Chapter 13 Magnetic Resonance Imaging

Cihan Duran, Piotr S. Sobieszczyk, Frank J. Rybicki

Magnetic resonance angiography (MRA) is widely accepted for the majority of vascular imaging applications and is the modality of choice for many of them, particularly peripheral and renal arteriography.¹^–³ Gadolinium-based contrast-enhanced (CE) MRA yields high-quality spatial resolution images with relatively short acquisition times. Detailed comparison with other imaging technologies is beyond the scope of this chapter, but general advantages and disadvantages appear in Table 13-1. These become important for many applications such as aortography, where multiple modalities (e.g., computed tomography [CT] and MRA) are routinely diagnostic and capable of providing images for planning most interventions.

^{Table 13-1} Advantages and Disadvantages of Vascular Imaging Methods

Method	Advantages	Disadvantages
MR	No ionizing radiation	Expensive hardware Acquisition requires technical expertise
CE	High signal from gadolinium-based agents Less nephrotoxic than comparable iodine-based imaging (CT)	Nephrogenic systemic fibrosis can occur rarely in patients with severe renal insufficiency
Non-contrast	Eliminates toxicity concern	Longer acquisition times compared to CE protocol Increased risk of artifacts that are largely mitigated with CE-MRA
CT	Rapid, high signal, high-quality image acquisition Less technical expertise required compared to MR	Ionizing radiation Nephrotoxicity of iodine
DSA	Intervention can be performed at time of diagnosis Highest spatial resolution	Invasive Nephrotoxicity of iodine Projectional data may be inferior to volumetric acquisitions (CT, MR) that can be viewed in any plane
Sonography	No ionizing radiation Less expensive Flow information readily obtained Portable	Better for superficial imaging Limited by artifacts from bone, air, and sonographic interfaces Lower CNR, SNR compared to CT and MRI Operator-dependent

CE, contrast-enhanced; CNR, contrast-to-noise ratio; CT, computed tomography; DSA, digital subtraction angiography; MRA, magnetic resonance angiography; MRI, magnetic resonance imaging; SNR, signal-to-noise ratio.

Basic Principles

Magnetic resonance imaging (MRI) relies upon the inherent magnetic properties of human tissue and the ability to use these properties to produce tissue contrast. Magnetic resonance imaging detects the magnetic moment created by single protons in omnipresent hydrogen atoms. Because any moving electric charge produces a magnetic field, spinning protons produce small magnetic fields and can be thought of as little magnets or “spins.” When a patient is placed in the bore of a large magnet (i.e., MRI scanner), hydrogen protons align with the externally applied static magnetic field (B₀) to create a net magnetization vector. On a quantum level, most protons will distribute randomly, either with or against the scanner’s B₀. However, a slight excess of spins aligns with the field, causing net tissue magnetization. The time required for this alignment is denoted by the longitudinal relaxation time, T1. T1 variations between tissues is used to provide contrast.

Spinning protons wobble or “precess” about the axis of B₀. The frequency of the wobble is proportional to the strength of B₀. If a radiofrequency (RF) pulse is applied at the resonance frequency of the wobble, protons can absorb energy and jump to a higher energy state. This RF pulse deflects the protons, creating a new net magnetization vector distinct from the major axis of the applied magnetic field. The net magnetization vector tips from the longitudinal to the transverse plane (transverse magnetization). The protons are “flipped” by the RF pulse, and the net magnetization vector is defined by a “flip angle.” The stronger the RF pulse applied, the greater the angle of deflection for the magnetization. Common flip angles for spin echo are 90° and 180°. For gradient echo (GRE) MRI, flip angles typically range between 10° and 70°. After the RF pulse tips the spinning protons out of alignment with the main magnetic field, new protons begin to align with the main magnetic field at a rate determined by the T1 relaxation time.

