Está en la página 1de 19
Tutorial A Perspective on K-Space' * State of the Art 1 essential difference between ragmetic resonance (MI image ingand other medical imaging mo. dates x the contro thatthe user has over hovr the data sre acquired, ma- nipslated, and reconstructed for sewing, ust by changing some soft ware controle~such asthe pulse se gqence timing the order of data ae Guiston, and the strength and rate of change of sary magnesie neds the user can modi resolution, field ef view (FOV), contrast, speed ofa uisiton, effect of anifact an so on The agent ofthis control ikespace, the abstract platform onto which dats are acquired, positioned, and then transormed into the desired image Nosuch agent exists for xray inal ing ltrasonography, or poston emission tomography, and while an analog does exist forcomputed to mnography (CT), access tt i crue nd mechanical and lacks the ense fn flexibility of the MR techniques, These methods cannot support the ch interaction between the user and the image svalable wth Mit imaging “The price for this ich interaction is the need for an understanding-bet yet an ntlve understanding of the concepts and mechanisms of space manipulation, The problem that k-space is an abstract notion, not Indexterms: Magnetic resonance (MR, space + Magnetic resonance (MR), physics = Magne renance i hk oor + Radiology 1995; 195297-315 "om the Lanse Imaging Cee, Robert son Mail Sool Unies nd Deity of Nev ere, 1 FrencSt New Brunswick Nf OSU tothe Radiology Croup of New Branowict Ny Re ceived Soe St 194 revlon reused De ember? svsion receives Janay 2199, fecrped January 27 Address eprint requests tibet, esshin oes something one can touch o fee or even shake to get a idea of how it sworks. Even when the contents of Espace are vstalized, the data have Title meaning and no apparent rele tionship tothe MR image. The math ematical constructs that expan espace in great deta are fairly sophistcated tnd do litle to give an ntutve fel for what s going on Tm actual fact, the concepts ofthe k-space are simple and have been known and understood for several centuries, The concepts that ofthe Fourier transform which ira series of equations developed in the 180050 help understand the flow of heat and Which has been extended over the past hundred years explain the flow of corrent in electric circuits, the design of muscel instrument, the spread of microwaves rom a satelite antenna and the ceeaton ofan image by passing igh through alens twas introduced to MR with the invention ofthe spinsvap technique by Edelstein eta (1) and hasbeen exploited with ever increasing sophis Eecation over the past fee year, In this article I will view the con- cepts of espace. To avoid dealing With abstract notions, Toil take antage ofthe fac thatthe concepts and equations involved with espace are exactly the sane as those involved with the actions ofa simple lens-—the kind used in telescopes, microscopes, and photographie eameras Simple Concepts suchas resolution and FOV, ormore sophisticated matters of m- age wrap, ghosts motion arfacts or segmentation are not unique to Mit imagers but have been understood and used by lens designers and even camera buflsfor centuries. Exploiting this lose relationship between Row k-space works and what happens ina Jens, Twill beable to explain the ab strec notions of kespace by relating them tothe nations understood by everyone who has ever used aca Finally, I should note that there are many excellent and quite detailed discussions of k-space to be found in the literature (2-7), My goal here is to give the reader an intuitive under- standing of k-space so that it may be incorporated into everyday practice WHY IT IS CALLED K-SPACE ‘The short answer stems from the fact that "k" isthe name of a term. found in the general equation that describes the MR signal. That term, which has the units of spatial fre- quency (cycles per centimeter), has three components, usually called k, k,, and ko, These three components are coordinates in and actually define a domain, or more colloquially a ‘space,” which we have come to call k-space, The choice of the letter” is based ona tradition among physicists and ‘mathematicians to use that letter to sand for spatial frequency in other similar equations (eg, equations that describe the propagation of light or sound o radio waves) and has no particular significance at all The first use of the letter k in such equations seems to be lost in history and in any event predates the invention of MR The long answer, including a deri vation of the general equation, has been relegated to the Appendix. .