Neuro MR: Principles - Queen's University · tem (CNS) and the harnessing of these physical principles to form clinical imaging protocols, tailored to suit a range of clinical presentations

Review Article

Neuro MR: PrinciplesTimothy P.L. Roberts, PhD1* and David Mikulis, MD2

The principles of contrast mechanisms and fast pulse se-quences underlying neurological and neuroradiological ap-plication of MRI are introduced in this part of a two-partreview. In particular, the relaxation time constants T1, T2,and T*2 are introduced, along with pulse sequences har-nessing their contrast, spin-echo, fast spin echo (FSE),inversion recovery (IR), gradient recalled echo (GRE), andecho planar imaging (EPI). The use of gadolinium (Gd)-based contrast is discussed in both T1- and T*2-weightedsequences. Tradeoffs between speed and image quality arediscussed, particularly in the context of long echo trainsequences (FSE imaging and EPI). The influence of parallelimaging strategies is also discussed. T*2 sensitivity is dis-cussed as both a source of artifact as well as a contrastmechanism (perfusion imaging with contrast agents, sus-ceptibility weighted imaging [SWI], and blood oxygenationlevel dependent [BOLD] imaging). Finally, the contrastmechanism of diffusion is introduced, as well as the con-cept of anisotropy. From these principles, the other part ofthis two-part review draws upon the pulse sequences andcontrast mechanisms to design disease and indication-specific protocols for state-of-the-art clinical use.

Key Words: pulse sequences; contrast mechanisms; neu-roimaging; physics; techniquesJ. Magn. Reson. Imaging 2007;26:823–837.© 2007 Wiley-Liss, Inc.

THIS REVIEW along with its partner review, “Neuro MR:Protocols,” form a combined platform serving to illus-trate the physical bases of image acquisition and con-trast manipulation for MRI of the central nervous sys-tem (CNS) and the harnessing of these physicalprinciples to form clinical imaging protocols, tailored tosuit a range of clinical presentations. As such, themanuscripts seek to address a broad readership, with astrong clinical focus, but recognizing the need for thor-ough physical grounding. In writing such reviews, it isinevitable that some facets cannot be completely ex-

plained and the interested reader is pointed to an arrayof texts that describe both physical and clinical aspectsof this domain in more detail (e.g., Refs. 1 and 2).

The various contrasts of MRI form an artist’s paletteof colors that can be drawn upon to compose the clin-ical neuroradiology protocol (Fig. 1). In this part of thereview, we introduce the component contrasts, examinetheir interpretation, and explore the properties and lim-its of their implementation. The completed “palette” ofcontrast possibilities will then be passed on to the clin-ical neuroradiologist to create disease-specific proto-cols drawing on the available contrast mechanisms (seeNeuro MR: Protocols).

In particular, we revisit T1, T2, and T*2 and focus onspecific neuroradiological issues associated with eachrelaxation time constant. We then proceed to examinethe contrast mechanism of diffusion-weighting. Otherarticles in this issue of JMRI address flow-related en-hancement and the basis and applications of magneticresonance angiography. Intravenous administration ofgadolinium (Gd)-based contrast media (“magnetic dye”)is introduced as a means of gaining additional physio-logically relevant image contrast. The associated ap-plied use of T1 and/or T*2 mechanisms additionally allowus to develop approaches to “perfusion-sensitive” imag-ing, imaging of microvascular permeability, and imag-ing sensitive to the oxygenation level of blood (bloodoxygenation level dependent [BOLD] imaging). Some in-troductory applications of these approaches are intro-duced to yield maps of white matter fiber paths (basedon diffusion-weighted imaging [DWI]), functional brainmapping (based on BOLD), and microvascular perme-ability (based on kinetic analysis of dynamic contrast-enhanced MRI). An introduction to magnetizationtransfer contrast (as it relates to T1- and T2-based re-laxation) is offered, as well as a suggestion regarding itsutility (particularly in the study of multiple sclerosisand in general for background tissue suppression inMR angiography). In general, an emphasis on the quan-titation of physiologically-relevant parameters (such ascerebral blood flow [CBF] and apparent diffusion coef-ficient [ADC]) derived from multiple image acquisitionsis stressed, since quantitation of tissue-specific param-eters provides a basis for intersite comparison as wellas for longitudinal studies of disease progression andresponse. Finally, recent advances in parallel imaging(with multiple receiver coils) are introduced, as thetechnology pertains to implementation issues of imag-ing with the above contrast mechanisms (3,4). Magnetic

1Department of Radiology, Children’s Hospital of Philadelphia, Univer-sity of Pennsylvania School of Medicine, Philadelphia, Pennsylvania,USA.2Department of Medical Imaging, Toronto Western Hospital, UniversityHealth Network, University of Toronto, Toronto, Ontario, Canada.*Address reprint requests to: T.R., Department of Radiology, Children’sHospital of Philadelphia, Wood Bldg., Suite 2115, 34th St. and CivicCenter Blvd., Philadelphia, PA 19104.E-mail: [email protected] February 11, 2006; Accepted May 3, 2007.DOI 10.1002/jmri.21029Published online in Wiley InterScience (www.interscience.wiley.com).

JOURNAL OF MAGNETIC RESONANCE IMAGING 26:823–837 (2007)

© 2007 Wiley-Liss, Inc. 823

resonance spectroscopy and spectroscopic imaging arebeyond the scope of this neuroimaging review. Further-more, attention is drawn to the increasing availability of3T MRI systems, and to the impact that higher fieldstrength has on the underlying physical principles dis-cussed (5).

RELAXATION TIME–BASED IMAGE CONTRAST

One of the main distinctive features of MRI compared toother cross-sectional imaging modalities (e.g., com-puted tomography [CT] and nuclear medicine tech-niques, single-photon emission computed tomography[SPECT], and positron emission tomography [PET]) isthe dependence of the MR signal on properties of thetissue, collectively known as spin relaxation. Charac-terized by three relaxation time constants, T1, T2, andT*2, tissue spin relaxation describes the process and rateat which the nuclear spin population (from which thenuclear magnetic resonance [NMR] signal originates)evolves in different tissues. While the details of nuclearspin relaxation are discussed in multiple NMR and MRItexts, the key feature to be retained in this discussion isthat T1 and T2, in particular, are properties of tissue,distinct, for example, between gray matter and whitematter, and, perhaps more importantly, often altered inthe context of pathological changes to tissue. Of greatinterest, also, is T*2. This relaxation time constant,which is related to T2, also reflects the tissue microen-vironment (particularly the inhomogeneity of the localmagnetic field). This may not only be a property oftissue, per se, but may be influenced by external factors(such as poor magnet construction and shimming), bythe nearby presence of tissues of different magneticcharacter, and, as is discussed below, by the presenceof endogenous or exogenous materials that transiently

disrupt magnetic field homogeneity by virtue of theircompartmentalized nature.

T2 Contrast and T2-Weighted Imaging

Considering first of all the so-called transverse, or spin-spin, relaxation time-constant, T2, we encounter aproperty of tremendous impact to the field. Not only doT2 values allow distinction between gray matter, whitematter, cerebrospinal fluid (CSF), scalp fat, and othertissues of the neuroimage, but T2 values of tissues arenotably altered in a number of pathophysiological situ-ations. Consider, simply, that transverse relaxation de-scribes the process by which the coherence of a popu-lation of excited mobile nuclear spins (the proton nucleiof the hydrogen atoms of water molecules) is lost due tothe transient experience by one spin of a magnetic fieldperturbation associated with proximity to another mo-bile excited spin. The rate at which this coherence islost characterizes the rate at which potential signal islost after initial excitation. Importantly, among otherfactors, the rate of coherence loss depends on the du-ration of such fleeting spin-spin interactions. That is, inhighly mobile situations, spin-spin interactions are sofleeting that magnetic field perturbations are less pro-nounced. Combined with a unifying “averaging” effectassociated with high mobility leading to more uniformdistribution of such perturbations across the spin pop-ulation, highly mobile fluid environments are charac-terized by relatively slower loss of transverse coherenceand are characterized by longer transverse relaxationtime constants, T2. More structured or constrained en-vironments (e.g., ordered cell assemblies and micro-structures) may be associated with more sustainedspin-spin interactions, more effective magnetic fieldperturbation, and more rapid loss of coherence acrossthe spin population (and thus shorter time constants,T2). Of clinical significance, many of the pathophysio-logical processes that occur in disease of the CNS areassociated physiologically with disruption of suchstructure in the microenvironment (and may be asso-ciated, for example, with cell lysis, mass fluid shift andpooling, or matrix breakdown). Consequently, suchprocesses are associated with prolongation of tissue T2