Energy is given off as the spins move from high to low energy states. The absorbed RF energy is retransmitted at the resonance frequency and can be detected with RF antennas or “coils” placed around the patient. These signals are compiled, and after mathematical processes become the MR images. Proton excitation with an externally applied RF field is repeated at short intervals to obtain signals. This MR parameter is referred to as repetition time (TR). For conventional MRI, TR is typically 0.5 to 2 seconds, whereas for MRA, TR ranges from 30 to less than 5 milliseconds. When the spins are tipped to the transverse plane, they all precess in phase. The speed of wobbling depends on the strength of the magnetic field each proton experiences. Some protons spin faster while others spin slower, and they quickly get out of phase relative to one another. Throughout the dephasing process, the MR signal decays. This loss of phase is termed T2 relaxation time or transverse relaxation. T2, like T1, is unique among tissues and is used for image contrast. In addition to the intrinsic T2 of tissue, inhomogeneity of B₀ results in rapid loss of transverse magnetization. The relaxation time that reflects the sum of these random defects with tissue T2 is called T2*. To obtain an MRI signal, these spins must be brought back in phase and produce a signal or echo. The time at which it happens is referred to as echo time (TE). In spin echo imaging technique, the echo is obtained by using a refocusing 180° RF pulse, after which the spins begin to dephase. Another 180° RF pulse can be applied to generate a second echo and so on. Signal loss at longer echo times reflects tissue T2. In GRE imaging, the echo is obtained by gradient reversal rather than RF pulse. Because this includes effects from tissue homogeneity, TE-dependent signal loss reflects T2*. Recently, GRE sequences (balanced GRE steady-state free precession [SSFP]) have been developed that are insensitive to magnet field inhomogeneities and reflective of actual tissue T2.

Longitudinal and transverse magnetizations occur simultaneously but are two different processes that reflect properties of various tissues in the body. Since T1 measures signal recovery, tissues with short T1 are bright, whereas tissues with long T1 are dark. Fat has a very short T1. In contrast, T2 is a measure of signal loss. Therefore, tissues with short T2 are dark, and those with long T2 are bright. Simple fluids, such as cerebrospinal fluid and urine, have long T2. To differentiate between the tissues based on these relaxation times, MR images can be designed to be T1-weighted, T2-weighted, or proton-density weighted. Exogenous contrast such as gadolinium-based agents are routinely used to alter tissue conspicuity. Spatial encoding of signals obtained from tissues is required for imaging. Additional external time-varying magnetic fields are applied to spatially encode the MR signal. Spatially dependent gradients are used to locate the MR signal in space. In two-dimensional (2D) MRI, these are slice-selection, frequency-encoding, and phase-encoding gradients. In three-dimensional (3D) MRI, the slice-selection gradient is replaced by a second phase-encoding gradient.

Magnetic resonance echoes are digitized and stored in “k-space” composed of either two axes (for 2D imaging) or three axes (for 3D imaging). K-space represents frequency data and is related to image space by Fourier transformation. An important feature of k-space is that tissue contrast is determined by the center of k-space (central phase encoding lines), whereas the periphery of the k-space encodes the image detail. The order in which k-space lines are collected can be varied, strongly influencing tissue contrast. For example, in CE-MRA, the central contrast-defining portion of k-space may be acquired early in the scan (centric acquisition) during peak intraarterial contrast concentration to maximize arterial contrast. In addition to simple line-by-line k-space acquisition schemes, more complex schemes have been described. In spiral imaging, data acquisition begins at the center of k-space and spirals to the periphery. Slice-selective gradients applied along the z-axis will form axial images. Those along the y-axis will yield coronal images, and the x-axis gradients will provide sagittal images. An oblique slice can be selected by a combination of two or more gradients.

Magnetic Resonance Angiography Techniques

Magnetic resonance imaging relies on selective imaging of moving blood where signals from blood vessels are maximized, whereas signals from the stationary tissues are suppressed. Algorithms then enable reformatted images similar to those found in conventional x-ray angiography (Table 13-2).