+ AND THERE WAS LIGHT Consider first the ordinary optical (je, photographic) image (Fig 1). Cre- ation of such an image is a two-step process. Firs, light illuminates (ex cites) the object, and the scattered light is collected by the lens. Second, the lens processes (operates on) the Abbreviations: FOV = field of view, RARE ‘pid squition with rlaaton ennoncement, RFs mado frequency, TE = esha timer TR = repeution time ® Figure 1. (a) A two-tens systom favored for Fourier transfer experiments, With the object placed atthe front focal plane ofthe lft ens, the Fourier transfer of the ight from tar object {is formed at its back focal plane (). The second lens performs a Fourier transfer ofthe light at that plane, which creates the image a its back foal plane. The plane hallway between foe lenses is calle the Fourier transfer plane. Simple fiers light stops and aperares placed at the Fourier transform plane can be used to create complex patterns and pesforin optical fanc- tons, (b) The simple, single lens is functionally dential to the doble lens of a. The Fourier ‘transfer plane isinside te lens. Although filter, stops and so on cannot be placed a the Fou: Fier plane, placement of them close to the ens hat smi effects The ins dlameter, for ex. ample, affects image resolution by plane. () Inthe case of two lense ling the alfected diameter of the Fourier transfer 1 sane foal lengthy the larger Tene will have better "Tesoluton (e, wil focus the light to smaller point, light to create the image. The charac- teristics of the image, such asits res0- lution, size, contrast, and so on, are in large measure determined by the pro- cessing, As one important example, the maximum resolution of the image is inversely proportional to the size of the lens: The larger the lens, the ‘smaller the detail that can be resolved in the image, as evident to anyone who has seen or used a “close-up” lens. Its instructive to understand this relationship, ‘The mechanism by which the lens actually does the processing—trans- forming the light passing through it into an image ata plane some dis- tance behind it—is based on interac- tion between the lens curvature and the speed of light in the lens (which is slower than that in air). This interac- tion bends the light passing through the lens, slightly at the center and ‘more at the edges. f the lens has been designed correctly, the light from ev- erywhere on the lens surface will meet at one point. Ina perfect lens (that is, one without aberration), the definition (ie, sharpness) of that point will depend on how steep an angle the light makes with the fens (or im- age plane); the steeper the angle, the more precisely the image point can be defined. An analogy can be made to an ax: One with a wide head can be honed to a sharper, more defined edge than one that is thin. In any event, imagine a lens whose plane is completely illed with light from a distant object: The larger the lens, the ‘greater the angle of light arriving at 298 « Radiology the image point. Thus, an inerease in lens size causes a decrease in the size ‘of the image point and so improves the resotution, This concept, which applies as well toa"burning’ tens used to start fre as to a photographic lens used to take a picture, was undoubtedly appreci- ated in very early times but was not really understood until sometime around the 17th and 18¢h centuries, when several physicists and math- cematicians (Huygens, Fresnel and Young, among others)8), began to explore the wave nature of light. As part of that exploration, some of them experimented with diftaction grat- ings and noted that ight passing through such-a grating will also be bbent—somewhat like light passing through a lens—bat the amount iis bent is proportional to the spatial fre- ‘quency (the number of transparent and opaque segments per millimeter) of the grating (Fig 2) They reasoned thata lens could actually be replaced by a collection of gratings with the spatial frequencies that are included proportional tothe curvature and rs- dius ofthe lens and, more important, with the highest spatial frequency. proportional to the maximum radius of the lens ‘As they continued with this reason- ing it became clear tha there was an alternative explanation for the action ofa lens. One could imagine that it was te light that was actualy com- posed of collections of different spa- tial frequencies and that the lens Was actually a special sort of filter that let more of fever ofthe frequencies * through fn ths constr the bigger thelesrthe more hequendes ae lowed fo pas throught ad the more teens tat could pus throug te beter would ete tite nmge rslaton The ns, cours has tebe moe i jus Seri aleo mest convert and com: bine dhe spatial quence of the light pal though tint the pein fig tht nae up ths ZeerThe sof th altertatve sel for he on shat these Seuble Kanone tern converson= are rathematcaly destibed by sn lgo Stn known Sethe Fouseane {orm The csence of enya least faras the atheputian eco shale the uologs) cone cemed that iti seethng tat tes dota ep unas waves gh thatesit nape and perormes Fourier andl on tet ete tnimagein another plane (Re) "he ata thatthe en operates on, shetight atthe lene plane's repost, scolstion o undies or packer pattem of ight each st aifecene {pata equine). The collection cn iteprneted sa to-dinenlnal tray (gs). The coordinator of he aeray ane spatial reguencies nthe two orthogonal drectong an the Inighinetat each port nthe ary i Proportional toh enegy inthe ight tr ikove pal eeguens. Since ts i atwocdimensiond aay twos tol equentesareasoadied wth feck pantone in (ote horton tdimension) and ons iny Weve May 1995 tai te car ine spe on the kes Vo! collection ot gratings wits aero spatial ‘eoquences Figure, m8 8. ere Os iversely proportional tothe ine spacing of he pratng elements.) Lars rt Simulated by s collection easy to achieve inthe MR imaging system cal dimension). This plane is called the Fourier transform plane Consider now the MR image (Figg 3c). It too is created