values compared with nonpathological states. As such,T2 changes can be seen to reflect pathological changesat a submicroscopic tissue level (inferences aboutchanges to the tissue microenvironment are made atdimension scales well below those of the “actual” im-ages acquired). This ability to make submicroscopicinferences about the integrity of the tissue microenvi-ronment is an inherent characteristic of MRI and amajor advantage of the technology. Since the parameterT2 reflects such important differences between tissuetypes and in pathophysiologic circumstances, a majoradvance is to “weight” MRI signal intensity, according tothe local T2 value—creating a “T2-weighted” image, sen-sitive not only to proton density, but also reflective oflocal tissue T2 properties. It should be borne in mindthat a “T2-weighted” image is not a pixel-by-pixel map ofT2-values, but rather is a proton density image in whichpixel intensities are additionally made sensitive to thelocal T2 value, with the degree of such sensitivity being

Figure 1. The art of MRI. Contrast mechanisms provide apalette of colors that protocols can be drawn upon to paint thedesired tissue scene.

824 Roberts and Mikulis

under the control of the MR operator, via the MRI pulsesequence parameters.

In the formation of the MR image, spin populationsare excited by application of an radio frequency (RF)pulse that is considered to rotate the magnetic mo-ments of the spins from an equilibrium (canonicallyz-axis) alignment into a transverse plane (in whichplane, the above-discussed loss of coherence, or relax-ation, occurs). A widely used MR pulse sequence ma-nipulates this transverse spin population through ap-plication of a 180° refocusing pulse to form the “spin-echo,” at a time, TE, after initial excitation. TE is knownas the “echo-time” and is a characteristic control pa-rameter of the MRI sequence. The signal recovered atthe echo time TE depends not only on the theoreticallyavailable signal (relating to the amount of water in thetissue being imaged), but also on the T2 value of thetissues, since the signal at time TE is attenuated by anexponential factor TE/T2. That is to say, for a given“proton density,” !, the signal collected at the time ofthe spin echo (TE) is given by equation 1:

S " !exp(#TE/T2) (1)

The signal S, then, indeed reflects both proton den-sity and local T2 values, with the degree of sensitivity toT2 (the “T2-weighting”) controlled by the parameter TE(imagine if TE $ 0, S is proportional only to !; on theother hand, for high TE values, S is strongly determinedby the ratio of TE/T2, thus introducing sensitivity to T2.Given the above described sensitivity of spin-echo se-quences to the parameter T2, and the fact that T2 ischaracteristically altered in a number of CNS patholo-gies, it is not surprising that spin-echo imaging hasbecome the mainstay of clinical neuro-MRI.

However, conventional spin-echo imaging is also as-sociated with relatively long scan acquisition times(typically several minutes). Ultimately, this limits spa-tial resolution, due to the inability of patients to avoidinvoluntary motion over the required long acquisitionperiods. Furthermore, in the case of noncompliant pa-tients, image quality with conventional spin-echo ac-quisition strategies may be limited, due to motion overthe acquisition time. This has led to the development ofalternative “T2-weighted” imaging sequences, which re-duce acquisition times by, for example, the collection ofmultiple echoes from a single excitation. Since manyechoes are required to form an image, the collection ofmore than one echo per excitation RF pulse has a cleareffect on reducing scan time (for a given spatial resolu-tion). These multiple spin-echo sequences are achievedby application of multiple 180° refocusing RF pulsesand are collectively known by a variety of acronyms,such as RARE (rapid acquisition with relaxation en-hancement), FSE (fast spin-echo) and TSE (turbo spin-echo). Figure 2 shows a schematic single-echo pulsesequence and corresponding schematic “multiple”spin-echo acquisition sequences, with echo trainlengths (ETLs) $ 2 and 4. The ETL directly relates to theacceleration of the multiple spin-echo sequence over itsconventional alternative.

From Fig. 2d it can be seen that each of the echoesformed in the multiecho train (e.g., Fig. 2c) fills a differ-ent line of k-space, with the vertical position of the linedetermined by the amount of phase encoding gradientapplied during formation of that particular echo.

Figure 3 compares a conventional spin-echo and FSEimage of the same patient, using an ETL of 16 for the

Figure 2. Pulse sequence timing diagrams for (a) spin-echoand FSE variants with ETLs of (b) 2 and (c) 4. (d) The schematicfilling of k-space in the four echo pulse sequence shown in (c).

Figure 3. A patient with a large cystic lesion acquired using (a)a spin-echo sequence with an acquisition time of seven min-utes and 17 seconds and (b) an FSE sequence with an ETL of16 and an acquisition time of 34 seconds. Note that imagecontrast and SNR appear comparable.

Neuro MR: Principles 825

FSE approach. Correspondingly, image acquisitiontime was reduced from seven minutes and 17 secondsin the conventional sequence to a mere 34 seconds inthe FSE variant. In general, both image (tissue) contrastand signal-to-noise ratio (SNR) are largely conserved.

While T2-weighting appears an advantageous utiliza-tion of FSE imaging, it should be remembered that theseparate phase-encoding of individual echoes in theecho train may influence image characteristics.Namely, the central portions of k-space (low phase en-code values) typically confer most of the image’s grossform, signal to noise and shape. Peripheral regions ofk-space carry information relating to sharp spatialtransitions in the image (small structures, details,edges, etc.). Thus, by applying different TEs to differentportions of k-space, we are not only influencing signalcontrast but also the relative weighting of gross vs. finestructures. Particularly considering a finite ETL, it isclear that if “early, short TE” (high SNR) echoes areapplied to the central regions of k-space, then outerregions of k-space will be represented by lower SNR(longer TE) echoes. Upon Fourier transformation, thismay be manifest as an underemphasis of image finestructure or detail, with resultant “blurry” appearance(see Fig. 4). Conversely, if later echoes (with longer TEs)are assigned to the central portions of k-space (in orderto achieve the desired T2-weighted contrast), residual(shorter TE) echoes must be utilized for the periphery ofk-space, leading to apparent overemphasis of the pe-ripheral portions of k-space (with the transformed ap-pearance of “edginess”).

With improvements in MR scanner hardware, partic-ularly gradient strength and receiver bandwidth, it hasbecome feasible to consider single-shot FSE imaging.Particularly in conjunction with partial k-space recon-struction algorithms (which rely on the Hermitian sym-metry, or conjugacy, of k-space), a high spatial resolu-tion image may be achieved from a single-shotacquisition (e.g., consider that a 256 % 256 matrixmight require only slightly more than 128 echoes to beformed; with an interecho spacing of 2–3 msec, thisresults in an overall acquisition time of somewhat lessthan 500 msec, within the limit imposed by tissue T2

values). The resultant image, obtained in a single shot,

must be considered in terms of intrinsic blur (as de-scribed above) vs. the possible (motion-related) artifactsincurred by selecting a similar contrast with a multi-shot (e.g., conventional FSE) approach. This may be ofparticular significance in imaging the noncooperativepatient, where conventional (multishot) sequenceswould be seriously degraded by patient motion (seeFig. 5).

FSE imaging has been a particular beneficiary of therecent advent of parallel imaging. Whether referred toas sensitivity encoding (SENSE), iPAT, or ASSET, theprinciples of multiple RF coil signal reception with sen-sitivity encoding allow fewer lines of k-space (fewerphase encoding steps) to be acquired, while still recon-structing an image with the desired spatial resolution.Although the details of parallel imaging are beyond thescope of this article (3,4) its implications for neuroradi-ology are presented in the context of influenced se-quences.