Table 13-2 Types of Magnetic Resonance Angiography Sequences

Magnetic resonance angiography methods can depict blood as either black or white. For those “black-blood” methods that use standard spin echo (SE) sequences (Fig. 13-1A), the excitation RF pulse is applied at 90° and followed by a refocusing pulse at 180°. If the imaging slice cuts across a vessel, then depending on the flow velocity and time interval between the pulses, the blood volume originally excited by the first pulse may not “see” the second pulse. This results in a black-appearing signal void in the vessel lumen. The use of thin sections or long echo times can further emphasize this flow void. This technique allows detailed examination of arterial wall morphology. Fast spin echo (FSE) sequences, in which a long train of echoes is obtained by use of repeated 180° pulses, produce images more rapidly. The double inversion recovery (DIR) FSE offers a new approach to enhance black-blood sequences. The technique uses two consecutive inversion pulses: the first nulls or blackens the blood everywhere in the coil, and the second restores magnetization in the slice being imaged. Between these pulses and image production, blood within the slice is replaced by nulled blood from outside. This produces more reliable black blood than conventional approaches, making this sequence ideal for examining wall thickness, dissection flaps, and the presence of mural thrombus or inflammation.⁴ This provides a clear advantage over traditional x-ray angiography.

Figure 13-1 “Black-blood” and “bright-blood” imaging techniques. A, Cross-sectional T1-weighted image of ascending and descending aorta; lumen appears black. B, Time-of-flight (TOF) image of carotid-vertebral system. C, Maximum-intensity projection (MIP) reconstruction of arterial phase of contrast-enhanced magnetic resonance angiography (CE-MRA) images in same patient shows normal carotid-vertebral arteries at higher spatial resolution than corresponding TOF acquisition.

“Bright blood” MRA techniques use GRE sequences and are generally divided into those measuring signal amplitude (time-of-flight [TOF]) and those based on phase effects (phase-contrast [PC]). In each GRE sequence, a single RF pulse is applied in short time intervals, eliminating signal loss due to flow void. The stationary protons occupying a given tissue slice do not have sufficient time to relax to their equilibrium state.

TOF-MRA techniques depend on the inflow of unsaturated protons in blood from outside the field of view (FOV) into the stationary tissue within a section already saturated by its exposure to repeated RF pulses. These “saturated” protons are unable to contribute signal to the image. The signals in the stationary tissues of GRE images used in MRA are therefore typically low. The “unsaturated” protons in blood flowing into the imaging plane have not experienced the RF pulses and yield maximum signal. The unsaturated blood appears bright compared to background tissue (Fig. 13-1B). The time required for blood to flow through an image slice and its effect on the resulting signal is known as TOF. Saturation of signals can occur in vessels with slow-moving blood as a result of repeated RF excitation in the acquisition plane. This can create artifacts in vessels with stenotic lesions or reduced blood flow.

Time-of-flight techniques can be obtained in 2D or 3D. The 2D TOF utilizes multiple sequentially acquired, overlapping thin slices to form an image. The patient is instructed to hold his/her breath to minimize motion artifact. However, spatial misregistration may occur if patients cannot hold their breath at the same level each time. Thus, only one or two slices are typically acquired per breath-hold. 2D TOF has good sensitivity for identifying vessels with slow flow because blood must move only 3 to 5 mm to refresh a slice. 3D TOF consists of GRE acquisition of a volume into which blood is flowing. The advantage of this technique is higher signal-to-noise ratio (SNR) and improved resolution. The thick volumes of tissue imaged require rapid flow to fully refresh signals within the arteries. The technique is flow dependent and superior in vessels with rapid steady flow without respiratory motion. Additional saturation pulses can be applied to eliminate signal from veins. Segmented GRE sequence with cardiac triggering can be used to eliminate arterial pulsation artifacts.

