CHAPTER 17 Computed Tomography and Magnetic Resonance Imaging of the Brain
Computed Tomography of the Brain
History and Fundamentals
Computed tomography (CT), also called computed axial tomography (CAT), was developed in the early 1970s by Sir Geoffrey Hounsfield and his colleagues in England.1 It was possibly the single most important advance in medical imaging since the discovery of x-rays by Professor Wilhelm Roentgen. It represented the first commercially available imaging equipment that used the emerging technologic advances in computing to generate digital images displayed in gray scale. Its development revolutionized the evaluation of patients with neurological diseases and allowed noninvasive visualization of the inner body, which led to important diagnoses of diseases and abnormalities and played a key role in the diagnosis, management, and treatment of patients on a daily basis in the practice of medicine all over the world.2 Although its place in imaging of the brain and spine have been somewhat supplanted by another revolutionary technology known as magnetic resonance imaging (MRI), CT remains a workhorse and important first study of choice in many aspects of neurosurgery. Furthermore, important advances in CT technology during the past decade, such as multidetector configurations in newer CT scanners and ever increasing speed of computer technology that now allow very fast CT scanning of a patient in seconds rather than minutes, have resulted in a strong resurgence in its use. Such advances have led to the development of CT angiography (CTA) and perfusion CT (pCT), which have become important in noninvasive evaluation of cerebrovascular diseases. In addition, portable CT scanners can now provide high-quality images for point-of-care imaging in an intensive care unit setting and thereby avoid potential risks associated with transport of critically patients.
CT can be performed in various planes that depend on patient position and the CT gantry angle within its limited arc. For example, direct coronal-plane CT imaging of the paranasal sinuses or brain can be performed with the patient in a supine, “hanging-head” position with the head of the patient literally hanging over the edge of the CT scanner table or with the patient in the prone position and the neck hyperextended. However, most commonly, CT imaging of the brain and spine is performed in the axial plane with the patient in a supine position on the scanner table and the head and neck in a neutral position. The need for a direct coronal patient position is less important since the advent of high-resolution multiplanar reconstruction capabilities on newer generation CT scanners. These reconstruction capabilities can generate axial images in 0.5- to 0.6-mm increments, which can then be reformatted into the sagittal, coronal, and oblique planes with image quality nearly identical to that obtained from direct scanning.3,4
CT is most often the first study of choice for evaluation of a patient with suspected acute intracranial pathology because of its ready availability, ease of use, short acquisition time, and high sensitivity for detection of acute hemorrhage and fractures. It can provide a wealth of information about the brain, including ventricular size, presence of brain edema, mass effect, presence and location of hemorrhage or masses, midline shift, evolving ischemic injuries, fractures, benign and malignant osseous pathology, and the paranasal sinuses. Its availability and short acquisition time also allow frequent repeat scanning of the brain, which can contribute to the management and follow-up of patients in the acute, subacute, and chronic phases in both inpatient and outpatient settings.5–7
In neurosurgery, CT of the head is used for preoperative and postoperative evaluation of patients for hemorrhage, infarction, hydrocephalus, mass effect, fracture, and postsurgical assessment.8–12 CT is the study of choice to evaluate for acute hemorrhage because it has higher sensitivity and specificity for this indication than MRI does. Intracranial hemorrhage is typically described in terms of its location within the head, such as epidural, subdural, subarachnoid, intraventricular, and parenchymal, with each of these different types of hemorrhages having sufficiently distinct appearances and locations. Epidural hemorrhage has a biconvex contour of its borders (Fig. 17-1A) in relation to the cranial vault and adjacent brain parenchyma and is usually the result of acute trauma associated with an acute fracture across branches of meningeal arteries that hemorrhage into the epidural space. Less commonly, rapid venous hemorrhage into the epidural space may occur and cause an epidural hematoma. The extent of an epidural hematoma is usually limited by periosteal dural insertions at the major sutures. However, an epidural hematoma can extend across the midline in the frontal region anterior to the coronal suture because it is not limited by the dural reflections within the anterior interhemispheric fissure (Fig. 17-1C). A subdural hematoma (SDH) is more common than an epidural hematoma, particularly in older patients, and is generally associated with acute head trauma with or without an associated fracture. Its shape is different from an epidural hematoma because its deeper border against the brain parenchyma is concave and approximates the contour of the adjacent cerebral hemisphere convexity. An acute SDH is typically a result of venous hemorrhage and is not limited by the periosteal dural insertions at the major sutures. However, it is limited by the midline dural reflections within the interhemispheric fissure. The density of the blood in acute, subacute, and chronic SDH changes over time from hyperdense, to isodense, to hypodense (Fig. 17-2). However, a hyperacute SDH or an acute subarachnoid hemorrhage (SAH) in the presence of coagulopathy may sometimes appear isodense or hypodense.
