Magnetic resonance imaging methodology
- 1.6k Downloads
Magnetic resonance (MR) methods are non-invasive techniques to provide detailed, multi-parametric information on human anatomy, function and metabolism. Sensitivity, specificity, spatial and temporal resolution may, however, vary depending on hardware (e.g., field strength, gradient strength and speed) and software (optimised measurement protocols and parameters for the various techniques). Furthermore, multi-modality imaging may enhance specificity to better characterise complex disease patterns.
Positron emission tomography (PET) is an interesting, largely complementary modality, which might be combined with MR. Despite obvious advantages, combining these rather different physical methods may also pose challenging problems. At this early stage, it seems that PET quality may be preserved in the magnetic field and, if an adequate detector material is used for the PET, MR sensitivity should not be significantly degraded. Again, this may vary for the different MR techniques, whereby functional and metabolic MR is more susceptible than standard anatomical imaging.
Here we provide a short introduction to MR basics and MR techniques, also discussing advantages, artefacts and problems when MR hardware and PET detectors are combined. In addition to references for more detailed descriptions of MR fundamentals and applications, we provide an early outlook on this novel and exciting multi-modality approach to PET/MR.
KeywordsMagnetic resonance Imaging Spectroscopy Artefacts Multi-modality imaging PET/MR
Since magnetic resonance imaging (MRI) entered the clinical arena in the early 1980s, MRI has experienced dramatic advances associated with an increasing number of clinical applications that, in turn, are driving further technical developments. Shortly after its introduction, MRI became one of the most important diagnostic imaging modalities and continues this role today. This success is due to the inherent characteristics of anatomical MRI, featuring excellent soft tissue contrasts that are based on multiple contrast parameters, the ability to image in oblique orientations and the capability to provide two-dimensional (2D) as well as 3D data. MRI is considered a non-invasive diagnostic imaging modality that generates cross-sectional images of the human body without the use of ionising radiation. The diagnostic importance of MRI is impressively reflected by an ever growing number of MRI scanner installations. Informal market research studies suggest that, today (2008), about 20,000–24,000 MRI systems are installed and operational worldwide. These studies also indicate that about 60–80 million MRI examinations are being performed each year.
As the MR signal depends on a variety of physical tissue parameters, MRI may also provide a broad range of functional information on diffusion, perfusion, flow rates, temperature, magnetic susceptibility, etc. beyond displaying morphologic and structural information. Owing to this spectrum of information, the clinical applications of MRI encompass neurological, psychiatric, abdominal, cardiac, vascular as well as musculoskeletal applications, covering the entire human body. Furthermore, metabolic changes in and around tumours, degenerative brain diseases and accompanying the metabolic syndrome may be visualised using MR spectroscopy (1H, 13C, 31P) or spectroscopic imaging (1H, 31P). With the advent of whole-body MR imaging technologies , whole-body tumour staging and screening for metastasis has become a viable option today [2, 3, 4].
In the context of combined PET and MR imaging, it is important to understand the underlying principles and limitations of both imaging modalities. To provide a basic understanding of MR technology, we shall describe the most important MRI hardware components, their interaction with the human body and some of the most common measurement techniques and parameters that may be varied by the user to obtain the diagnostic information of choice.
We conclude by summarising current knowledge on advantages and challenges in combining the two imaging modalities.
Main magnet—homogeneous static magnetic field
The magnet of a whole-body MR scanner should have a high main magnetic field strength, B 0, to provide sufficient equilibrium magnetization and therefore a high potential SNR for good image quality. MR systems with field strengths above 1.0 T, or better yet 1.5 T, are currently assumed as the clinical standard and are increasingly being supplemented with 3 T systems in clinical use. Beyond clinical application, a limited but increasing number of 7 T, or even higher field whole-body MR systems have been installed for research investigation in human high-field MR. The homogeneity of the basic magnetic field over the examination volume should be as high as possible to ensure low image distortion and high signal uniformity. Also, the homogenous examination volume should be as large as possible. A cylindrical main magnet with the homogeneous volume centred on the central axis fulfils all these requirements and, therefore, represents the most frequent magnet design today (Fig. 1).