A successful TOF image requires the section to be thin enough to allow for sufficient inflow between RF pulse repetitions, but thick enough to ensure adequate SNR and anatomical coverage. Section thickness of 3 to 4 mm is used for large vessels, and 1 to 2 mm for smaller vessels. Spatial presaturation pulses are applied above or below the imaged slice or volume to eliminate unwanted signal from arteries or veins, depending on which part of the vascular tree is being imaged. Optimal TR for TOF is 20 to 50 milliseconds. Short TR keeps background tissues saturated, but it must be long enough to allow for satisfactory inflow of unsaturated blood between successive repetitions. The best flip angle is usually 30° to 60°. With phasic flow in the extremities, systolic flow signal may be increased (because of greater transverse magnetization created), and distal flow may be decreased, creating view-to-view intensity changes and phase artifacts from pulsatile variations. This pulsation artifact is greatest at higher flip angles. Cardiac gating can be used to minimize these artifacts at the expense of increased imaging time.

Whereas TOF uses differences in signal amplitude to differentiate between stationary and flowing spins, the PC technique observes the phase shifts of signals. Moving spins experience different phase shifts in the presence of the applied magnetic fields used in MRA. Strength and orientation of the applied magnetic field are varied to encode different phase shifts for flowing protons relative to stationary protons. The faster the spins are moving, the greater their phase shift, and protons of flowing blood may be discriminated from stationary protons. The phase shifts result in a contrast between moving and stationary tissues and form the basis for PC imaging. Pairs of images are acquired that have different sensitivities to flow and are then subtracted to cancel background signal, leaving only the signal from flowing blood. Phase shift is proportional to velocity, allowing flow quantification with this modality. Phase-contrast acquisitions may be acquired in two or three dimensions; although used rarely in angiography today, Phase-contrast offers a reliable way to quantify amount and direction of flow. It requires long imaging times: two data sets in each direction are acquired by using flow-encoding gradients of opposite polarity, and up to three measurements in the orthogonal planes are needed to image flow in all directions.

Visualization of the arterial system with PC and TOF is adequate ⁵ but has limitations. Acquisition times can be long and prevent imaging within the time span of a single breath-hold. This increases the chance of movement artifacts. Some of the limitations are caused by flow-related artifacts such as in-plane saturation and phase dispersion. Flow-based imaging also has limits in areas of slow flow, such as aneurysms. Overgrading of stenotic lesions is most commonly a manifestation of signal loss in the areas of complex flow. Undergrading is a matter of inadequate spatial resolution. Complex turbulent flow patterns in areas of stenoses can create signal loss and mimic a critical lesion. This is due to “intravoxel dephasing.” An accelerated flow across a stenosis consists of a wide distribution of velocities and thus a large distribution of proton phases. In the smallest volume element, a “voxel” of the image, this distribution of phases can result in cancellation rather than coherent addition of signals, accounting for the presence of signal voids at the site of stenosis. A short TE minimizes flow-phase dispersion artifacts. Phase dispersion is further decreased when voxel size is minimized using thin sections. Small voxels and short TE are most easily obtained with 3D TOF methods. The biggest drawback of the thick volumes used with 3D techniques is that slow or recirculating flow can become saturated. The MOTSA (multiple overlapping thin-slab acquisitions) technique of sequential 3D TOF gives better flow enhancement than single-slab 3D TOF techniques and less dephasing than 2D techniques. However, the need for substantial overlap of adjacent slabs increases acquisition time.

Contrast-Enhanced Magnetic Resonance Angiography

The introduction of CE-MRA has revolutionized MRA.^6,⁷ This technique overcomes many of the limitations of traditional bright blood modalities: respiratory motion artifacts, poor SNR, and flow and saturation-related artifacts (Fig. 13-1C). Gadolinium increases the signal intensity of blood on contrast-enhanced 3D T1-weighted (spoiled) GRE images. Blood contrast is not flow dependent. It is determined by the concentration of contrast agent within the arterial system while imaging data are being collected. Reliable images can be acquired irrespective of whether flow is laminar, turbulent, or stagnant. This technique acquires large-volume data sets in coronal or sagittal orientation within a single breath-hold during the first pass of the contrast material. The contrast agent, gadolinium, is a heavy metal but becomes inert when bound to a chelator. Intravenous (IV) administration of gadolinium-diethylenetriamine pentaacetic acid (DTPA) results in a marked reduction of the T1 or longitudinal relaxation time of blood, therefore reducing the effects of spin saturation. Signal reduction is also problematic in 3D TOF sequences. Moreover, the very short TE reduces spin dephasing and allows accurate evaluation of vascular stenoses.