by means of a two-step process. In the first, radio ‘waves excite the object and the resul- fant energy from the object is col- lected. In the second, the data are processed into an image. The details of how the radio waves are generated and acquired are interesting but with few minor exceptions are of litle in- terest to us now. What is of great in- ferest is that the second step of the imaging sequence, the data process ing, isa Fourier transform, the very same Fourier transform used in the second siep of the optical image se- quence. However, this means that there must be a data plane, analogous to the Fourier transform plane in the camera, that consists of signal or data intensity patterns, each at different Spatial frequencies, that is operated ‘on by the Fourier transform to create the image. That plane exists; we call it k-space, and itis exactly analogous in i Volume 195 * Number 2 Lela a cone oe eer tomas colaton of etl ' 7 om ee (4) Simple optical grating, Part ofthe light incident from the left is diffracted at an tings High spatial frequency (le, Small line spacing) of gratings ‘upper and lower aspocs drat ight at steep angle, and lower spatial Requorey oto ings near center difracts ight aa shallow angle. With cteful selection ofthe lange and kane tion ofthe diferent spatal frequencies, all the diracted beam focus) which simulates « conventional lens. This arrangement i Fresnel lens (et acting on a homogeneous bean of light homogeneous fier (eg, something as simple asa pane of clea gas) acting on apataly ‘modulated beam of ight. Such spatial modulation i dificult to achieve with meet at one point (the led & Fresnel fens (e) The If functionally identical to» Tight buts vey function and concept to the two-d mensional array of lightat the lens plane (Fig 4). ‘The implication, and indeed the reality, of this is that the entire pro- cess of MR imaging is closely analo- {gous to optical imaging, and concepts from one can be used {6 explain con- cepts in the other. The mostly abstract concepts that are basic to MR imag ing, especially those that inwolve k- space, can be explained and under- stood by reference to the mutch more familiar and concrete concepts used in photography, microscopy, and ‘other forms of optical imaging, [shall use these analogies to explain how. k-space controls some of the basic fea- tures of the image such as resolution and FOV, and Ishall demonstrate how k-space is manipulated in some of the more sophisticated imaging sequences such as echo-planar iniag- ing, fast spin-echo imaging, seg- mented k-space imaging, "key hole” imaging, spiral imaging, and others, I hope, through the-use of the optical analog, to give the reader an intuitive feel for these and other k-space manipu- lations so that these manipulations can ‘be incorporated into daily practice. BASIC IMAGE CONCEPTS. [wil start by explosing some of the basic concepts of imaging, whether it is witha camera or with the MR imag- ingasystem, First we must recognize that there js no one-to-one correspondence be tween a point in espace and the inne age, just as there is no point-to-point correspondence between a politi the Fourier transform plane of the camera and the image, Point in the Jmage receive contributions from ev- ery point in k-space, and data from every point in k-space contribute every point in the image Places of dst or lint on a camera Jens (and here Tmean the lens sel, not the filter or cover glass up front) do not appear as spots in the image but rather cause a cifse “haze” in the image due to light scattered over the whole image plane. People with vacuoles or eataracs in the fens of their eyes do not see spots or focal de- fects but complain of bersiness or de- creased contrast Theres no one-to-one correspondence betwen aspoton the Knrantnopotonthefimeree ‘This is tte not only for discrete spots atthe lens but also for large ar- tes, Cloning theirs ofa lens (whether ina camera, microscope, or eye) dacs not obscure the outer rin of the im: age bat rather decreases the total brightness and the total sharpness, or resolution, of the image. Every point inthe image that was present before ~ the iris was narrowed willbe present afterward the outer points as well as the inner points. However, every point in the image will be affected by the changes at the Fourier transfoors plane (or k-space) ‘This brings us fo the second point, which is that while every point in the Fourier transform plane (or kspace) contributes to the entire image the zany in which i contributes depends ‘on where itis inthe Fourier transform plane. in particule, it tums out that points at the edges of the Fourier transform plane contribute to, and indced determine, image resolution The farther a point is from the center of the Fourier transform plane, the more it contributes to the sharpness or resolution of the image The more points there areat the edge, orequivay Jently the larger the diameter ofthe Jens, the better the resolution. Ina telescope, the bigger the lens or m= dria tobe wanstomes (iouier ‘rancor lane) Fleures@) Simple optical lens.) The “fective” lens which consists of patily mod {ated light ator nea the Fourier trnaform plane and after that perionns s Rowse trove on the igh at the transform plane (2) Lens analog MR imaging stem that cone are