Trauma is the most common cause of SAH, whereas rupture of an intracranial aneurysm is the most common nontraumatic cause of SAH. SAH extends freely within the subarachnoid spaces around the cerebral hemispheres, brainstem, and cerebellum and frequently, by reflux of cerebrospinal fluid (CSF), extends into the intraventricular spaces. It often leads to acute, subacute, or chronic hydrocephalus because the blood products disrupt and obstruct the normal CSF drainage pathways (Fig. 17-3).
Parenchymal hemorrhages have many causes, including trauma, hypertension, vascular anomalies such as arteriovenous malformation (AVM) or cavernoma, infarction, neoplasm, infection, or vasculitis. They can be small or large and single or multiple, and patient prognosis depends on the cause, number, size, and associated mass effect of the hemorrhage, among other variables (Fig. 17-4).
Computed Tomographic Angiography
Advances in CT scanner technology have allowed an ever improving capacity for higher resolution images in the submillimeter range with shorter acquisition times. Such technologic improvements have led to imaging techniques such as CTA, which permits relatively noninvasive imaging of the major arteries and veins of the neck and brain after an intravenous injection of iodinated contrast material rather than the traditional catheter-based intra-arterial angiogram technique. This venous injection helps avoid the small risk for complications such as vascular dissection, renal injury, allergic reaction, and iatrogenic embolic strokes associated with traditional catheter angiography. CTA of the neck or brain is performed with a multidetector CT scanner, which allows rapid dynamic imaging of the anatomy of interest after a bolus intravenous injection of iodinated contrast material through a large-bore intravenous catheter (i.e., 18 gauge). Typically, submillimeter axial images are obtained and then reformatted into 2D sagittal and coronal image data sets at 1- to 2-mm intervals. 3D reconstruction images are usually obtained, but interpretation of the study is based primarily on the original axial data set and the 2D sagittal and coronal reformatted images. The diagnostic sensitivity and specificity of CTA approach that of catheter angiography for both the extracranial and intracranial vasculature.13,14 Although CTA cannot entirely replace traditional catheter angiograms, it is a very useful noninvasive screening study for the evaluation, management, and follow-up of patients with definite or possible aneurysms, as well as the evaluation of vasospasm, AVMs, traumatic dissection, stroke, and carotid or vertebral artery atherosclerotic stenosis (Figs. 17-5 and 17-6).15
Perfusion Computed Tomography
pCT is an example of new advances in imaging that provides physiologic information in addition to anatomic information. pCT is performed with the latest-generation multidetector CT scanners, which allow very rapid CT imaging of a particular anatomy, such as the cerebral hemispheres. During a bolus intravenous injection of iodinated contrast material at a rate of 4 to 5 mL/sec, rapid serial CT images of a chosen volume are obtained in multiple phases over an approximately 1-minute period. At the end of this acquisition, multiphase time-density curves corresponding to each voxel are generated within a 2D image of a multilevel image data set. The data from these images are further postprocessed with a mathematical algorithm that allows displays of the data in color maps representing such physiologic cerebral perfusion parameters as cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT). The CBF, CBV, and MTT maps generated from this CT technique are, in part, quantitative; that is, the numerical values obtained from these images may be expressed in mL/100 g per minute for CBF, mL/100 g for CBV, and seconds for MTT.16 pCT technology has been validated against other proven in vivo techniques such as xenon-enhanced CT and positron emission tomography (PET).17,18
pCT has been used to evaluate acute stroke, central nervous system (CNS) neoplasms, and ischemic sequelae of SAH-related vasospasm. The most common use of the pCT technique is for the evaluation of a patient with an acute stroke. The various color maps of cerebral perfusion help determine the presence of salvageable ischemic penumbra during the first few hours after stroke, which may lead to more aggressive therapy such as an intra-arterial thrombolysis or thrombus extraction to permit rapid recanalization of occluded large intracranial arteries such as the supraclinoid segment of the internal carotid artery (ICA) or M1 segment of the middle cerebral artery (MCA) (Fig. 17-7).