Figure 6a shows an axial slice through the brain of ahealthy volunteer acquired at 3T. Importantly for thescan protocol prescribed (TR $ 3500, TE $ 102 msec)with an ETL of 32, a maximum number of slices of 13was incurred (seriously limiting anatomic coverage).Furthermore, the incurred specific absorption rate(SAR) was 1.34 W/kg (close to the maximum tolerablelevels). Overall acquisition time was one minute andthree seconds. Figure 6b shows a similar slice from the

Figure 4. FSE images of a healthy volunteer with (a) protondensity weighting (TEeff $ 30 msec) and (b) T2-weighting(TEeff $ 105 msec). Contrast is manipulated by altering theposition in the echo train at which the central portion of k-space is acquired. There is a slight tendency toward blurring,especially in the proton density–weighted image.

Figure 5. FSE of a moving patient. a: Multishot FSE (ETL $16) is degraded by considerable patient movement, obscuringa brainstem lesion. b: Single-shot FSE (using partial k-spaceacquisition) allows an “instant” image to be acquired, freezingmotion.

Figure 6. Parallel imaging in FSE. a: An FSE image (ETL $ 32)acquired at 3T. b: Using parallel imaging acceleration canreduce the scan time by a factor of two. c: Parallel imaging canalso be used to reduce ETL to 16, thus reducing the SAR, oraccommodating more slices per TR, but not reducing overallacquisition time.


same acquisition, in which parallel imaging was incor-porated (acceleration factor %2) to reduce the numberof RF excitation needed to acquire sufficient data forimage reconstruction. Consequently (with the ETL re-tained at 32), the overall scan time was reduced by afactor of two (overall acquisition time 35 seconds), withno impact on SAR (per unit time) or the maximumnumber of tolerated slices. Conversely, Fig. 6c showsthe same axial slice in which the parallel imaging ben-efit has been applied to reduce the ETL from each RFexcitation. Consequently, the overall acquisition timeremains one minute and three seconds, but (with anETL of 16), the SAR is reduced to 0.71 W/kg. Alterna-tively, the maximum number of slices could be in-creased from 13 to 26. In common with all parallelimaging approaches, a penalty in SNR is inevitably in-curred. However, this strategy for reducing SAR ishighly important, particularly in the context of multiplespin-echo sequences and particularly in the climate ofincreasing B0 magnetic field strength from 1.5T to 3Tand above (since SAR values tend to increase as thesquare of magnetic field strength). Additionally, reduc-ing the ETL has an important image quality benefit: byreducing the signal intensity modulation across echoes(e.g., 16 different TE values rather than 32), images areless susceptible to blurring degradation.

A summary of T2-related pulse sequences and theirfeatures is offered in Table 1.

T1 Contrast and T1-Weighted Imaging

A second key contrast mechanism relates to the tissueproperty of longitudinal (or spin-lattice) relaxation. Inbrief, this process describes the recovery of magnetiza-tion after RF excitation back toward its equilibriumstate (longitudinal). This exponential recovery processis described by an exponential time constant, T1, andafter a 90° excitation, recovery of the longitudinal (Mz)component of magnetization over time, t, is given by Eq.[2], where ! again relates to the proton density.

Mz " !{1 # exp(#t/T1)}. (2)

Analogous to T2, tissues differ in their T1 values.Since longitudinal relaxation describes the process ofmagnetization recovery to equilibrium after excitation,it is apparent that the microenvironment greatly influ-

ences this process. Relaxation is facilitated by the pres-ence of relatively large moieties (such as macromole-cules and membranes), which can effectively interactwith the excited nuclear spins on passing water mole-cules. In the absence of such microstructural entities(e.g., in free fluids) longitudinal relaxation is slower.Thus, as with T2, T1 processes shed light on the micro-structural environment of water molecules on a dimen-sional scale that is not directly resolvable.

In an imaging sequence, if 90° RF excitation pulsesare applied at intervals, TR (repetition time), then thesignal, S, obtained is given by Eq. [3]:

S " !{1 # exp(#TR/T1)}. (3)

To the extent that S is less than its theoretical max-imum (due to attenuation by the term exp(–TR/T1),such images are described as “T1-weighted.” In general,the shorter the choice of TR value, the greater the de-gree of T1-weighting. Figure 7 shows a schematic T1-recovery curve for various tissues and indicates re-gimes that can be considered “T1-weighted” vs. “protondensity–weighted” (i.e., where T1 differences betweentissues do not exert much influence on signal charac-teristics—generally where TR is considerably greaterthan all T1s present).

It can be seen that choosing a TR value on the orderof 500 msec (Fig. 7; arrow) allows clear separation of thethree illustrated tissue types, with CSF showing only&20% of its theoretical maximum (fully relaxed) signaland therefore appearing relatively hypointense, withtissue exhibiting approximately 50% of its fully relaxedpossibility (mid-intensity) and with fat having under-gone considerable longitudinal relaxation (since fat isassociated with a short T1 time) and contributing morethan 90% of its fully relaxed maximum (and thus ap-pearing hyperintense). It is worth noting that increasingthe TR will tend to increase signal from all moieties butthat the benefit will be small for the fat tissue for exam-ple, in which there is only 10% more signal “available.”Another key point is that tissue contrast as described inthis example is achieved by selectively discarding SNR.That is, we achieve contrast between fat, brain tissue,and CSF by only allowing tissues to contribute a per-centage of their theoretical maximum. Thus, we havetraded signal for contrast. Hence, it is common in quan-

Table 1Summary of T2-Weighted Sequences

Sequence Common acronyms Feature Key variables

Conventional spin-echo SE, CSE Proton density to T2-weighted contrast. Longacquisition times

TE

Fast spin-echo FSE, TSE, RARE Typically 8-16 times faster than CSE. Can lead to blur ETLSingle-shot fast spin-echo ssFSE, HASTE Uses long ETL and partial k-space to obtain all k-space

data in a single shot. Can be blurry. High SAR.Overall acquisition time

(relates to interechospacing and ETL)

Fluid-attenuated spin-echo FLAIR Uses IR prepulse and long TI to “null” signal from CSF TI (to exactly null CSF)Parallel imaging SENSE, ASSET, iPAT Can accelerate all above sequences by factors of

typically 2-4, or reduce ETL by similar factor inmultiecho sequences

Acceleration factor

IR $ inversion-recovery.


tifying MRI not only to describe an SNR, but also todescribe a contrast to noise ratio (CNR) between twotissue types. It is the soft tissue contrast achievablewith MRI that is largely responsible for the wide adop-tion of this imaging modality.

To increase the dynamic range of T1-weighting evenfurther, it is possible to apply not a 90° RF pulse butrather a 180° RF pulse, to “fully invert” the longitudinalmagnetization. The Mz component then proceeds to re-covery by longitudinal relaxation, as described above,but now starts from a value of –1, rather than 0 asbefore. A schematic inversion recovery curve is shownin Fig. 8.

Since longitudinal magnetization (Mz) is initially neg-ative and recovers to its eventual positive equilibriumvalue, it must at some point pass through an instanta-neous value of zero. This is known as the “null point”and occurs at a time, t $ 0.69 T1, for each species. If animaging sequence (initiated by a 90° RF excitation) iscommenced a time t after a 180° inversion pulse, thesignal intensity of tissues in the image will be scaled bythe instantaneous value of Mz at time t. If t is chosen tobe the null point for a given tissue, then this tissue’ssignal intensity will be scaled by zero in the subsequentimage and it will be “suppressed.” This powerful tech-nique has several indications, but two variants arewidely encountered: short tau inversion recovery(STIR), in which fat signal is suppressed by choice of ashort inversion recovery time t; and fluid attenuatedinversion recovery (FLAIR), in which the inversion re-covery time t is chosen to “null” the signal of the long T1

species from the CSF (see arrow in Fig. 8) (6). FLAIRimages (by virtue of allowing most species to fully re-cover) offer a great deal of clinical utility as they retain

the diagnostic qualities of a T2-weighted spin echo im-age (sensitive to edema, inflammation, etc.), as de-scribed above, but with the signal from the CSF elimi-nated, improving the delineation of a lesion that mightotherwise be approximately isointense with CSF on aconventional T2-weighted image and for which it wouldthus be difficult to determine the margins (see Fig. 9).As can be seen, the FLAIR sequence draws on both T1

and T2 contrast mechanisms to improve diagnostic util-ity.