Multiple refinements have resulted in a technique that is much faster than TOF-MRA. The development of high-performance gradient systems with ultra-short repetition TR and TE has shortened acquisition time in CE-MRA to allow imaging within a single breath-hold and minimize motion artifacts. Administration of agents shortening T1 allows selective visualization of contrast-containing structures and better visualization of circuitous collaterals. Digital subtraction, spoiling, and fat saturation techniques suppress background signal and enhance signal from the contrast agent in the vessels. The subtracted data sets can be postprocessed to provide 3D projectional images. CE-MRA still provides a luminogram, and conventional or FSE images are needed for a complete study so that true lumen diameter and presence of thrombus can be established.

Optimal images are generated when gadolinium concentration is highest in the vessel of interest. To make blood bright compared to background tissues, the gadolinium bolus must be administered in a way that ensures the majority of the contrast to be present in the arterial tree. This requires exact timing of the arrival of the gadolinium bolus. Acquisition prior to contrast arrival creates a “ringing” artifact, whereas late acquisition creates venous and tissue enhancement, contaminating the arterial signal. This is especially problematic in MRA of the extremities, where the images are obtained in multiple segments. Contrast transit time can be affected by low cardiac output, valvular regurgitation, large abdominal aneurysms, and flow-limiting stenoses. Proper timing can be achieved by empirical estimation of transit time or a test bolus in the anatomical field of interest. Alternatively, with automated triggering, a pulse sequence can be designed to sense the arrival of contrast and automatically trigger image acquisition. Magnetic resonance fluoroscopy allows the user to visualize arrival of the contrast bolus directly on the image and manually trigger the start of the scan. Areas that require higher spatial resolution, such as the lower extremities, also need larger doses of contrast for longer acquisition times.

Imaging during the arterial phase of gadolinium infusion takes advantage of higher arterial SNR and eliminates overlapping venous enhancement. This is a brief moment in time, but several methods allow slower MR image acquisition to capture that moment. The phase reordering (mapping k-space) technique acquires central k-space data (i.e., the low spatial frequency data) when contrast concentration is high in arteries but lower in veins. This allows a relatively long MR acquisition to achieve the image contrast associated with the shorter arterial phase of the contrast bolus. It is critical to time the contrast bolus to achieve maximum arterial gadolinium concentration during acquisition of central k-space data.

CE-MRA is limited by venous and soft-tissue enhancement. Contrast media not only passes into venous structures, dependent on the arteriovenous transit time of the tissue, but also rapidly leaks out of the vascular compartment, creating significant tissue enhancement. New “blood pool” agents, which are currently undergoing clinical trials, are retained within blood vessels and selectively enhance the blood pool on T1-weighted MR images. These use either gadolinium compounds that bind to albumin, or are large enough to stay within the vascular space or ultra-small iron particle. Another agent, gadobenate, has a higher T1 relaxation time because of its capacity for weak and transient interaction with serum albumin. This may enhance vascular signal intensity and thus increase diagnostic efficacy at doses comparable to those used for current gadolinium agents. It is approved for imaging use in Europe but is under clinical investigation in the United States. It provides a higher and longer-lasting vascular signal enhancement in the abdominal aorta compared with gadolinium, which does not interact with proteins.⁸

Gadolinium-based contrast has a very favorable safety profile. However, gadolinium is nephrotoxic. For patients with underlying chronic renal insufficiency, gadolinium chelates can cause acute renal failure. Nephrogenic systemic fibrosis (NSF) is linked to gadolinium-based contrast agents ⁹ and largely involves the skin, though it may also affect the muscle, joints, or internal organs such as the lungs, liver, and heart in patients with renal failure. Nephrogenic systemic fibrosis occurs in patients with severe renal disease who are exposed to high doses of gadolinium agents, or in patients who receive multiple standard doses of contrast agents in a short period of time. The reported prevalence of NSF among patients with glomerular filtration rate (GFR) less than 30 mL/min is 3% to 5%.¹⁰ Therefore, the MR protocol should aim to minimize contrast volume, especially in patients with moderate to severe renal failure.¹¹