The availability of physiologic data also helps in the diagnosis, management, and treatment of patients with a ruptured aneurysm and subsequent vasospasm, which may contribute to acute or subacute ischemic injury. Evaluation of these patients has typically relied on serial clinical assessment, non–contrast-enhanced head CT, and transcranial Doppler (TCD) ultrasound. There are recognized limitations with this evaluation protocol; in particular, non–contrast-enhanced CT and TCD may not accurately reflect the state of cerebral perfusion at an early enough stage to allow successful intervention for reversal of oligemia and ischemia. Baseline pCT and follow-up pCT can demonstrate the size and extent of brain areas at risk for stroke in patients in a neurological intensive care unit often before symptoms develop and permanent infarction occurs (Fig. 17-8). This early detection of at-risk areas may in some patients permit earlier medical and catheter-based intervention for vasospasm and thus prevent delayed ischemic injury.19–21
Magnetic Resonance Imaging of the Brain
Physics and Techniques of Magnetic Resonance Imaging
History
The interaction of the intrinsic magnetic moment of the nucleus with an externally imposed magnetic field results in the phenomenon known as nuclear magnetic resonance (NMR). Two independent groups, Felix Bloch working with liquid water22 and Edwin Purcell working with solid paraffin,23 detected the hydrogen nucleus resonance in 1946 in bulk matter. Bloch further described the processes and time constants (T1 and T2) by which the resonance would dissipate.24 This set the stage for the eventual development of MRI 30 years later. Between 1946 and 1976, NMR became a useful laboratory tool for probing molecular structure. Laboratory NMR instruments had small spaces for the sample, usually a small test tube. Use of NMR for larger objects, such as humans, required the development of larger magnets with larger sample spaces. The term magnetic resonance imaging is now used rather than NMR to allay patient anxiety about a test that has the word “nuclear” in it.
Basic Physics of Magnetic Resonance Imaging
Creating the Signal
When immersed in the strong, constant magnetic field, the spins in the sample experience a slight polarization. This polarization, known as M0, increases with increasing magnetic field. At 1.5 T, this polarization is very slight, about 1 × 10−5. This translates to just 10 in 1 million nuclei being polarized.25 The polarization competes with the randomizing effect of the thermal vibration (Fig. 17-9). It is only the polarized nuclei that contribute to the MR signal; hence, MRI systems with higher field strength produce better images.
We will use the classic model created by Bloch to describe the motion of the spins. Spins that align with the main field precess around the direction of the main field in a manner similar to the spinning of a toy top (Fig. 17-10). The rate of this precession is a product of the intrinsic magnetic moment (the gyromagnetic ratio) of the spin and the strength of the main magnetic field. The rate of precession is known as the Larmor or resonant frequency.26 The spins aligned along B0 are rotated into a plane transverse to the direction of the main magnetic field by the action of a time-varying magnetic field called B1 (Fig. 17-11). The B1 field is created by the radiofrequency (RF) transmitter and the antenna, known as the RF coil. The frequency of the B1 field matches the Larmor frequency of the spins. The B1 field is of brief duration, typically 1 to 10 msec, and is thus referred to as an RF pulse. The angle through which the spin is rotated is called the flip angle. The flip angle depends on the duration and amplitude of the RF pulse.
