This tutorial explains how magnetic resonance imaging (MRI) systems use the reaction of hydrogen atoms moving in a magnetic field to produce a detailed medical image. The types of magnetic fields typically used are described. The note explains why today's highest-resolution MRI systems are based on superconducting magnets. The note also discusses three-dimensional imaging through proper alignment of gradient coils and their interaction with RF signaling. A functional block diagram of a typical MRI system is presented.
Magnetic resonance imaging (MRI) systems provide highly detailed images of the body's tissues. The systems capture and process the signals generated when hydrogen atoms, which are abundant in tissues, are placed in a strong magnetic field and excited by a resonant magnetic excitation pulse.
Hydrogen atoms have an inherent magnetic moment due to their nuclear spin. When placed in a strong magnetic field, the magnetic moments of these hydrogen nuclei tend to align. In simplified terms, hydrogen nuclei can be thought of as a taut string in a static magnetic field. Nuclei have a resonant frequency or "larmor" determined by the strength of their localized magnetic field, just as a string has a resonant frequency determined by the tension on it. For hydrogen nuclei in a typical 1.5 T MRI field, the resonance frequency is about 64 MHz.
Proper stimulation by an RF or resonant magnetic field at the resonant frequency of the hydrogen nuclei can force the magnetic moments of the nuclei to be tilted partially or fully in a plane perpendicular to the applied field. When the applied RF excitation field is removed, the magnetic moments of the nuclei in the static field precess as they realign. This realignment creates an RF signal with a resonant frequency determined by the magnitude of the applied field. This signal is captured by the magnetic resonance imaging system and used to create an image.
Static Magnetic Field
For MRI, the patient must be placed in a strong magnetic field to align the hydrogen nuclei. There are generally three methods of creating this field: fixed magnets, resistive magnets (current flows through a conventional coil of wire), and superconducting magnets. Fixed magnets and resistive magnets are generally limited to field strengths below 0.4 T and cannot generate the higher field strengths normally required for high-resolution imaging. As a result, most high-resolution imaging systems use superconducting magnets. Superconducting magnets are large and complex; They require the coils to be soaked in liquid helium to bring their temperature down to near absolute zero.
The magnetic fields generated by these methods have to be not only strong, but also spatially very uniform and temporally stable. A typical system should have less than 10 ppm variance over the imaging range. To achieve this level of precision, most systems generate weaker static magnetic fields by using special tuning coils to "tune" or "adjust" the superconducting static field, thus correcting for field inaccuracies.
To produce an image, the MRI system must first stimulate hydrogen nuclei in a given 2D imaging plane in the body, and then determine the position of those nuclei within that plane as they return to their static state. Both tasks are accomplished using gradient coils that cause the magnetic field to change linearly within a localized area as a function of spatial position. As a result, the resonance frequencies of the hydrogen nuclei within the gradient are location dependent. The area of the body to be stimulated is controlled by varying the frequency of the excitation pulses. The location of the stimulated nuclei as they return to their static state can also be determined using the emitted resonant RF frequency and phase information.
An MRI system must have x, y, and z gradient coils to generate gradients in three dimensions and thereby create an image slice in each plane within the patient's body. The application of each gradient field and excitation pulses must be correctly sequenced or timed to allow collection of an image data set. For example, by applying a gradient in the z-direction, the resonant frequency required to excite a 2D slice in that plane can be changed. Therefore, the spatial position of the 2D plane to be imaged is controlled by changing the excitation frequency. After the excitation sequence is complete, another correctly applied x-direction gradient can be used to spatially alter the resonant frequency of the nuclei as they return to their static position. The frequency information from this signal can be used to locate the position of the nuclei in the x-direction. Similarly, a correctly applied gradient field in the y-direction can be used to spatially alter the phase of resonance signals, and thus used to detect the position of nuclei in the y-direction. By correctly applying the RF and gradient excitation signals in the correct order and at the correct frequency, the MRI system produces images of a three-dimensional section of the body.
To achieve adequate image quality and frame rate, the gradient coils in the MRI imaging system must rapidly change the strong static magnetic field in the region of interest by approximately 5%. High voltage (operating at a few kilovolts) and high current (100 amps) power electronics are required to drive these gradient coils. Despite the high power requirements, low noise and stability are important performance metrics, as any ripple in the coil current introduces noise in subsequent RF acquisition. This noise directly affects the integrity of the images.
To distinguish tissue types, MRI systems analyze the size of the received signals. The excited nuclei emit a signal until the energy absorbed during the excitation phase is released again. The time constant of these exponentially decaying signals varies from a few tens of milliseconds to more than a second; recovery time is a function of field strength and tissue type. It is the fluctuations of this time constant that make it possible to identify different types of tissue.
The transmitting and receiving coils are used both to stimulate the hydrogen nuclei and to receive the signals generated as the nuclei are recovered. These coils must be optimized for the specific area of the body to be imaged, so they are available in a wide variety of configurations. Depending on the area of the body to be imaged, separate transmit and receive coils or combined transmit/receive coils are used. Additionally, to improve image acquisition times, MRI systems use multiple transmit/receive coils to acquire more information in parallel, using the spatial information associated with the position of the coils.
high frequency receiver
An RF receiver is used to process the signals from the receiver coils. Most modern MRI systems have six or more receivers to process signals from multiple coils. The signals range from approximately 1 MHz to 300 MHz, and the frequency range is highly dependent on the strength of the applied static magnetic field. The bandwidth of the received signal is small, typically less than 20 kHz, and depends on the magnitude of the gradient field.
A traditional MRI receiver setup has a low noise amplifier (LNA) followed by a mixer. The mixer mixes the signal of interest at a low-frequency IF frequency for conversion using a high-resolution, low-speed 12-bit to 16-bit analog-to-digital converter (ADC). In this receive architecture, the ADCs used have relatively low sample rates below 1 MHz. Due to low bandwidth requirements, ADCs with higher sample rates from 1 MHz to 5 MHz can be used to convert multiple channels on a single ADC by time division multiplexing the receive channels through an analog multiplexer.
With the advent of more powerful ADCs, newer receiver architectures are now possible. High-resolution 12-bit to 16-bit ADCs with high input bandwidth and sample rates up to 100 MHz can also be used to directly sample signals, eliminating the need for analog mixers in the receive chain.
The MRI transmitter generates the RF pulses needed to oscillate the hydrogen nuclei. The frequency range in the transmit drive pulse and the magnitude of the gradient field determine the width of the image slice. A typical transmit pulse produces an output signal with a relatively narrow bandwidth of ±1 kHz. The time-domain waveform required to generate this narrow band of frequencies often resembles a traditional timing function. This waveform is usually digitally generated at baseband and then converted to the appropriate center frequency using a mixer. Conventional broadcast implementations require relatively low-speed digital-to-analog converters (DACs) to generate the baseband waveform because the bandwidth of this signal is relatively small.
Again, other potential transmission architectures are feasible with recent advances in DAC technology. High-speed, high-resolution DACs can be used for direct RF generation of transmit pulses up to 300 MHz. Therefore, waveform generation and upconversion over a wide frequency band can now be achieved in the digital domain.
image signal processing
Both frequency and phase data are collected in what is known as k-space. A two-dimensional Fourier transform of this k-space is computed by a display processor/computer to produce a grayscale image.