Gd-Based Contrast Agents for Enhancement of T1-Weighted Images

Despite the power and utility of endogenous contrastmechanisms exploiting tissue-specific differences in T1

and T2 values, there are situations in which the diag-

Figure 7. Schematic T1 recovery curves for fat, tissue, andCSF as a function of increasing TR (approximate T1 values of200 msec, 700 msec, and 2000 msec, respectively, are simu-lated). The lightly shaded area between 0 and 1000 msec canbe considered to provide T1-weighting (with marked differencesin Mz values between tissue types); a typical TR choice of 500msec is marked with a vertical arrow; the longer TR regime(illustrated as TR $ 3000–4000 msec) can be seen to be littleinfluenced by T1 and thus images acquired with such TRs willexhibit contrast that is largely dependent on proton density.(Note that even at TR $ 4000 msec, some species such as CSFare still not fully relaxed and thus will be partially suppressedin intensity).

Figure 8. Schematic T1 inversion recovery curves for fat, tis-sue, and CSF as a function of increasing TR. Note that Mz isinitially negative, but recovers over time to its positive equilib-rium value. At a characteristic time (marked by arrow for CSFcurve), Mz must pass through zero. This is termed the “nullpoint”.

Figure 9. FLAIR. By suppressing otherwise hyperintense sig-nal from CSF, this T2-weighted image allows delineation ofpathologies such as brain tumor (arrow).


nostic utility of MRI can be further improved by theadministration of an exogenous contrast medium. Acontrast medium can be considered as a “magnetic dye”that is administered by intravenous injection and canbe visualized as it distributes. All presently available(U.S. Food and Drug Administration [FDA]-approved)contrast agents are based on low molecular weight(&500 Da) organic molecules that chelate or encase anion of the rare earth metal, Gd. The Gd3' ion, by virtueof its electron configuration, is a highly potent “en-hancer” of longitudinal relaxation. That is to say, wher-ever the nuclear spins of water molecules encounterand interact with the Gd3' ion, their T1 value is sub-stantially shortened. Such T1-shortening can be re-vealed on T1-weighted images, as described above, andis commonly observed as a relative hyperintensity. Allapproved contrast agents can also be considered asnonspecific markers of the extracellular fluid (ECF).Sometimes they are referred to as “ECF agents.” Theyfreely distribute in the extravascular, extracellularspace of the body, but do not enter healthy cells. Im-portantly for neuroradiology, they do not cross an intactblood brain barrier. Thus administered intravascularly,they remain intravascular on passage through thehealthy brain. Given the low fractional blood volume ofbrain tissue, it is relatively difficult to discern the pres-ence of the contrast agent in tissue parenchyma (signalin larger vessels, however, may well be enhanced). If,however, the blood brain barrier integrity has beencompromised (as is the case in a large number of brainpathologies), contrast agent will extravasate out of theintravascular compartment and into the tissue space.The contrast agent concentration may thus accumulatelocally and have a substantial T1-shortening effectwithin that region, leading to conspicuous hyperinten-sity on T1-weighted images (see Fig. 10).

In fact, not only the presence, but also the degree ofblood brain disruption can be interrogated with MRI.Use of sequential T1-weighted images allows the pro-gression of tissue enhancement to be characterized.Upon appropriate kinetic modeling of the dynamic en-hancement data, it is possible to estimate both thevascularity (fractional blood volume) and the microvas-cular permeability of tissues (7), revealing subtle butphysiologically significant hyperpermeability of theblood brain barrier (Fig. 11).

In fact, some studies suggest that the microvascularpermeability of tumors may reflect the degree of angio-genic activity within the tumor (Fig. 11b). In principle,therefore, monitoring tumor microvascular permeabil-ity might offer a useful strategy in the assessment ofantiangiogenic therapies. Recent data also suggeststhat evidence of hyperpermeability of brain microves-sels in regions of acute cerebral ischemia may be pre-dictive of subsequent hemorrhagic transformation. Ta-ble 2 summarizes considerations for T1-weightedsequences.

T*2-Weighted Image Contrast

While the use of Gd-based contrast agents in conjunc-tion with T1-weighted imaging as described above tendsto lead to local hyperintensity, or “positive enhance-ment,” Gd-based contrast agents are also widely usedin another guise as “negative enhancing agents.” Thisexploits the strong paramagnetic nature of the Gd3' ionand a final relaxation mechanism, described by thetime constant T*2. Related to T2, as it describes the rateat which transverse coherence is lost after initial RFexcitation, T*2 embodies all mechanisms that lead tosuch dephasing, not only the interactions between ex-cited nuclei, but also spatially-dependent variations inmagnetic field that lead to different spin precessionrates, and thus to dephasing and signal loss. Such

Figure 11. a: Schematic of microvascular permeability deri-vation as the slope of progressive enhancement in tissue. Byplotting the ratio of tissue enhancement to reference vascularenhancement, progressive extravasation of contrast agent intotissue can be resolved, even in the presence of ongoing plasmaclearance. In practice, the time axis is often “stretched” toreflect the decreasing plasma concentration of tracer (and apseudotime, or “Patlak” time axis is generated to maintain thelinearity of progressive enhancement implied by the sche-matic. b: Permeability (Kps) map of patient in Fig. 10. Thispermeability map was generated by pixel-wise kinetic model-ing of the dynamic enhancement data as described in Ref. 7and shown schematically in (a).

Figure 10. T1-weighted image showing hyperintensitypost-Gd administration revealing blood brain barrier disrup-tion in this patient with a high-grade glioma.


variations may arise from limits to the quality of magnetconstruction or to inherent differences in the magneticcharacter of tissues—different tissues exhibit differingdegrees of dia-/paramagnetism (i.e., their magnetic re-sponse to being placed in an external magnetic field).The degree to which a tissue (or indeed any entity)responds magnetically to being placed in a magneticfield is described by its “magnetic susceptibility,” (.Differences in magnetic susceptibility between soft tis-sues and bone, or air, thus lead to strong local variationin magnetic field at the interface between them. Thisfield variation leads to rapid spin dephasing, describedby a short T*2 value, and signal loss. Resultant signalvoiding is often termed “magnetic susceptibility arti-fact” (see Fig. 12).

While some magnetic field inhomogeneity can beimproved by the process of shimming (application ofexternal magnetic field gradients to attempt to com-pensate for observed field variation), magnetic sus-ceptibility effects are largely corrected by the forma-tion of the spin echo (as in T2- and T1-weightedimages). They remain, however, potentially signifi-cant in gradient recalled echo sequences and echoplanar imaging (EPI) (see below). Gradient recalledecho techniques (with short TE values) are nonethe-less typically preferred over spin echo sequences for

the rapid acquisition of volumetric data (three-di-mensional [3D] sequences) as well as MR angiographyand MR venography. EPI, due to its tremendousspeed of acquisition is widely used to study dynamic(or functional) changes, or to “freeze” involuntary pa-tient motion. Since magnetic susceptibility–related“dephasing” necessarily evolves over the time betweenexcitation and echo formation (the echo time, TE), it isclear that effects can be minimized by selection of shortTE values. Practically, TE can generally be shortenedby: 1) use of higher receive bandwidths (increasing therate of data acquisition and thus decreasing the overallduration of each echo); or 2) using partial k-space tech-niques to only partially sample the echo.

However, rather than seeking to ameliorate or reducethe sensitivity of an MR sequence to magnetic fieldinhomogeneity, there are notable circumstances inwhich such sensitivity can be positively exploited. In-deed, this contrast mechanism forms the basis of threeexciting emerging MRI applications: perfusion imaging,susceptibility-weighted imaging, and functional brainmapping.

Gd-Based Contrast Agents as Negative EnhancingAgents

As discussed above clinically-available contrast media(e.g., Gd-diethylenetriamine pentaacetic acid [DTPA]),injected intravenously, remain intravascular duringpassage through the brain, with extravasation (leak)prohibited by the intact blood brain barrier. With T1-weighted imaging, the low blood volume fraction ofbrain tissue makes enhancement subtle and generallynot resolvable. However, since the magnetic field–dis-turbing effect of the paramagnetic Gd3' moiety extendswell beyond the physical location of the Gd3' ion itself,spin dephasing and signal loss occurs not just in theimmediate vicinity of the Gd3' ion (i.e., intravascularly)but also within the extravascular space of parenchymaltissue. As such “negative enhancement” can be consid-ered to be occurring in both intra- and extravascularcompartments, considerably amplifying sensitivity of

Table 2Summary of T1-Weighted Sequences


Conventional spin-echo

SE, CSE Proton density to T1-weighted contrast.Weak contrast at 3T and above.