Gradient system—strong, fast and linear (orthogonal) gradient fields
Three orthogonal gradient fields (G x , G y and G z ) are needed for spatial localisation. For performing fast MR imaging, two parameters of the gradient system are critical: a fast gradient slew rate [given in (mT/m/ms)] combined with a high gradient amplitude [given in (mT/m)] are the prerequisites for short repetition (TR) and echo times (TE) and, therefore, for coverage of a large examination volume in the shortest time possible. This can be seen as a fundamental prerequisite for clinically acceptable examination times and for covering large volumes in multi-station or continuously moving table whole-body MRI applications. Ultrafast gradients are also essential in functional (echo-planar) MRI. Fast switching of high-amplitude gradients may, however, also increase the risk of peripheral nerve stimulations . A high degree of gradient linearity over a large volume is required by the gradient system to keep image distortion in and around the image volume to a minimum. These gradient system requirements can be best realized with a cylindrical design to conform to the magnet geometry (Fig. 1).
RF system—homogeneous spin excitation across the body
The radiofrequency (RF) system, in general, consists of a RF transmit coil installed within the MR housing for selective RF excitation of the spins as well as a RF receiver coil system for receiving the weak RF signals leaving the patient. RF excitation of the tissue volume of interest should be as homogeneous as possible. Here, it is important that the RF signal strength for tissue excitation (indicated by the flip angle) remains consistent across the image. Large cylindrical volume transmit RF coils with a conductor geometry similar to a birdcage and fitted within the cylindrical magnet tunnel are typically used (Fig. 1).
MR protocol parameters—timing of RF and gradient pulses determines contrast and location
Short RF pulses of a few milliseconds duration are required to excite the nuclear spins. This energy is deposited in the sample and is then irradiated back to a signal-receiving coil. A considerable share of this energy is dissipated into thermal energy and is referred to as the specific absorption rate (SAR). As this process causes tissue warming, which may be harmful to the patient, SAR limits are constantly controlled throughout an MR scan to ensure compliance with international MR safety guidelines. As indicated by the term “resonance”, spin ensembles will absorb this RF energy effectively only if the transmission frequency is equal to the Larmor frequency of the spins, which scales linearly with the magnetic field strength. For hydrogen nuclei (1H), this frequency increases from 64 MHz at 1.5 T to 128 MHz at 3 T. It is possible to limit excitation to certain parts of the subject through spatial variation of the magnetic field (via gradients) during RF excitation, thus allowing selection of a volume of interest, as only these irradiated regions will emit RF signals back.
In the case of MR spectroscopy, we are interested in metabolite concentrations rather than water content. Hydrogen nuclei bound to metabolites exhibit slightly different Larmor (resonance) frequencies (“chemical shift”) depending on the density of the local chemical environment and, hence, on the density of the electrons. The differences in electron density result in variable shielding of the hydrogen nuclei and, consequently, lead to different resonance frequencies. It is possible to apply a specially designed RF pulse to suppress the water signal without altering the metabolite signals. This is required as the mobile water content in human tissue is typically about 50 mol, resulting in a significant water peak, which would mask metabolite peaks detectable by 1H-MRS (sensitivity in the mM range). Fourier transformation of the acquired MR signal then provides the full spectrum of metabolites in the specified voxel. The duration for single-voxel MRS is just a few minutes, and the spatial resolution is 1 cm3 or less. In some sequences, referred to as chemical shift imaging (CSI), additional gradients are applied prior to signal acquisition to introduce spatial encoding within the preselected volume, enabling acquisition of spectrums from multiple individual voxels to provide insight into the spatial distribution of the spectroscopic information. This method allows performing MRS with a spatial resolution of 0.1 cm3 or less , which is, however, associated with a significant increase in measurement time up to 1 h or more.
Imaging sequences require additional gradients to introduce spatial encoding (Fig. 3; right). Depending on whether the gradients are applied before or during signal acquisition, they are referred to as phase or readout gradients, respectively. MR signals for imaging sequences are acquired within k-space, which may be regarded as a mathematical model for spin-gradient interactions. The final image is reconstructed after 2D Fourier transformation of the k-space signal. In standard imaging sequences, a single line of k-space is acquired after each RF excitation, i.e., for N image lines (the final image matrix has N pixels along one side), the sequence has to be repeated N times, leading to image acquisition times of up to several minutes. In current ultrafast imaging techniques like echo-planar imaging (EPI) or Turbo Spin Echo (TSE), however, multiple k-space lines or even the entire k-space data are acquired following a single excitation (“single-shot” methods).