Detecting the Signal
To detect the MR signal, an RF coil is placed as shown in Figure 17-12. This may be the same coil used to transmit the B1 pulse. A time-varying magnetic field will be created at the coil by the magnetic field of the precessing spins (rotated in the transverse plane) as they pass by the RF coil. Magnetic induction (Faraday’s law) causes the RF coil to produce an electric current, which can then be amplified and detected (Fig. 17-13). An interesting and important disparity should be noted. The amount of power used to produce the RF pulse is in the range of 100 to 20,000 W. The signal that is received from the object is on the order of 10−12 W. This received signal is no greater than the signals from radio or television stations, among other sources. To prevent interference from these external sources, MRI systems are enclosed in electrically shielded rooms, which are six-sided copper boxes.
Physics: Localizing the Signal
To this point, the sample has been polarized and excited and a signal detected, but the location of the spins that created the signal remains unknown. Suppose that two objects are in the magnetic field, both of which create a signal. To force each object to give off a unique signal, the magnetic field can be modified to vary as a function of position along the x-axis (Fig. 17-14). The resonant frequency of the spins is a function only of the magnetic field at that point in space. Thus, the spatial origin of the signal can be determined by the frequency of the received signal. In practice, this is done by creating a linear gradient that adds or subtracts to the main field as a linear function of offset from the origin. An MRI system has three gradients: x, y, and z. The gradient system serves two purposes in the MRI system. The first is to limit the excitation to a plane or slab (Fig. 17-15). If an RF pulse is transmitted during the time that a gradient is applied, only a slice or slab will be excited, the thickness of which depends on the amplitude of the gradient and the bandwidth of the RF pulse. The second purpose is to encode the spatial location of spins to form the MR image. In a slice, the two directions of encoding required are frequency and phase. Location along the frequency-encoding axis is accomplished by applying a gradient during the signal readout time. The orthogonal axis is encoded by applying a gradient somewhere between the time of excitation and reception and is called phase encoding. One unique value of phase encoding is applied every time that the readout gradient is applied. Thus, the excitation and readout must be repeated to make an image. The rate at which the excitation is repeated is the TR time.
The Origin of Image Contrast
The intensity of a voxel in an image arises from the three principal factors. The first is the number of protons in the voxel, known as proton density. If this were the only image mechanism, MRI would be little more than a CT scanner. However, intensity in the voxel also depends on the relaxation rates of the spins. The first is T1, or the longitudinal relaxation rate (Fig. 17-16A). After excitation, the magnetization in the slice returns along the axis of the main magnetic field by interaction with other nonmoving hydrogen spins, typically those attached to large molecules. The magnetization returns along B0 as an exponential function of the ratio of T1 and the rate at which the excitation is repeated, or TR. The second relaxation rate is T2 (Fig. 17-16B). This rate describes the rates at which spins that have been excited into the transverse plane lose coherence. Each spin precesses at a rate that is determined by the magnetic field at the location of that spin. Macroscopic and microscopic field gradients, created by differences in the magnetic susceptibility of tissue, cause some of the excited spins to precess faster and some slower. Eventually, the spins are spread out uniformly, which produces no signal in the RF coil used to detect the spins.
Spin Echo
After the spins have been rotated into the transverse plane by the initial 90-degree RF pulse (Fig. 17-17A), they begin to lose coherence because of the effects of local inhomogeneities (contributed by changes in tissue), inhomogeneities secondary to imperfections in the B0 field, and diffusion of water molecules (Fig. 17-17B). If a second RF pulse is transmitted at twice the amplitude of the first pulse, the relative direction of the spins can be reversed (Fig. 17-17C). Then, at the echo time, the spins will have nearly regained coherence (Fig. 17-17D). The effects of magnetic field inhomogeneities are thus cancelled out. One loss, that caused by diffusion, cannot be reversed but is usually negligible in routine imaging. The resulting coherence produces the spin echo as described by Erwin Hahn in 1950.27 If the 180-degree pulse is not used, the signal decreases with increasing echo time as T2*. With the 180-degree pulse included, the decrease is T2, with T2 always being greater than T2*.