TR

IR-prepped 3Dgradient-recalledecho

MP-RAGE, FSPGR, FFE 3D sequence with IR-prep to achievedesired T1 contrast

TI

TOF-MRA 2D or 3D gradient recalledecho

Uses inflow enhancement of blood(and suppression of stationarytissue) to yield angiographic images.Multislice 2D or 3D. Usually 3Dpostprocessed (MIP)

Flip angle, slabthickness (toameliorate taperingsaturation of inflow)

' Gd SE, 3D GRE Uses T1-shortening effect of i.v.-administered GdDTPA to probe BBBdisruption

TR (to ensure T1-weighting)

Permeabilityimaging

Gd-enhanced dynamic 3DGRE

Can image progressive enhancementof tissue and model this to estimatemicrovascular permeability

Flip angle, temporalresolution

IR $ inversion-recovery, TOF-MRA $ time-of-flight MR angiography, BBB, blood brain barrier.

Figure 12. a: Magnetic susceptibility schematic and (b) arti-fact (solid arrows) at tissue:air interface. Note hypointensity incortical veins (dotted arrows) resulting from endogenous,paramagnetic deoxyhemoglobin.


detection. In fact, dynamic T*2-weighted imaging (gradi-ent-recalled echo [GRE] or, more typically, EPI) is widelyused to serially image brain tissue, before during andafter injection of a short bolus of Gd-containing con-trast medium, and effectively to “track” its passage into,through, and out of the brain (Fig. 13) (8,9). The time-varying signal intensity, S(t), of a given pixel can bedescribed by Eq. [4].

S(t) " ! exp(#TE/T*2), (4)

where TE controls the sensitivity of the sequence to theparameter T*2. Since T*2 is shortened by the presence ofthe paramagnetic contrast medium, according to Eq.[5]:

1/T*2 (t) $ 1/T*2(0) ' r*2[Gd], (5)

(where r*2 is a constant of proportionality, known as theeffective transverse relaxivity), the time-varying signalintensity S(t) can be related to [Gd] through the surro-gate, )R*2, where R*2 $ 1/T*2. Thus,

[Gd](t) " )R*2(t) $ (1/TE) ln(S(t)/S(0)). (6)

So the signal intensity curve is transformed into atime-varying function, )R*2, which is taken proportionalto the instantaneous tracer concentration (Fig. 14).From this time-varying estimate of tracer concentra-tion, standard tracer kinetic modeling approaches canbe employed to derive physiologically-specific parame-ters of the vascular system (10,11).

Analysis of such “bolus-tracking” experiments canyield relative, and increasingly quantitative, absolutemeasures of cerebral hemodynamics, such as cerebralblood volume (CBV), mean transit time (MTT), and CBF.So-called “MR perfusion” imaging is widely used in thestudy of acute cerebral ischemia, and is also encoun-tered in brain tumor imaging (however, the disruptionof the blood brain barrier in brain tumors leads tocontrast agent extravasation and considerable compli-cation in the kinetic analysis and interpretation of such

data). In fact, there are alternative approaches to MRassessment of CBF and perfusion that are emerging asadvantageous in some circumstances. Particularly, themethodology of arterial spin labeling (ASL), originallyintroduced by Detre et al (12) offers a quantitative ap-proach to CBF measurement entirely without the use ofexogenous contrast agents. A full discussion of thisapproach is beyond the scope of this article, especiallysince this technique is not yet widely adopted in clinicalprotocols. Nonetheless, a brief description is war-ranted. Rather than dynamically imaging (or “tracking”)a bolus of contrast agent as it passes through the brain,inflowing arterial blood is “magnetically labeled,” typi-cal by applying a spatially-selective 180° inversion RFpulse, for example at the level of the carotid artery.Such labeled blood then travels toward the more supe-rior slice or slices of interest in the brain, carrying withit a population of spins with inverted magnetization(i.e., a z-component of magnetization, Mz, with a valueof –1). Since stationary (unlabeled) tissue spins have anMz of '1, the signal intensity in a subsequent image willbe weighted by the fraction of inflowing labeled spins(effectively decreasing the net or ensemble Mz value). Acomparison image is acquired without such labeling (inwhich the net magnetization Mz remains 1). The differ-ence between the two images (obtained by pixel-wisesubtraction) is then entirely attributable to the frac-tional content of inflowing/perfusing water molecules.Thus the magnetically-labeled water can be consideredas a freely diffusible tracer, allowing subsequent quan-titative modeling of perfusion parameters. A substan-tial limitation of ASL is the requirement to wait between“labeling” and “imaging” for the blood to arrive and forthe labeled water to exchange into tissue. The impact ofthis waiting period arises from the T1-based recovery ofthe inverted “labeled” spin population, effectively caus-ing the “label” to fade over time (according to the T1

value of blood). This effect appears to motivate the con-tinuing development of ASL in the context of higher field

Figure 14. a: Dynamic signal intensity time course from rap-idly acquired sequential T*2-weighted images reveals a tran-sient signal loss associated with the presence of the contrastmedium bolus. This rapidly recovers toward the baseline. b:Transforming the signal intensity time course into the time-varying measure of )R*2, which is taken as proportional to [Gd],we obtain a concentration–time curve for the tracer. This isamenable to simple parametric description (e.g., time to peak,magnitude of peak, area under the curve, peak full width athalf maximum [FWHM]) as well as more sophisticated kineticanalysis, to yield parameters such as MTT and CBF.

Figure 13. A series of six T*2-weighted echo planar imagesacquired during passage of a bolus of 0.1 mmol/kg Gd-DTPA.Temporal resolution of the sequential images is two seconds.The negative enhancing property of the Gd-DTPA may be mostclearly visualized in (c) at the moment the local concentrationof Gd-DTPA peaks. Absence of contrast media in the ventriclesis seen as preserved signal.


MRI systems (3T and above, see below), since a phe-nomenon of higher field strength is the prolongation ofT1 values, which in this setting will lead to stronger andlonger persistence of the magnetic label.

In fact, the signal loss associated with the transientpresence of paramagnetic Gd3' would in principle oc-cur with any paramagnetic species. At high magneticfield strength, this was originally reported by Ogawa etal (13) as causing the marked hypointensity in cerebralveins (see Fig. 12b). It was hypothesized that the in-creased deoxyhemoglobin content of veins compared toarteries would lead to increased magnetic field disrup-tion and signal loss, due to the paramagnetic nature ofdeoxyhemoglobin (while oxyhemoglobin is diamagneticand has little negative impact on magnetic field homo-geneity). For visualizing cerebral veins, this phenome-non has been enhanced in the work of Reichenbach etal (14) and Haacke et al (15), whereby the magnitude ofsignal is masked by the phase image (since phase isalso disturbed by the deoxyhemoglobin). The resultantpostprocessed images are typically “MinIP’d,” using analgorithm similarly to maximum intensity projection(common in MR angiography to identify the brighteststructures in a 3D data set), where the MinIP algorithmidentifies and forms a projection image of the darkestvoxels within a 3D dataset. Figure 15 illustrates theresultant, so-called “susceptibility weighted image”(SWI) from a healthy volunteer, acquired at 3T.

BOLD Contrast—Functional MRI

However, the major application resulting from the dif-fering influences on local image signal intensity of de-

oxy- vs. oxyhemoglobin is the field of functional mag-netic resonance imaging (fMRI), which exploits the“BOLD contrast” and a remarkable physiological phe-nomenon. Secondary to neural activity associated withperipheral stimulation, task performance or cognitiveactivity, there is a transient increase in regional CBF (inresponse to the increased demands of the active neu-rons). This increased CBF is not, however, balancedwith a commensurately increased demand for tissueoxygen, O2. Consequently, there is a net increase ofoxyhemoglobin in the capillary bed, compared to theperiod prior to the neuronal activity (Fig. 16). An in-crease in oxy-hemoglobin is associated with a relativedecrease in deoxyhemoglobin. Since deoxyhemoglobin,being paramagnetic, is associated with signal loss onT*2-weighted images, its transient decrease in concen-tration is visualizable as a transient increase in localsignal. That is, in the capillary bed of tissue closelycoupled to the neurons that were active, there will be anincrease in signal intensity. Thus the field of humanbrain mapping with fMRI was initiated (16).