Structural/anatomical MR imaging
Anatomical imaging by MRI and CT is the clinical standard for all stages in the management of tumour patients, e.g., detection, characterization and staging of the lesion, control of therapeutic response and determination of recurrence. However, it is well known that imaging of the anatomical structure alone suffers from many shortcomings. The evaluation of macroscopic changes in lesions and tissues can lead to a false assessment of the pathologic processes and their progression on the cellular and molecular levels. Furthermore, macroscopic abnormalities commonly are nonspecific and reveal no information about physiological and biological processes.
This technique uses ultrafast imaging methods, mainly echo-planar imaging (EPI), to collect functional information on blood oxygenation or flow changes [BOLD functional MRI (fMRI)], perfusion or diffusion. fMRI based on the blood oxygen level-dependent (BOLD) effect has been increasingly used to investigate brain function since its introduction in 1990 . It shows a wide field of applications ranging from the investigation of basal and higher brain functions in healthy subjects and animals to pharmacological studies and the preoperative localization of eloquent cortical areas in patients. However, BOLD fMRI assesses changes in the deoxyhemoglobin level, which is associated with hemodynamic response due to changes in neuronal activity. Stimulation of a particular brain area via a defined paradigm leads to a local increase in cerebral blood flow and volume and to a decrease of deoxyhemoglobin content in the capillary bed and the venous-draining vessels. Due to the paramagnetic nature of deoxyhemoglobin, tissue T2* relaxation time is increased, and subsequently an increased signal intensity on T2*w images is also observed. In order to observe the BOLD signal change, a T2*w sequence is needed, typically an echo-planar imaging sequence. MR protocols with a TE approximately equal to T2* result in maximum BOLD sensitivity upon brain activation.
Compared to functional PET studies, BOLD fMRI experiments show similar spatial resolution and superior temporal resolution. Electrophysiological methods such as EEG and MEG offer better temporal resolution, but EEG has less spatial accuracy and MEG is less available. However, when performing BOLD fMRI, one has to keep in mind the potential confounds, which can occur through drug or disease-induced modulation of physiological events responsible for the generation of the BOLD signal. BOLD fMRI studies are strongly recommended for the preoperative diagnostics of brain tumour patients to delineate functionally important neuronal tissue, which should be preserved during treatment, as well as to obtain information on tumour-induced brain plasticity [10, 11].
This technique is used to investigate differences in the blood supply of lesions and healthy tissue. In principle, noninvasive arterial spin labelling (ASL) can be used to magnetically tag inflowing blood in order to get a direct measurement of organ perfusion . ASL suffers, however, from intrinsically low SNR, and, thus, most applications employ exogenous contrast agents. Dynamic contrast-enhanced MRI (DCE-MRI) techniques require the acquisition of images before, during, and after intravenous administration of a paramagnetic contrast agent, usually a gadolinium (Gd) chelate bolus .
Contrast-enhanced perfusion-weighted MR imaging (PWI) can be performed in two ways. The first is called T1-based perfusion and utilizes the shortening of the T1 relaxation time of tissues by the paramagnetic Gd-based contrast agent. This “T1 effect” is proportional to the amount of contrast media passing through the region under investigation. To observe the changes in the T1 relaxation time, a fast T1-w sequence has to be executed, and the dynamic course of the changes in T1 times is evaluated. T1-based PWI is applied to study the contrast uptake of organs and tissues in the abdomen and thorax (e.g., breast, heart, prostate), usually using a 3D spoiled gradient echo sequence, since this approach allows the acquisition of dynamic information within a breath-hold with good image quality and spatial resolution. It is used to improve tumour diagnosis, evaluate response to therapy, and to detect residual or recurrent tumour. The second PWI method is called T2*-based perfusion and exploits the influence of paramagnetic contrast agent on the local magnetic field which leads to a reduction of the T2* relaxation time in surrounding tissue . A very fast T2*-sensitive sequence has to be performed to measure these changes in the T2* relaxation time, i.e., EPI sequences are commonly applied. During the passage of the contrast agent bolus, the MR signal decreases due to the reduction of T2*, but the signal recovers at least partially after the passage. The differences in occurrence and magnitude of the signal changes are evaluated via an arterial input function (AIF). T2*-based PWI allows for imaging of flow at the capillary level. The measurement of brain perfusion helps to differentiate between acute stroke, hypoglycaemia, hyponatremia, seizure, tumour or subdural hematoma. In acute stroke, for instance, the difference between the diffusion abnormalities measured with DWI and perfusion abnormalities measured with PWI provides a measure of the ischemic penumbra or the brain tissue at risk.