ror the smaller the star that can be Seen. Ina microscope, large ens allows visualization of small deta In a camera seo gp lens or open of teins inereases the resolution The reason for this, ofcourse is that aninceasein the dinmeter of he lensor Ainerease in the aperture of espace [pets higher pata eguentcs pas tothe mage and tee higher Spatial frequencies (ts jst dsc in ‘regard to the fens) that detemnne sharp hess orreslution ofthe mage This relationship between ies size and resolution is sed on thes Suumption that thee is no aberaton, Iman actual ens, because of a. matic aberration, coma asigmmbm, ad soon, the relationship valid only upto a point beyond that point resolution actually deteriorates Fina, Pily forthe mont part these shore tons do not exis n espace (nonn form gradient magnetic field actully will ease aberatone slr to those scenisagasslens but tsi easer to shape magnetic fields thon lace elements, and most manaiocueon have largely eliminated these distor tons conse sae ppsponesofthisaride that space as ho aberrations “Tolllustate the relationship be. Lween resolution and lene aperacare, Figure 3 shows a sequence a Mig ages taken with diferent aperanes ry espace The fit (Fig 5a, 50) shows thelimage and the fully Opened sper- turein k-space, the second (Hg 9, 3) shows the image wth te apettere diameter reduced by hale the fed shows the effect of ducing the sper ture by half again and the oun shows the effect of reducing the aper- turehy yet another half for tort reduction in aperture diametes of one-eighth). The points to ote are that each aperture reduction causes a reduction ininage sharpness but that even with sth large redactions Inthe effective area of Espace the entire image's seen. This of couse, i exactly wht happens when ne, duces the lens ns ona caress oF changes objectives ina miconcope tnd should come ano surprise One further thing to note i that in addition toa reduction in resolution, the effect of decreasing the apertu is to create and increase “edge secs” (cringing at shoxp marply these tre cated by an ieuticent moor of spatial frequencies. A sharp edge or line contains many (theoreti on infinite number) of spatial frequen. Gies_both high and low. By dere ing the apetare one eiminetesthe high spatal frequencies (an s0 catse uring br dere sap. nes) but leaves the low frequencls, Some of those lovr frequenctesctny important information about the con- trast and width of the line, but some of those low spatial frequencies carsy information about edge sharpmese > information that was o be sare Ge, compensated fr) bythe (now cling hated) high spatial requencles Ie these “uncompensated low spatial ences that ive set the “edge ce iasrnls Pe Tithe edlgesof espace control image resolution, a easonbe question tose ‘Swhat sit hat he censr oF apace controls? It turns out that thé center of k-space determines overall contrast, the rato of ight to darn the mage Continuing the experiment de. sated in Figure 3 if we mate the apertare i, diameter) of espace very Stall (1/32 39% ofthe ful diametel 28 in Fig 69,6) the image contana Title ta bt does show vaguely, Thecontan at deren: port of hee” image. By increasing the aperture in stages, one increases the detail (Fig 60-69). Taken together, the results of Figures 5 and 6 show a continuum of effects, with data farther from the center controlling resolution and data loser to the center influencing, con- trast. It is important to note that there is no sharp divide in the sense that points beyond a certain aperture dic ameter control resolution and points within control contrast; rather, the effects of aperture changes are grace- ful, with a gradual change in image quality (mostly resolution) with changes in aperture diameter, This, of course, is the experience with photo- praphie cameras, in which even fairly large changes in the aperture (some- times called f-stop) cause relatively small changes in resolution. (Lam ig- oring for now the other effect of changing the iris, that of changing the image brightness. The reason for this, is thatin MR imaging the image brightness is artificially manipulated at the image console or in the laser camera, and there really is no direct relationship between expected image brightness due to changes in k-space aperture and actual image brightness) If we accept the conclusion that the location of data in k-space (ie, toward. the edge or near the center) affects image resolution and contrast, then the next question to be resolved is, how do we in fact control that place- ment of data in k-space? FILLING K-SPACE MER image generation is, as Thave said, a two-step process, The last step, which kespac Fourie age.It exactl, isimp: that th the fir: vide tt data a (0,-1) Figure 4. The Fourier transform pl the spat fourier transform plane spatial | Feawony in y direction spatial frequency in x direction | __ lane (Kspace). The spatial frequencies in he plane vary With location; they are higher atthe margins and lower atthe enter, At the very cen (0.0), frequency slow (hatch marks separated), while at the comers (HLL lends ‘1/41; +1,=1) the spatial frequencies ae high (hatch marks close together). At the ezner of {he right margin (+10), the spatial frequencies will be high along the x direction ana low Song the y direction. At the center ofthe