Table 3 summarizes considerations and applicationsof T*2-weighted MRI.

EPI

To achieve the above-desired magnetic susceptibility–weighted contrast, and at the same time to haveadequate temporal resolution to monitor the transientdynamic changes in oxy-/deoxyhemoglobin concentra-tion with whole brain coverage requires a high perfor-mance imaging sequence. Most widely used is the GRE

Figure 16. fMRI. a: Schematic of typical “block design” para-digm in which periods of task performance (black) and rest(open) are alternated. b: The theoretical basis of BOLD fMRI isthe replacement of paramagnetic deoxyhemoglobin (filled cir-cles) with diamagnetic oxyhemoglobin (open circles) duringregional neuronal activation associated with the task. c: Con-sequent fluctuations in signal intensity on T*2-weighted images(signal increasing locally during task performance) also reflectthe “hemodynamic lag” or delay in the BOLD response withrespect to neuronal electrical activity. d: Identification of pixelswith signal intensity time courses significantly correlated withthe model in (c) allows construction of statistical parametermaps effectively identifying functional centers. e: Such mapscan be reconstructed in 3D, for example for use in surgicalplanning.

Figure 15. Susceptibility-weighted image. Minimum intensityprojection through a set of axial T*2-weighted images of ahealthy adult volunteer. Images were acquired at 3T and post-processed to weight the resulting image not only by the signalmagnitude, but also according to its phase.


EPI sequence (17). In much the same way as the FSEand TSE variants of spin-echo imaging collect multiplespin echoes (i.e., lines of k-space/phase encodingsteps) per RF excitation, EPI collects multiple gradientrecalled echoes. Unlike the FSE approach, which ap-plies multiple 180° RF refocusing pulses, the EPI ap-proach applies multiple reversals of the readout gradi-ent polarity (see Fig. 17).

Most implementations of EPI are as (relatively) low-resolution single-shot acquisitions (with a matrix size of64 % 64 or 128 % 128). Indeed, if one considers that thelimitation to an EPI echo train is ultimately governed bythe tissue T*2 (as in Fig. 17), increasing receiver band-width is also a major factor in reducing the amount oftime required to collect each echo, ultimately shorten-ing the total acquisition time. Importantly, in addition,

the recent advent of parallel imaging makes it indeedfeasible to reconstruct single-shot EPI images with amatrix size of 256 % 256 or more. The use of high-bandwidth receivers allows the acquisition of 128 ech-oes faster (and therefore with less signal loss and blur);the use of parallel imaging allows such echoes to bereconstructed into a 256 % 256 matrix. Thus in combi-nation (see Fig. 18), it is now possible to acquire higherspatial resolution single-shot EPI images with lowersignal degradation (compare images Fig. 18a and c).

DWI

Exploiting the rapid image acquisition of single-shotEPI, visualization of an additional tissue contrastmechanism is offered. Images can be made sensitive to,or weighted by, the process of water molecule diffusion.By addition of a pair of magnetic field gradient pulsesprior to the image readout module of the EPI sequenceit is possible to probe the movement of water molecules,by “labeling” the proton position during the first gradi-ent pulse and “unlabeling” it during the second pulse. Ifthe proton position does not change between gradientpulses (typically &40 msec apart), then signal will beretained in the subsequent echo. To the extent that the

Table 3Summary of T*2-Weighted Sequences


Gradient recalledecho

SPGR, GRE, FLASH, FFE Sensitive to T*2 shortening associated with iron(e.g., hemorrhage). Sensitive to magneticsusceptibility artifact at tissue interfaces.

TE

Echo planar imaging EPI Essentially a single-shot GRE. Highlysensitive to magnetic susceptibilitydifferences. Tendency toward anatomicdistortion and blur. Historically, lowresolution. Demanding of hardwareperformance.

TE, ETL, overallacquisition time

Perfusion EPI Uses dynamic imaging with T*2-sensitivity totrack Gd-bolus passing through brain due to“negative enhancing” effect of paramagneticGd.

Temporal resolution.Choice of GRE(sensitivity) vs. spin-echo (increasedcapillary specificity)

BOLD EPI Uses T*2-sensitivity to track endogenouschanges in deoxy- and oxyhemoglobinconcentration (secondary to neuronalactivity, or breathhold, or O2 administration)

TE (for sensitivity), TR(for temporal resolution)

Figure 17. EPI. Successive echoes (and thus lines of k-space)are acquired by reversal of polarity of the readout gradient.Each echo is associated with a different history integral ofphase encoding (effectively shifting the vertical coordinate ink-space). Echo amplitude decreases according to T*2, sincethere is no spin echo refocusing.

Figure 18. EPI images acquired with: (a) 128 % 128 matrixand no parallel imaging, TE $ 69 msec; (b) 128 % 128 matrixwith a parallel imaging acceleration of %2 shortening the ETLto 64 and reducing the TE to 60 msec; and (c) 256 % 256matrix with a parallel imaging acceleration of %2 (i.e., an ETLof 128), resulting in a TE of 65 msec.


proton’s position has altered over this time, the “unla-beling” step will be incomplete and overall signal willdrop. The greater the displacement, the greater the sig-nal attenuation (Fig. 19). Again, this contrast mecha-nism, exposed over a spatial resolution of the order of 1mm, allows inference about water molecule movementon a spatial scale of 5–10 *m (since the mean free pathof diffusion over &40 msec has a magnitude measuredin *m).

The most widespread application of DWI is in theidentification of acute ischemic stroke (wherein waterdiffusion in the ischemic region is restricted, perhapssecondary to cytotoxic edema or reduced cytosolicstreaming) (18,19). This early manifestation of cellmembrane pumping failure (most of these mechanismsrequire energy consumption and adenosine triphos-phate [ATP] dephosphorylation) becomes apparentearly after the ischemic event (minutes in animal mod-els and within approximately one hour in clinical set-tings). Figure 20 shows an example of a DWI imageacquired in an acute stroke patient.

It is clear from Eq. [7] that the signal intensity onsuch a DWI image in fact depends additionally on pro-ton density and also on T2 (since diffusion weightingwas effectively created by augmenting a T2-weightedpulse sequence with diffusion encoding gradientpulses).

S " ! exp(TE/T2).exp(#b.ADC). (7)

These additional weightings may confound interpre-tation of a DWI image in isolation—it may be difficult todistinguish between hyperintensity on a DWI imagethat arises from reduced ADC vs. prolonged T2, for ex-ample (20). Fortunately, the ability to quantify param-eters such as the ADC can resolve this ambiguity. In-stead of obtaining a single DWI image with a certain

b-value, one is required to acquire two images withdifferent b-values (typically, one image has a b-valueclose to 0 seconds/mm2). From such a pair of imageswith b-values of for example 0 and 1000 seconds/mm2,it is possible to eliminate confounding weighting (fromproton density, and indeed T2) and derive on a pixel-by-pixel basis the value of the ADC (Eq. [8]), which can berepresented in grayscale as the ADC “map”:

ADC $ (#1/b) ln(Sb/S0), (8)

where Sb is the signal intensity in a given pixel for theimage acquired with the high b-value (e.g., 1000 sec-onds/mm2) and S0 is the signal intensity in the corre-sponding pixel of the image acquired without diffusionweighting (b $ 0) (Fig. 21), where b is the sensitivity ofthe diffusion weighted sequence to the process of diffu-sion.

The b-value is dependent on several imaging se-quence choices, namely the duration, separation, andamplitude of the diffusion encoding gradient pulses.For maximum sensitivity to changes in diffusion coeffi-cient, while retaining adequate SNR, a choice of b $1000 seconds/mm2 is widely used (note typical tissueADC values are approximately the reciprocal of this,such that the product b.ADC + 1).