Diffusion-weighted MRI (DWI) assesses the Brownian motion of water molecules and provides information about tissue structure and intra/extracellular space via the “apparent diffusion coefficient” ADC . DWI shows broad clinical application in diagnosis of acute cerebral infarction, characterization and therapy monitoring of tumours, and differentiation of gliomas from brain abscesses or lymphomas . A DWI sequence detects the signal attenuation due to diffusion of water molecules. The essential parameters for the “diffusivity weighting” are the strength and timing of the diffusion gradients and can be expressed with the “b-factor”. Higher b-factors lead to stronger signal drop-off in areas with higher diffusion. The observed signal decay can be used to calculate the diffusion properties or ADC.
Depending on the anatomical region, diffusion can occur unrestricted and equal in magnitude in all spatial directions, which is referred to as isotropic diffusion. However, mobility of the water in biological tissue can show a preferred direction governed by structures, e.g., restriction of diffusion by myelinated sheets of nerve fibres, which is referred to as anisotropic diffusion. The DWI method provides information about the magnitude of water diffusion but not about its direction.
Diffusion tensor MR imaging
Magnetic resonance spectroscopy
MR-based spectroscopy (MRS) is a noninvasive diagnostic tool that allows for the detection of amount and spatial distribution of various molecular compounds, which are involved in the metabolism of pathologic and healthy tissue. In proton MRS (1H-MRS), the most important metabolites are creatine (Cr), choline (Cho), N-acetyl-acetate (NAA), citrate (Cit), lactate (Lac) and lipids . Unfortunately, no marker specific for cancer and detectable via in vivo MRS exists today. However, one can detect typical metabolic profiles in MRS spectra that provide an indication for cancer . Various studies have assessed the potential of 1H-MRS to improve diagnosis and prognosis of pathologies in the central nervous system, especially of brain tumours (Fig. 5 B). In recent years, 1H-MRS has been applied also for diagnosis of other common cancer types such as prostate and breast carcinoma [22, 23].
3D 1H-MRS sequences with a spatial resolutions comparable to that of PET  suffer from low signal-to-noise and have the drawback of long acquisition times. The application of so-called parallel imaging techniques (e.g., SENSE)  is a strategy to overcome this problem. However, 1H-MRS is very sensitive to susceptibility artefacts, which excludes the use of this method in patients with subcranial, frontobasal, temporal, etc. lesion locations or in patients after surgery. Unfortunately, these problems are more evident at higher magnetic field strength.
The sensitivity of 1H-MRS is in the range of millimolars. Positron-emitting analogues of molecules, which are involved in metabolic processes, can be detected with PET in picomolar concentrations. A wide variety of PET tracers tracers have been developed to target the transport and metabolism of glucose (FDG) and amino acids (18F-fluorothymidine or 18F-fluoroethyltyrosine; Fig. 5 E). On the other hand, 1H-MRS may be performed with any modern MRI system without the need for exogenous tracers, while the availability of PET is still limited in part due to the substantial infrastructure required for tracer production.
Whole-body tumour staging
Whole-body tumour staging utilizing MRI has now become feasible due to several technical innovations: whole-body coverage with RF surface coils (Fig. 2), multiple RF receiver channels, extended and fast table translation. Prior to these innovations, MR has been limited to examination of a single organ region, which fit within the homogeneous FOV of the magnet. It is now possible to perform complete TNM-staging using multiple pulse sequence 15–30 min, depending on the exact protocol and image resolution [3, 25].
Many practical imaging decisions are based on a compromise between the SNR and the time available for image acquisition, as the SNR increases with the square root of acquisition time. SNR also increases approximately linearly with the main magnetic field strength . A higher SNR translates into improved spatial resolution or reduced scan times. Any modification of the MR environment, for example, by adding PET components to the MR, must avoid significant degradation of the SNR.
Furthermore, a MR system is a very sensitive radio receiver tuned to a frequency band centred around 64 MHz at 1.5T and 128 MHz at 3T (for hydrogen protons and less for other nuclei). Any electronic noise at the reception frequency will cause banding artefacts in the images (Fig. 7b), or elevated noise levels. Common sources of such noise are any active electronics, and all electronics must be properly shielded to avoid interference.