bottom margi they dinection and low along the dieecion. The important point sth spatial frequencies are high slong, the epatial frequen siesin x and are independent snd are determined in each case by means ofthe distance from the asi hich Ihave just described involves taking data that have been placed in espace and performing a tansform a Fourier transform, to create the ite age Its not important to this exactly how the data got to kapece I isimportant only to the transform that the data be there, eis the task of fhe fist step of the MR process to pro- vide that dt and to ensure that the data ae properly placed in space ‘Volume 195 * Number? ‘The gradients, and in particular gradi- ent strength, control the placement of data in k-space. Image data (ie, the radio-frequency [RF] echo from the excited spins in the patient) acquired with (in the presence of) small gradients will be placed near the center of k-space, while data acquired with large applied gradients will be placed farther from the center. Just by manipulating the gradient strength that is applied ‘while data are acquired, we can con- trol the resolution and contrast of the image. Gradient strength in the MR system is the equivalent of iris diam- eter in the photographic camera ‘There are, in the usual case, two gradients that directly influence im- age quality—the frequency gradient and the phase-encoding gradient. (The third gradient, the section-selec tion gradient, controls image location, orientation, and thickness, but none of these affects or is affected by ma- nipulations in k-space and so this gra- dient will be ignored for now. We simply assume that somehow or an- other appropriate sections are se~ lected and that we are concerned only with image quality in that section. I will return to this point later.) While there is no correlation be- tween points in k-space and points in the image, there isa direct one-to-one relationship between points in kspace and gradient strength. Image data acquired with the phase-encoding, gradient equal to +Gp will be along. the rightmost margin in k-space (line in Fig 7), and image data acquired with the phase-encoding gradient ‘equal to 0 will be along the central vertical axis in k-space (line b in Fig 7). ‘The line in k-space at the leftmost, margin (line cin Fig 7) willbe filled when image data are acquired with the phase-encoding gradient set to its ‘maximum negative value G,). By simply adjusting the gradient strength (and sense, ie, whether pos tive or negative), we precisely deter- mine placement of data in k-space. Which points in k-space are filled and the order in which they are filled are determined by the pulse se quence. The remarkable thing about (MR imaging is that the pulse sequence designer—and through the designer, the radiologist—has complete control over image quality by virtue of the control over how and when k-space is filled. If k-space is made lange (le, a large-diameter aperture), then the mage will have high resolution, If only the central portions are filled, then the image will have lower reso- lution. If, for some reason, only the edges but not the center is filed, then the image will have high resolution but poor contrast (it will, in fact be an edge-enhanced image, as seen in. Fig 6b), ‘One might wonder that since the {goal of imaging is usually to produce a picture with the highest quality pos- sible, why would the pulse sequence designer not ensure that k-space be made as large as possible (commenst: Radiology * 301 L F iat comands seep chaning the size oF pace.) A256 x 56 sagital MR image an (hs shite Iepact data (aca of he tn cause propane! the photograph the peripheral pets ofthe k space dts are ie ose) Sera cis east en oa cde ar ane gh etal The sae of espace i reduced by one-half ne and d by another one-half ne aah oan he | oa Sharp main cone by asa ado oa reditionnresltin there am incense inthe “eg eae eee ne i sharp margins caused by an insuficient numberof spatial equcnece a tt | & st Figure 6. (2) MR image and (b associated k-space data d onstrate that reduction of k-space tothe extreme (1/32 of mainder prea esa it preservation of cont (a) (ed) The vere situation wih th ener eC a | ‘ele preserve ress inan mage with pred deal bl ls fcena Ths banat oa RB oF | de evaiabe) and Wat ah yeas, hat the paint or at eas part ofthe A second answeristhatinceasing | Heda gend hatall of k-space be patent) will move during the acqust. the ssermcymatver that incre si filled all the time? One answerissim- tion. Patient motion eon Eluh aca ate kespace decreases the size of each im. | a Erotic makergccPece ke ne and cortheimage ust as with photo” agecoment Shee hace cal tee es Ss -,C Aime filing alarge Kspace ifitislkely Fesdluton con eotonee a Ee Figure 7. Filling of kspace. Line is illed when the phase-encoding gradient ampl tudes large and postive Line bis filled ‘when the phase-encoding gradient ampli= tudes zero. Line cis filled when the phase. encoding gradient amplitude sarge but negative. $6f indicates maximal positive and ~Gfindicates maximal negative spatial fe ‘quencies along the frequency encoding di fections. Gp indicates maximal positive and ~Gpindicates maximal negative spatial fre quences long the phase-eneotng ietons ment size reduces the number of hy- Srogen atoms contained in each