Anisotropy and Diffusion Tensor Imaging (DTI)

Since diffusion encoding gradient pulse pairs can beapplied on any (or all) of the gradient axes, the resultantimage can be made sensitive to diffusion processes indifferent directions. This is of particular utility in thecharacterization of anisotropic diffusion (i.e. diffusionin which one direction is preferred, e.g., within whitematter tracts; Fig. 22a) (21). Acquiring DWI images withdiffusion-sensitization in different directions allowsboth the magnitude and orientation of anisotropy to bedetermined. To fully describe the degree and directionof anisotropy, diffusion is often considered as a tensorquantity (a 3 % 3 matrix) rather than as a single scalar(22). DTI is thus the term applied to successive DWIacquisitions, with encoding in different directions toallow determination of the diffusion tensor. Note atleast six different directions of diffusion encoding arerequired in order to solve the diffusion tensor. Figure22b illustrates a typical representation of the degree of

Figure 20. DWI image in acute stroke. Reduced diffusion inthe region of ischemia (arrow) is visualized as hyperintensityon the DWI image (b-value $ 1000 seconds/mm2).

Figure 19. Schematic of DWI imaging. a: Diffusion sensitiza-tion is achieved by applying a pair of magnetic field gradientpulses to a spine echo (or spin echo EPI) sequence. The degreeof sensitization (called the b-value) depends on choice of gra-dient amplitude (G), duration (,), and separation ()). b: Sche-matic of spin dephasing and incomplete rephasing secondaryto diffusion (or displacement).


diffusion anisotropy using one metric, the fractionalanisotropy (FA).

Taking such pixel-by-pixel information on diffusionanisotropy, it is possible to track the pathway of whitematter tracts through the brain. One algorithm, fiberassignment by continuous tracking (FACT) considers

the direction of anisotropy (the principal eigenvector ofthe diffusion tensor) and proceeds from an initial seedpoint in the direction of the principal eigenvector frompixel to pixel (updating the direction at each new pixel)(23). This continues until a predetermined lowerthreshold of FA is encountered, at which point the fiberpath is terminated (as the pixel no longer is consideredto be dominated by white matter). Figure 22c showsexamples of white matter fiber tracts, with differentcolors representing different initial seed points. It canbe seen that this method offers a powerful approach tothe study of brain connectivity and is expected to pro-vide synergistic insights when used in combinationwith functional center identification using fMRI.

Magnetization Transfer Contrast

T1, T2, and indeed ADCs provide MR-accessible param-eters of tissue that in various ways reflect the physico-chemical nature of the microenvironment. The insightsthat can be drawn from resultant image contrast can beseen as reflecting the power and scope of MRI andaccounting for its exquisite soft tissue contrast. An ad-ditional level of insight into the tissue microenviron-ment can be obtained using a mechanism known asmagnetization transfer contrast. In this approach, themicroscopic composition of tissue is probed, specifi-cally the content and accessibility of macromolecularand microstructural components such as myelin. Inbrief, the physical phenomenon exploited is the ongoingexchange between protons of appropriate macromolec-ular entities and protons of free water. Effectively, two“pools” of protons can be considered—the free pool andthose protons “bound” by a chemical bond to the mac-romolecular entities (typically as part of a hydroxyl&OH) group. If considered separately, the T2 values ofthe two pools would be found to be very different (withthe relatively immobile bound protons having a veryshort T2 value). Consequently, imaging of a pixel con-taining both pools reveals only the free pool content(since signal is lost very rapidly from the short T2 spe-cies). Essentially, the bound pool protons are invisible.Another feature of the short T2 species is that a chem-ical spectrum of this population would reveal a verybroad resonance (extending over several kHz) is contra-distinction to the sharp resonance of the free water

Figure 21. a: A DWI image reveals focal signal hyperintensity,in this case of active multiple sclerosis. However, inspection ofthe baseline (b $ 0), T2-weighted image reveals hyperintensityattributable to T2 prolongation. It is not clear whether thehyperintensity on the DWI image reflects reduced diffusion orprolonged T2. By constructing a ratio image (c), formed bydividing the DWI image by the T2-weighted image, we are leftwith an image, weighted by exp(–b.ADC), which reveals hy-pointensity (consistent with elevated diffusion). Synthesizing apixel-by-pixel map of the ADC itself from the natural logarithmof (c) confirms the slightly elevated ADC value in this zone.This phenomenon (termed T2-shinethrough), leading to ambig-uous interpretation of the DWI image in isolation, is resolvedby the subsequent image analysis (ratio formation and/orADC mapping).

Figure 22. a: Anisotropy of diffusion schematic. b: The degree of anisotropy is calculated on a pixel-by-pixel basis and can berepresented as a map of metrics such as FA. c: Pixel-wise computation of the direction (as well as degree) of anisotropy allowsstepwise reconstruction of fiber tract paths (typically by advancing from a seed point in single pixel steps), the direction of whichis determined by the principal diffusion direction of each pixel.


protons. The peak of both resonances might be coinci-dent, but the linewidths differ considerably. If a strongpulse of RF radiation is applied at a frequency shiftedfrom the spectral peak of the two pools in such a waythat it impacts the broad line of the bound pool but isshifted beyond the narrow line of the free pool (so-called“off-resonance” excitation), one might expect to exciteand ultimately saturate the magnetization of the boundpool protons, with minimal impact on the protons of thefree pool. Initially, one might expect this process to haveno effect on subsequent image signal intensity (sincethe bound pool protons were “invisible” anyway, be-cause of their short T2 values). However, under appro-priate chemical and microstructural circumstances, assuggested above, protons in the two pools exchange.For the purpose of signal intensity this means thatsome protons of the bound pool carry their saturatedmagnetization with them as they join the free pool,while corresponding protons of the free pool join theinvisible group of bound protons (as they experience theshort T2 environment). The availability of unsaturatedmagnetization in the free pool is therefore decreasedand, consequently, signal intensity decreases (Fig. 23).The degree of reduction depends on the amount andavailability of protons in the bound pool and the effec-tiveness of the proton exchange. It can be quantified byacquiring two images, one with and one without theextra (off-resonance) RF irradiation. The loss of signalassociated with magnetization transfer is describedmathematically in the magnetization transfer ratio(MTR) given by Eq. [9]

MTR $ (Mo # Ms)/Mo, (9)

where Mo is the signal intensity without off-resonanceirradiation and Ms is the signal intensity in the imageacquired after off-resonance irradiation. MTR valuestend to be highest in white matter (since myelin appearsto provide the necessary chemical and physical envi-ronment to promote effective magnetization transfer).Quantitation of MTR appears very promising in thecharacterization of white matter disorders such as mul-tiple sclerosis (24). Nonetheless, most tissues exhibitsome degree of MTR. More fluid moieties (e.g., blood),however, tend not to exhibit magnetization transfer ef-fects and retain similar signal intensity after such off-resonance irradiation. This difference between bloodand tissue is commonly exploited in MR angiography(the subject of another article), in which extra contrastbetween (bright) blood and (dark) background tissue isachieved by applying the off-resonance irradiation andexploiting the magnetization transfer effect to furtherreduce the signal intensity of background tissue struc-tures.

High-Field MRI

Recent advances in MRI hardware have seen the in-creasing clinical adoption of MRI systems at fieldstrengths higher than 1.5T. The vast majority of thesesystems operate at a field strength of 3T (although sys-tems at 7T and higher are emerging). While it is beyondthe scope of this review to deal thoroughly with issues of

high field MRI, a brief summary of the influences ofhigher field strengths on the above image contrasts andacquisition mechanisms is warranted. A number of re-cent texts and collections address these issues in sig-nificant detail (e.g., Ref. 5). The driving motivation to-ward high-field MRI lies in the Boltzmann statisticsunderlying MR signal intensity. While noise scales ap-proximately linearly with magnetic field strength, “sig-nal” increases approximately quadratically; as such,the SNR of an image might be expected to increaseapproximately linearly with magnetic field strength. Asis generally true in MRI, this increase in SNR can beexploited in several ways: either to simply increase im-age quality directly, or to permit smaller image pixels(with reduced water content)—i.e., to increase spatialresolution, or to image faster (increase temporal reso-lution), which might otherwise come at a prohibitivecost in terms of SNR reduction.

