high, eddy currents can be generated in the conducting tissue that provoke peripheral nerve stimulation (PNS). The amplitude at which PNS occurs is dependent on the switching time of the gradient, which has led to limits on the performance of the gradient subsystem based on physiological considerations. Many systems have a dedicated stimulation monitor to shut down the system if excessive gradient waveforms are applied.

To circumvent these performance limits, one strategy is to reduce the linearity of the gradients, so that the field amplitudes at the periphery of the field of view (FOV) of the gradient coil are less; such nonlinearities can be corrected to a certain extent during image reconstruction, since the non-idealities can be well characterized during system calibration. Another strategy is to reduce the target FOV of the coil itself. Note that this means reducing the maximum physical FOV of the gradient system and cannot be achieved just by acquiring images at reduced FOV. Thus, head-only gradient systems can provide higher performance than whole-body gradient systems.

The goal of the RF transmission subsystem is to excite the polarized spins and drive them into the other allowed energy level. Viewed from the perspective of classical physics, the RF energy tips the spins from alignment along B0 to an orientation with components orthogonal to B0. Only these transverse components can be detected by the RF reception system. The frequency of the RF energy must be resonant with the precession frequency of the spins and ranges from about 64 MHz for hydrogen at 1.5 Tesla up to 500 MHz for hydrogen at 11.7 Tesla. Other nuclei possess lower resonant frequencies than hydrogen due to their lower gyromagnetic ratios.

An initial small-signal RF pulse is generated in the system exciter, which consists of an oscillator and a phase/amplitude modulator. This pulse is augmented to high power levels by an RF amplifier and applied to the transmit RF coil. Typically, the goal of RF excitation is to generate a uniform magnetic RF field B1 that is transverse to B0. Spatial uniformity ensures that the flip angle does not vary across the subject, and thus the image contrast is predictable and constant throughout the entire excited volume. At lower fields, birdcage coil geometries are commonly used, since these coils produce a very uniform transverse magnetic field in the central portion of the coil. They can also be easily driven to produce a circularly polarized field, which is the only component useful for spin excitation. The reverse circularly polarized component that is inherently present in a linear polarized field produces unwanted tissue heating due its electric fields but no spin excitation.

Tissue heating is in general the result of the RF electrical fields that compulsorily accompany all RF magnetic fields through Maxwell’s equations; RF power deposition into the tissue is characterized by the specific absorption rate (SAR) given in W/kg. MRI systems require monitoring of the RF power being transmitted to make sure that the RF exposure of the patient remains within regulatory limits. Such a SAR or RF exposure monitoring system should ideally be independent of other system components, and its correct function should be regularly verified.

Unlike low-frequency gradient magnetic fields, for which the body tissue is basically transparent, RF fields do interact extensively with the tissue through which they propagate. These interactions, which include absorption, refraction and reflection, are dependent on the distribution of tissue conductivity and permittivity, and they are more pronounced at higher RF frequencies. This implies that it is quite challenging to generate a uniform B1 field in MR systems with higher B0.

To account for this challenge, it is now common to excite the tissue with a set of independently controllable coil elements rather than a single monolithic coil. Such parallel transmit systems require at least the ability to apply RF power with arbitrary amplitude and phase weightings to each of the coil elements; this approach is often called RF shimming, since it is similar to the shimming strategy pursued for B0 using multiple shim coils. For more complex applications, the complex weightings are changed at each support point of the RF pulse, implying that different RF pulse forms are applied to each coil element. The effect of such parallel transmit pulses (Transmit SENSE) can only be understood in conjunction with knowledge about the trajectory through k-space, which is determined by the accompanying gradient magnetic fields. Accurate monitoring of RF exposure in parallel transmit systems is very challenging.

Once the spins have been excited using the RF transmission system, the RF reception system is utilized to detect the very weak magnetic fields issued by the spins precessing in the plane transverse to B0. In principle, the same RF coil as used for transmission could also be used for this purpose, but in practice separate coils are used as the design criteria differ for the two functions. During RF reception, sensitivity is more important than uniformity. Thus, RF coils for reception are often composed of multiple, smaller coil elements that are placed as close as possible to the volume of interest.

Not only do the smaller elements enhance the sensitivity (SNR) of the coils, they also enable use of parallel imaging techniques. Since each coil element only receives signal from a limited portion of the volume of interest, the coils inherently provide information about the spatial localization of the signal that they receive. This information can be used to supplement the spatial encoding performed using magnetic field gradients, thereby reducing the acquisition time significantly. It is important to have sufficient RF coil encoding capability along the phase-encoding direction or directions. For simultaneous multislice (SMS) techniques that rely on RF coil encoding, it is important to have coil elements distributed along the slice select direction.

The RF signals received by the RF coils are amplified by low-noise amplifiers followed by sampling and digitization. In some systems, the digitization occurs in the examination room very close to the coil itself. In other systems, the digitization is performed in the equipment room. In cases where the receiver may process signals from different-sized coil elements, the receiver must be able to cover a large dynamic range, since in some cases signal might be acquired from a large volume of tissue and in other cases only from a very small volume. If the receiver is not able to cover this range at fixed signal amplification, the system may have to provide multiple levels of gain between the coil and the receiver.

In addition to the major subsystems mentioned above, a clinically useful MR imager relies on multiple other hardware components to enable patient examinations. These include the patient table, which is usually motor-driven and able to move the patient to the appropriate longitudinal positon so that the images are acquired in the central portion of the magnet. For multi-station acquisitions, the table must also be able to quickly and accurately transport the patient from station to station.

Another set of important components are the physiological monitoring devices. In many instances it is necessary to synchronize the MR data acquisition or reconstruction to the ECG, peripheral pulse, and/or respiratory signals to reduce motion artifacts or produce CINE movies of periodic motion.

The part of a MR system that a user is most familiar with is the computer front-end. Software represents the main source of system complexity in clinical MRIs. It allows the user to control all of the other hardware by selecting pulse sequences and changing the parameters of the acquisition to achieve the desired coverage, resolution, and contrast. Imaging parameters include TR, TE, flip angle, FOV, readout bandwidth, imaging matrix and many others. After the data are acquired, they have to be reconstructed, which nowadays often occurs on fairly standard PC hardware, possibly in conjunction with GPUs (Graphics Processing Units). Beyond basic image reconstruction, advanced post-processing and analysis techniques are increasingly important in clinical routine and research. Most MR systems have dedicated computers for the front-end user interface, system control, and image reconstruction.MR imaging depends on the interaction between several hardware subsystems, including the magnet to generate spin polarization, the magnetic field gradients to provide spatial localization, and the RF coils to achieve both spin excitation and signal reception. In particular the available field of view depends on the interplay of the static field homogeneity, the linearity of the gradients, and the transmit volume of the body coil. The performance limits of the gradients and the RF transmit coils are set by physiological considerations, i.e. PNS and SAR. The limits of the static magnetic field, on the other hand, are still governed by technological and economic considerations. The physiological upper limit of B0 acceptable for MRI is still unknown.