ele- ment, which leads to a reduction in signal and therefore a reduction i signal-to-noise ratio, In situations in which signal-to-noise ratio is limited, the system designer and radiologist ‘may have to compromise on resolution. A third answer involves limits on the length of each line (and hence the resolution) along the read-out direc- tion. That length is proportional to the product of the read-out gradient strength and its duration (je, how Jong itis left on). Increasing the gradi- ent can put unsustainable stresses on the gradient amplifiers, which leads to inhomogeneities in the magnetic field. Increases in gradient strength also lead to increases in read-out ame plifier bandwidth (a consequence of the Larmor equation, which relates frequency to magnetic field), which tulimately leads to decreases in sig- nal-to-noise ratio. Increasing the du- ration can lead to undesirable in- {creases in echo time (TE) and, because of susceptibility (T2*) effects, further decreases in signal intensity, The real goal of pulse sequence de- sign then is to fillas much of k-space 48 quickly as possible while trying t0 balance image quality and speed, In the following sections I will ex- plore several approaches to image Volume 195 « Number2, acquisition and display that illustrate different methods of k-space manipu- lation. Some of the examples employ pulse sequences that stress speed at the expense of image quality (echo planar, keyhole, spiral), while others ‘im for high resolution in reasonable acquisition times (fast spin echo, eg. mented k-space). Along the way I will stop to examine some additional fea- tures and limitations of k-space ma- nipulation, SPIN-ECHO TECHNIQUE ‘The basic“spin-warp” technique fils espace one ine aa time, a rate of one line per 0 RF pulse (ig 8) (). The length ofthe lin (along the fequencyrencoding directions Proportional the maximum strength Of the read-out (which is the same as the frequeney-encoding) gradient and to the duration of that gradient The Postion ofthe line is determined by the value ofthe phase-encoding gree dient. Figure 8 shows eight lines kespace and the corresponding pulse sequences used to postion thee line. Since only online placed in kespace or ench 97 RE pus, the time (0 il espace i given by NTR, where Nis the numberof phase en coding lines and TR is the pulses a. 256lines ofl espace (and we ssl See ina moment what determines how many lines are actually ned), and i Tis 2 seconds, then would tae 512 seconds (over 8 minutes 0 Fllicspace, Notice that k-space isrecangular cather than round ike most camera lenses, and while that does nat realy change the analogy to the camer, does give espace manipulators some - extra freedom and lexility For ec ample, consider the pulse sequence designe who wants to decree the image acquisition time ofthe spin. cho sequence Since each hin ke Space along the phase-encoding de rection takes TR seconds to fll'a reduction in he mummber of phase: encoding lines would give a propor: tonal reduction in time Tein the exe ample above only 64 lines of k-space werefilled, the me would decrease from 84 minutes about 2 mines, a considerable savings. The consequence to the mage ofsuch a reduction de, pends on Row kespace i actually filed. There are two posses I we keep the spacing between each line of k-space constant (oy keeping the increment in magnetic strength between each phaseencod- ing giadiont constant) hen when we decrease the numberof tines in espace by four and the sic Kespace along the phase-encoing direction must alse decrease by ac toro four Grom 286 tei) allthe lines willbe bunched areal te center of eopace, Since the sie of space contol image resolutions however and since the sige of eepace that filled has decreased, talons thatthe detail or resolution in the te age must ao decrease Native that ie have deeresed the sie slong Only one dimension of espace the hase-encoding direction) and Sea Fest we will decrease the reetion only along ane dimension (xs) of, the image: The emarhabl thing 6 that while the resclation of te frage decreases the sie of the image does not change. If we assume fore so iment hat the nanier of ines he image equa tothe number of ines inlespace then what happens that fac linen the mage gee “blue” fate by js the factor of four ne ery fotaintan image ee ‘Alternatively the pulse sequence designer (or radiologis) cout deckte tokeep the sizeof espace constent anal choose to increase the spac between the ines slong the phase. encoding direction instead Phe reso. tution inthe image, which is deters mined by the sie of keopace, would Hot change, but the image ice would dently have to change and woul decrease by the factor of four (twill explain win jas moment ‘hese are very diferent rests, which were determined entirely by Ineans of how keapace lle ‘The reason the mage size would have to change in this last ane, and the basis for our asumpion thatthe numberof fines n espace eal so the numberof ines inthe image, con be explained by refering bach four opti anciog Inthe usta imaging process, data in the Fourier transfor plane are Continuous, not segmented into sepa- Tatelines The effect of egmenting the data into neste tne can be Sep resented by superimposing a gad Brating onto the Founertanstorm lin. This has a dramatic eee be- fase