MR imaging is also susceptible to some artefacts which may hamper proper co-registration of images with an independent imaging modality such as PET. These imperfections may be of particular consequence for attenuation correction algorithms based on the MR data, which rely on proper overlay of the complementary image sets.
First, since many MR imaging sequences are rather slow compared to physiological processes, cardiac and breathing motion often contaminate the images (Fig. 7d), which is a familiar problem from PET/CT imaging. In addition, signal localization in MR imaging is achieved by the application of ideally linear gradients in the magnetic field. However, due to hardware limitations and the potential of peripheral nerve stimulation during very fast imaging, a certain amount of nonlinearity is acceptable in the periphery of the nominal imaging FOV. As a consequence, the signal is mapped to the incorrect location in the reconstructed image, up to several centimetres at the periphery of large FOVs  (Fig. 7c). Although correction algorithms are available, they are typically limited to in plane correction, and they cannot account for extreme nonlinearities.
Finally, it is common in MR imaging to undersample the raw data space in order to save acquisition time. Undersampling in the spacing of raw data lines in the phase-encoding direction leads to fold-over artefacts if the effective FOV is smaller than the object being imaged. This strategy is followed when the imaging region is of limited extent, such as in cardiac exams (Fig. 7e). Obviously, MR-based attenuation correction based on images suffering from fold-over distortions will be quite challenging.
Discussion and outlook
high-resolution MR images that complement PET should be the least affected by a PET insert, in particular in the brain, as very strong gradient inserts (≥100 mT/m) may be used to compensate for geometric distortions or susceptibility artefacts,
functional MRI of the brain (i.e., BOLD fMRI, perfusion, diffusion) using EPI or spiral imaging would be very sensitive to SNR or CNR losses caused, e.g., by PET inserts. This may be compensated, in part, by longer measurement times or higher field strengths,
the most serious problems are to be expected in whole-body PET/MR at high field strengths. Limitations are due to inherent MR problems such as reduced homogeneity in large FOVs and dielectric losses, leading to signal losses, which would be further exacerbated by PET components that reduce SNR and/or increase susceptibility differences,
user-friendly 3D doctors, finally, integrated models will need to be developed and tested in order to better understand the origins of complex diseases.
Assuming these challenges can be addressed, the question still remains, how a practical examination could be performed on a combined PET/MR tomograph. Even if the hardware enables simultaneous artefact-free data acquisition for both modalities, would it make sense? Individual MR sequences for imaging the head often last minutes. If the longitudinal FOV of the PET is sufficient to cover the brain, then several individual MR contrasts could be acquired, while counts are continuously collected for reconstruction of a single, low-noise PET dataset. For imaging outside the brain with combined PET/MR, a similar strategy to brain imaging might be pursued if the area of interest is limited to a single organ.
For examinations requiring extended FOV coverage, on the other hand, the examination logistics may not be as straightforward. For most oncologic staging applications, multiple longitudinal FOVs must be sampled. The longitudinal FOV of most MR scanners is 40–50 cm, exceeding typical PET FOVs by a factor of 2–3. Even if the PET FOV can be extended to cover a larger extent, the acquisition of MR contrast agent dynamics may force movement between individual table positions every 20–25 s. PET data sampling will have to be repeatedly stopped and resumed when the table returns to the appropriate position. Current PET/CT implementations subdivide the problem into two sequential acquisitions: first CT then PET. This will most likely not be possible in PET/MR imaging, as there will be periods of “slow” MR data acquisition where it would be inefficient not to acquire PET counts.
Regardless of whether simultaneous PET/MR ultimately makes its way into clinical routine, the efforts to integrate these two technologies will surely expand our understanding how to best exploit the complementary strengths of these two powerful imaging modalities. The role of both PET and MR in medical diagnostics, whether combined or individually, will continue to expand.
Conflict of interest
There are no conflicts of interest for any of the authors.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- 23.Kurhanewicz J, Vigneron DB, Nelson SJ. Three-dimensional magnetic resonance spectroscopic imaging of brain and prostate cancer. Neoplasia (New York, NY 2000;2:166–89.Google Scholar
- 26.Haacke EM, Brown RW, Thompson MR, Venkatesan R. Magnetic resonance imaging: physical principles and sequence design. New York: John Wiley & Sons; ; 1999.Google Scholar