The spin echo sequence is the mainstay of routine clinical imaging. By adjusting TE and TR times, proton density–, T1-, or T2-weighted images can be selected (Table 17-1).
| SHORT TE | LONG TE | |
|---|---|---|
| Short TR | T1 weighted | Mixed contrast—do not use |
| Long TR | Proton density weighted | T2 weighted |
TE, echo time; TR, repetition time.
Gadolinium Contrast
Exogenous contrast enhancement is now a routine part of MRI. The most widely used is a gadolinium chelate. The gadolinium atom is strongly paramagnetic and acts to shorten the T1 relaxation time of nearby water protons in blood.28 The agent does not pass the blood-brain barrier and thus becomes a marker for abruptions in the blood-brain barrier caused by neoplasm, infection, trauma, and infarction. The T1-shortening effect is also used for rapid MRA, which is performed by using a fast scanning protocol after the bolus injection of a contrast agent. In the brain, bolus injection of gadolinium with repeated echo planar imaging (EPI) has been used to image perfusion29 and the blood volume of tumors.30
Fast Spin Echo
Fast spin echo (FSE) sequences decrease scan time by increasing the efficiency of data collection.31 The increased efficiency can be used to either decrease scan time or increase the signal-to-noise ratio of the resulting images. Improved scan efficiency has resulted in the frequent application of FSE sequences in radiology, particularly in imaging of the CNS.
To understand this improved efficiency, it is necessary to understand how the data are collected. In a conventional spin echo sequence (Fig. 17-18A), a single line of k-space samples (called a view) is collected. For images that are not fractional excitation, the number of views is equal to the matrix size in the phase that encodes image direction. The next view is collected at TR time later, when the sequence is repeated. The entire k-space matrix must be collected before the images can be reconstructed. The TR time is set in accordance with the desired contrast (Table 17-1). The FSE sequence collects multiple views in each TR time (Fig. 17-18B). The number of views collected per TR time is known as the echo train length (ETL). An FSE sequence with an ETL of 4 has a total scan time a fourth that of a conventional spin echo sequence with equivalent TR. The ETL can be equal to the number of views in the sequence. This allows the entire image to be collected in a single TR time.
The improved efficiency of FSE sequences comes at the expense of image contrast purity because each view is collected at a different echo time (Fig. 17-18B). The effect is to create image blurring that worsens with increasing ETL or decreasing tissue T2. The relatively long T2 times of tissues in the neuraxis allow ETL values of 16 to be used routinely. To image uncooperative patients, single-shot FSE sequences can be used, but with some increase in image blurring.
Inversion Recovery
Image contrast can be further manipulated by transmitting a 180-degree RF pulse before the pulse sequence. The effect of using this pulse is to rotate the spins from their orientation along +z to −z (Fig. 17-19A). The longitudinal magnetization (Mz) signal regrows from −z to +z at a rate determined by the T1 of the spins. The regrowth is plotted in Figure 17-19B for several tissues. If the 90-degree RF pulse is transmitted at the time at which Mz is 0, the tissue will produce no signal. Inversion recovery can be used to increase contrast between structures, such as gray and white matter, or to null signals that arise from protons of a known T1. A coronal image of a volunteer imaged at multiple T1 times is shown in Figure 17-20. The round object above the volunteer’s head is a jar of mayonnaise. Fluid-attenuated inversion recovery (FLAIR) imaging32 uses this method to null CSF while preserving signal from edematous tissue. The inversion pulse may be applied to a spin echo or gradient echo sequence.