the grating scatters the ight and causes the creation of multiple Copies (techie refereed toag, higher orders ofthe orginal image that are symmetrically paced tos. therside of that image (ig). (ine terested, the reader enn see thie by placings nylon stocking or pantyhose Over the head ora least in pont of the eyes and iooking ata bright ob- Jest agaist a dim bckgroun Mi Radiology + 303 tiple copies ofthe bright object willbe seen orented along te wep an woo! ofthe stocking tis bette do thisatnightand by lookingat tect light tight als be good to make sureno neighbors are aru to take in the sight of a radiologist wandering around at night with a stocking over his or her head.) The spacing between the higher orders is inversely proportional to the spacing of the grating lines (which is the same as the spacing of the lines in k-space). Placing the lines close to- gether spreads out the orders, while sncreasing the line spacing moves the orders close together. (The adventur- ous radiologist can demonstrate this, as well, by stretching the stocking and. noting that the spacing between the orders changes.) If the orders come too close together, they will overlap and thus cause the so-called wrap that is commonly seen in clinical stud ies when the FOV is too big, the num. ber of phase-encoding lines is too small, or the patient is malpositioned in the magnet. In fact, the image is not “wrapping” around at all. What ‘we see are the diffraction orders over- lapping (Fig 10a), ‘One way to understand this is to imagine that the monitor at the op. erator's console is really a window cut into the center of an opaque screen, All of the diffracted orders project onto the inside of that opaque screen, but only the images that project onto. the window can be seen (Fig 10b). In the ideal case, only the zeroth order projects onto the window and the higher orders hit the opaque screen and cannot be seen. If, however, the higher orders are not diffracted far enough away, they too will project onto the window, with part of the +1 order on one side and part of the —1 ‘order on the other side. It will look like “wrap,” but of couse itis not It is just the diffraction orders overlapping Figure 10cis an example of such “wrap.” The central image (between the vertical black lines) is what would appear on the console window fand be recorded on film). The images to either side of center are simulations of the +1 and ~1 diffracted orders (as they would be seen if we looked at the opaque screen that we imagine surrounds the console window) The ose pushing into the back of the pa fient’s head is not actually wrapping around from the front but is really caused by overlap of the diffracted order on to the central order. The spacing between phase-encoding lines is too wide to diffact the higher orders away from the central image. It nee: One lin is filed following, ach line depends on the phase: the lett of Sequence is conventional. section-select gradient i turned on concumast ak teats Re Peat gradients tured on concent with he sppeanace of tectin The eheee CHUNK radien is placed between the 9" and 180 RF pulses The strength of the enn coding gradients is sequentially stepped from beso steps determines how many: lange negative fo large postive values the num lines willbe in space, the value af the steps determines how close the lines willbe, and the maximum value (postive or negative) dene {Patlmum sizeof k-space and thus the mage resolution. ~g/isthe mount epee aah frequency along the equency-encoding deecion Is ike out friend with the stocking over his or her head pulling too hard and spreading out the stocking mesh, which brings the diffracted orders too close together. We could ask our friend to release the tension, but what we actually do can be illustrated by a Detter example. We imagine a trang- Parent elastic lens, one that can be stretched or squeezed along its ra. dius. We farther imagine that grid lines (corresponding to the phase- encoding lines we have been talking about) are etched in its surface. When ‘we stretch the lens to make it bigger the image resolution gets better, but if the grid lines get separated too far we get overlap, just like we see in Figure 0c. If we squeeze the lens to make it smaller the grid lines get closer to- ether, the diffracted orders spread apart, and we eliminate the overlap, just as shown in Figure 10d. The problem with this solution is that the image gets smaller and resolution de- creases (we did reduce the lens size, ‘which is the same as decreasing the ‘maximum spatial frequency in k-space), but that isa compromise we have to make to get rid of “wrap.” Another solution, which involves a different compromise, would be to mnaintain the lens size and decrease {he grid line spacing by increasing the numberof lines etched on ls surece This would correspond to increasing the nuntber of lines of kspace (8.10) The problem with this is that segue tion time s increased in proportan to the numberof lines added ht would be better to arrange data in k-space to avoid “wrap” in the hist place To avoid overlap, the FOV ofthe Smage mustbe ne larger than the separation between the orders, which means that asthe spacing between the lines in k-space gets bigger, the FOV ofthe image should get smaller. The inverse is also te~if the