Metal objects, such as surgical clips, lead to susceptibility artifacts in MRA. The increasing use of stents has important implications for MRA. Cavagna et al. evaluated CE-MRA of seven stent types in the aortic, iliac, and popliteal positions.¹² Few of the commonly used stents permitted visualization of the lumen. Susceptibility artifact results in significant signal loss that can preclude proper visualization of the stent lumen even with gadolinium-enhanced MRA. Some nitinol, tantalum, or polytetrafluoroethylene (PTFE)-based devices, on the other hand, cause less artifact on CE-MRA.

Postprocessing Techniques

Magnetic resonance data can be viewed as source images or be displayed in projections with any orientation. Image postprocessing allows reformation in any desired plane to improve conspicuity of overlapping vessels (Table 13-3). The origins of the left common carotid artery (CCA) and left subclavian artery, for example, can overlap in coronal projections, whereas the origins of the right subclavian artery and right CCA can overlap in some oblique views. The renal ostia are usually best seen in either coronal or slightly oblique view. The celiac axis and superior mesenteric artery (SMA) are best depicted on sagittal projections. One advantage of MR versus digital subtraction angiography (DSA) is that the latter may require multiple injections to assess the origins of these vessels. The details of image interpretation are beyond the scope of this review, but source image data, multiplanar reconstruction (MPR), maximum-intensity projection (MIP), and volume rendering (VR) are used (Fig. 13-2). Source images are the initial reconstructions and should be used for problem solving and to confirm findings. Interpretation often begins with a vascular survey using MPR and MIP data sets. Multiplanar reconstructions are very useful in volumetric acquisitions because the desired imaging plane can be prescribed to enhance vascular separations. Subtracted MIPs are routinely created from CE-MRA. The noncontrast (mask) images are subtracted from the enhanced images, and resulting high SNR data sets undergo projection of maximum intensity. By performing the projections of all angles around the z-axis of the patient, the data sets can be viewed in cine. These projections are referred to as rotating MIPs.

^{Table 13-3} Types of Postprocessing Techniques

Technique	Description
MPR	Production of cross-sectional images in planes different from acquisition plane
MIP projection	Production of full- or partial-volume images along any desired axis from a stack of image slices
Volume rendering	Manipulation of MRI slices to produce full volumetric images; structures segmented for viewing by application of intensity thresholds and removal of unwanted structures

MIP, maximum intensity projection; MPR, multiplanar reconstruction; MRI, magnetic resonance imaging.

Figure 13-2 Postprocessing techniques; contrast-enhanced magnetic resonance angiography (CE-MRA) of abdominal aorta and branches.

A, Coronal thin-section image of abdominal aorta. Summation of these images, projected with maximum intensity, is used for (B) coronal maximum-intensity projection (MIP) that includes normal renal and mesenteric arteries. C, Axial multiplanar reconstruction (MPR) image at level of left renal ostium does not include entire extent of both renal arteries, giving false impression of a proximal right renal artery occlusion (arrow). D, Three-dimensional volume-rendered (3D-VR) image shows entire course of abdominal aorta and its branches.

Clinical Applications

Extracranial Carotid and Vertebral Arteries

Atherosclerosis, dissection, and inflammatory diseases affect the extracranial carotid and vertebral arteries. Time-of-flight methods (2D and 3D) have been largely replaced by CE-MRA because CE-MRA allows imaging of the entire course of these vessels (Fig. 13-3). A problem with TOF imaging is turbulent flow at or near the carotid bifurcation where most of the lesions occur; this may lead to overestimation of lesion severity. Gadolinium-enhanced MRA has greater SNR than noncontrast imaging, is less sensitive to intravoxel dephasing from turbulence, and does not have signal loss from saturation effects. Phase-contrast sequences can supplement anatomical data for flow direction or quantification.