Echo Planar Imaging
The readout gradient used to form the echo in a gradient echo sequence can be repeated to collect additional k-space views in a manner similar to FSE.33 Between repetitions of the readout gradient, a small phase-encoding gradient is applied that allows a different view to be collected during the subsequent readout gradient (Fig. 17-21). Only a single excitation is used to collect all the views in k-space. The entire image can be collected in approximately 40 msec. Applications that are highly sensitive to even minor patient motions use this sequence. Like FSE, EPI suffers from blurring caused by acquiring views at different echo times, but without rephasing of the RF 180-degree pulses. This effect limits the resolution obtainable. Although an FSE image may use a matrix of 256 × 256, EPI is limited to 128 × 128 or more typically 64 × 64 over the same field of view. Despite these limitations, EPI is essential for diffusion, perfusion, and functional MRI (fMRI).
Diffusion-Weighted Imaging
Diffusion of water in the brain depends on the integrity of the cellular membrane and the osmotic balance of the neurons and astrocytes.34 In ischemic processes, the diffusion of water first decreases because of cellular swelling and thus decreases the extracellular space. During recovery and into the chronic stage, the loss of cell wall integrity causes the diffusion to increase.
Perfusion-Weighted Imaging
Two main approaches are taken to measure tissue perfusion. Both methods use EPI sequences because like diffusion, the effect can be easily degraded by minor patient motion. The first method for imaging tissue perfusion is to create an endogenous contrast. This is done by presaturating the blood flowing into the organ with an RF pulse. EPI of the organ is performed with the presaturation on and off. By subtracting the two images, a tissue perfusion image is created.35 The second method is by bolus injection of contrast material while repeatedly performing EPI. Time course images from EPI are then fitted to a model of the response of the tissue to the presence of the contrast agent, and an image of perfusion is made. This method is also known as a dynamic susceptibility contrast technique and requires a rapid intravenous bolus injection of gadolinium contrast material at a rate of 4 to 5 mL/sec, usually through an 18-gauge intravenous catheter, followed by normal saline injection. Manual or automatic postprocessing is required to generate color maps of various parameters of cerebral perfusion, such as CBV, CBF, mean time to enhance, and MTT.
Spectroscopy
The resonant frequency of a spin is determined not only by the B0 magnetic field and its associated imperfections but also by the molecule of which the nucleus is a part. The effect is slight, but not negligible. This phenomenon is known as chemical shift; that is, the chemical microenvironment of a given nucleus results in a slight change in the resonance frequency of the nucleus from that expected in its pure state. The difference between the resonant frequencies of the protons in water and the hydrogen atoms in the methylene groups in lipid are about 3.5 ppm. This translates to a difference of about 220 Hz for operation at 1.5 T. Between the water proton and fat proton resonances lie resonances of other moieties of interest in the brain. These include myoinositol, a sugar phosphate; choline, a key indicator of membrane turnover; creatine and phosphocreatine, part of the energy pool; glutamate and glutamine, the primary excitatory neurotransmitter and its astrocyte-recycled counterpart; N-acetyl aspartate (NAA), a key indicator of neuronal health; and lactate, an indicator of a shift to anaerobic metabolism. A spectrum of the resonances of the protons in these compounds is shown in Figure 17-22A. These resonances are included in every MR image of the brain. However, the concentration of each is lower than that of water by a factor of 5000 to 10000. To detect the resonances of the metabolites thus requires much larger voxels than imaging does: 2 cm3 versus 1 mm3 in imaging.
Two principal methods are used to collect spectroscopic data. The first, single-voxel spectroscopy, selects a voxel by using three successive RF pulses and then reads out the signal from the entire voxel. The second, chemical shift imaging, selects a larger volume or slice with one or more RF pulses. Phase-encoding gradients are then applied in one to three directions to yield 2D or 3D spectroscopy data sets. Reconstruction allows the selected region to be subdivided into separate voxels. An individual spectrum can then be extracted from a voxel, or a resonance may be selected and an image made of that resonance. Such 2D and 3D data sets can be postprocessed to generate color maps of metabolite distributions within the volume of brain evaluated with spectroscopy (Fig. 17-22B).