However, in practice, there are several other pertinentconsiderations associated with higher field MRI thatmay compromise the apparent gain predicted above.First, it is generally true that T1 relaxation times oftissue become prolonged at higher magnetic fieldstrength; thus for conventional spin echo pulse se-quences, as discussed earlier, longer TR values wouldbe required to elicit a similar degree of T1-weighting,thus increasing image acquisition times. Worse still,not only do tissue T1 values become prolonged, but theytend to converge, rendering good T1-based soft-tissuecontrast hard to achieve using such conventional spinecho approaches. The generally-adopted solution tothis problem is to learn from the STIR and FLAIR se-quences and to apply an inversion pulse with a recoveryperiod optimal for distinguishing between gray andwhite matter (700–900 msec), prior to a fast 3D gradi-ent recalled echo volumetric acquisition. Such an ap-

Figure 23. Schematic of the magnetization transfer effect. a:Free water protons are characterized by a sharp resonance,whereas the pool of “bound” protons has a much broaderresonance, which is still impacted by off-resonance RF excita-tion (arrow). b: After irradiation, the “bound pool” protons aresaturated (represented by gray lettering), while the free waterprotons are unaffected. c: Chemical exchange between theproton pools transfers saturated magnetization (gray “H”s) tothe free water pool, which is reflected in (d) in a diminishedspectral peak amplitude (solid sharp resonance), compared toa similar acquisition without the off-resonance irradiation(dotted sharp resonance).


proach is often referred to as a rapid gradient echo with“magnetization preparation” or MP-RAGE.

Similarly, sensitivity to magnetic susceptibility differ-ences increases with magnetic field strength (T*2 valuesare shortened). This can be considered as an advantageor a disadvantage. Sensitivity to tissue interface “sus-ceptibility artifacts” is generally exacerbated, requiringshorter TE choices in the pulse sequence. On the otherhand, sensitivity to BOLD contrast mechanisms and/orparamagnetic contrast media in perfusion studies isalso enhanced, improving the quality of fMRI examina-tions and potentially reducing the injected volume ofcontrast agent.

Other considerations include the increased spectralseparation between metabolite peaks in MR spectra,which is offset somewhat by the broader spectral lines(since linewidth relates to local T*2). Furthermore, insome settings (e.g., imaging of the orbits), this in-creased spectral separation can be manifest by increas-ing profound chemical shift artifact (fat–water misreg-istration). However, this is generally a minor concern inCNS imaging, and may be ameliorated, whetherthrough fat suppression techniques or through pulsesequence changes (increasing the acquisition band-width) that reduce the impact of the chemical shiftmisregistration.

Finally, the RF system becomes less efficient athigher field strengths (higher RF frequencies), suchthat stronger RF pulses are required. Increasing theamplitude of an RF pulse linearly increases the powerdeposited in the subject quadratically. Thus there isincreased potential for subject heating. The power dep-osition is generally described in terms of the SAR andsince safety limits prohibit exceeding defined SARthresholds this can lead to compromised scan perfor-mance (e.g., the need to decrease TR or reduce flipangle). A variety of solutions exist to address this issue,including modulating the RF flip angle (e.g., VERSE andhyperecho techniques), or invoking parallel imaging toreduce the number of RF pulses applied in the echotrain (typically in FSE pulse sequences).

Despite the potential confounds at higher fieldstrengths, it is becoming clear that many, if not all,neuroradiologic applications are considerably en-hanced by the migration from 1.5 to 3T and that higherfield strengths will become increasingly prevalent inthis domain.

In summary, MRI of the CNS offers a broad range ofcontrast mechanisms reflecting different aspects of thephysicochemical microenvironment and probing thefunctional viability of the tissue at the cellular andmicrovascular level, as well as emerging interpretationin terms of functional organization and connectivity ona larger-scale systems level. By analogy, when consid-ering these contrast mechanisms, especially T1, T2, T*2,and diffusion, as colors in an artist’s palette, this reviewhas described the principles of neuro-MR in such a waythat the tailoring of neuroradiological protocols can beconsidered as the selection and weighting of these con-trasts in the composition of the ultimate image.

REFERENCES

1. Kucharczyk J, Moseley M, Barkovich AJ. Magnetic resonance neu-roimaging. Boca Raton, FL: CRC Press; 1994.

2. Stark DD, Bradley WG. Magnetic resonance imaging, 3rd edition.St. Louis: Mosby; 1999.

3. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE:sensitivity encoding for fast MRI. Magn Reson Med 1999;42:952–962.

4. Sodickson DK, Manning WJ. Simultaneous acquisition of spatialharmonics (SMASH): fast imaging with radiofrequency coil arrays.Magn Reson Med 1997;38:591–603.

5. Special Issue: 3.0 versus 1.5T imaging. Roberts TPL, editor. Neu-roimaging Clin N Am 2006;16(2):217–370.

6. Hajnal JV, Bryant DJ, Kasuboski L, et al. Use of fluid attenuatedinversion recovery (FLAIR) pulse sequences in MRI of the brain.J Comput Assist Tomogr 1992;16:841–844.

7. Roberts HC, Roberts TPL, Brasch RC, Dillon WP. Quantitative mea-surement of microvascular permeability in human brain tumorsachieved using dynamic contrast-enhanced MR imaging: correla-tion with histologic grade. AJNR Am J Neuroradiol 2000;21:891–899.

8. Villringer A, Rosen BR, Belliveau JW, et al. Dynamic imaging withlanthanide chelates in normal brain: contrast due to magneticsusceptibility effects. Magn Reson Med 1988;6:164–174.

9. Rosen BR, Belliveau JW, Vevea JM, Brady TJ. Perfusion imagingwith NMR contrast agents. Magn Reson Med 1990;14:249–265.

10. Ostergaard L. Cerebral perfusion imaging by bolus tracking. TopMagn Reson Imaging 2004;15:3–9.

11. Calamante F, Gadian DG, Connelly A. Quantification of perfusionusing bolus tracking magnetic resonance imaging in stroke: as-sumptions, limitations, and potential implications for clinical use.Stroke 2002;33:1146–1151.

12. Detre JA, Leigh JS, Williams DS, Kortesky AP. Perfusion imaging.Magn Reson Med 1992;23:37–45.

13. Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonanceimaging with contrast dependent on blood oxygenation. Proc NatlAcad Sci USA 1990;87:9868–9872.

14. Reichenbach JR, Venkatesan R, Yablonskiy DA, Thompson MR, LaiS, Haacke EM. Theory and application of static field inhomogeneityeffects in gradient-echo imaging. J Magn Reson Imaging 1997;7:266–279.

15. Haacke EM, Xu Y, Cheng YC, Reichenbach JR. Susceptibilityweighted imaging (SWI). Magn Reson Med 2004;52:612–618.

16. Kwong KK, Belliveau JW, Chesler DA, et al. Dynamic magneticresonance imaging of human brain activity during primary sensorystimulation. Proc Natl Acad Sci USA 1992;89:5675–5679.

17. Mansfield P. Multiplanar image formation using NMR spin echoes.J Phys C 1977;10:L55–L58.

18. LeBihan D, Breton E, Lallemand D, Grenier P, Cabanis E, Laval-Jeantet M. MR imaging of intravoxel incoherent motions: applica-tion to diffusion and perfusion in neurologic disorders. Radiology1986;161:401–407.

19. Moseley ME, Cohen Y, Mintorovitch J, et al. Early detection ofregional cerebral ischemia in cats: comparison of diffusion- andT2-weighted MRI and spectroscopy. Magn Reson Med 1990;14:330–346.

20. Rowley HA, Grant PE, Roberts TPL. Diffusion MR imaging: theoryand applications. Neuroimaging Clin N Am 1999;9:343–361.

21. Moseley ME, Kucharczyk J, Asgari HS, Normal D. Anisotropy indiffusion-weighted MRI. Magn Reson Med 1991;19:321–326.

22. Basser PJ, Mattiello J, Le Bihan D. Estimation of the effectiveself-diffusion tensor from the NMR spin echo. J Magn Reson B1994;103:247–254.

23. Mori S, Crain BJ, Chacko VP, van Zijl PC. Three-dimensional track-ing of axonal projections in the brain by magnetic resonance im-aging. Ann Neurol 1999;45:265–269.

24. Grossman RI, Gomori JM, Ramer KN, Lexa FJ, Schnall MD. Mag-netization transfer: theory and clinical applications in neuroradi-ology. Radiographics 1994;14:279–290.


Documents

Neuro MR: Principles - Queen's University · tem (CNS) and the harnessing of these physical principles to form clinical imaging protocols, tailored to suit a range of clinical presentations