FOV i to increase the spacing between lines can decrease. This relationship be- tween FOV and spacing between the lines in k-space (B) can be wntten as ‘FOV al/D, (ly In this relationship, we assume that {he number of lines inle-space docs fot change, so that if the spacing be tween the lines gets bigger, espace ise (actually the mani spatial We previously fou that the rela p th to th i th Dy as du du vias ks tionship between aperture size in k- space (ivhich isthe same as maxirnam spatial frequency kg.) and line spac- ing in the image (dis gos Ud. @ If we remember that the line spacing ‘din the image can be calculated as a= FOV/Neand line spacing in ke space D can be calculated as D= kn/N {ihere Nis the number of lines in the image and Nis the nurnber of lines in espace), then we can show that N ° The consequence of the desire to prevent wrap” in the image is that the number af lines in k-space is equal ta the number of ines in the image. ‘Thus, in the example above in which the pulse sequence designer main- tained the size of espace but reduced the numberof lines (nd so increased D), the numberof lines in the image aswell asthe image size (FOV) was re- diuced proportionally. This had to hap. pen tomintain resolution. In the prior example, in which both the number of linesand the sizeof espace was re- duced together, D was not changed and soimage size was maintained, butimage resolution was reduced. Equations (1), @}, and (3) summa- rize the basic relationships between space and the image HALE-FOURIER AND. HALE-ECHO ‘As we have seen, itis possible to reduce image acquisition time by sim- ply reducing the number of lines the image, but this speed improve ‘ment usually comes with a propor- tional reduction in image resolution cor image size. There is one exception to this, and it stems from the symme- tty and inherent redundancy of data in k-space. Data stored at positive spatial frequencies are closely related to data stored al negative spatial fre- quencies and are nt tall indepen- lent, Knowing the value of data at ‘one positive spatia frequency, we can predict the value at a symmetric nega- tive spatial frequency. An intuitive sense that this must be so can be gained by considering a simple opti- cal experiment (again, the lens plane is the analog of k-space). Imagine the basic imaging system such as in Figure 1b. With the iris full ‘open we would get an image, and we could measure, or at least note, image quality, resolution, number of lines, and so on. Now imagine (or actually get out your camera and do it) that Volume 195 ¢ Number 2 sve cover half the lens—say the left half (as before, we cover half the lens, not the cover glass nearby, for the analogy to work). We notice that the image is now half as bright (afterall, we are blocking half the light), but the entire image is stil seen, just as we discussed before, I'we were careful to include the center line—the diameter (which corresponds to the very cen- tral line or zero frequency of kspace)— then we would notice that the con- trast inthe image has not changed. If wwe look very carefully we might no- tice that the resolution inthe vertical direction has not changed, but the resolution in the horizontal direction hhas changed and is now half whatit was before we covered haf the lens (This makes sense to, since we have reduced the numberof lines, or the bandwidth, in k-space) Now imagine (or go back to your camera and do if) that we remove the cover that is over the left half ofthe Jens and place it over the righ half Nothing changed! The image is stil half as bright, the entire image is stil visible (not just the right half, con- trast is preserved, and the resolution {halfas good in the horizontal as in {he vertical direction) has not changed. Either we did not do anything or the data inthe left half ofthe lens plane (espace) are the same as those in the right halt. Iti the later, of course, because of the inherent symmetry of allens (and of k-space) ‘The actual symmetry is a point sym- metry, which means that points in k- space are symmetric about the central point of k-space—poinis in the lower left, for example, are symmetric with (ie, the same as) points in the upper right. However, regardless of the ac- tual type of symmetry, the implica- tion i that half the data are redun- dant. Once we have acquired half of the data—that is filled half of k- space—we really do not have to ac- ‘uire any more because we know ‘what ie will bet ‘And that, of course is just what half Fourier (sometimes called half- NEX {number of excitations) is. Only half of kspace i filled (60 for 3 128- Jine image only 64 acquisitions would bbe needed) and the computer would fill in the rest of kespace (taking data from the lower let and copying ito points atthe upper right, and data from the upper left and copying it to the lower ight and soon) Weswould have to take care to include the center Tine to ensure contrast was preserved Once k-space was filled, the Fourier transform would be performed and the entire image, with fll resolution, ‘would be generated as if we had spent twice the time to fila of k- space! There is a cost for this dramatic im- provement in speed (there is no free lunch, and the cost ere is a redtie~ tion in signal-to-noise ratio. As I men- tioned, image brightness halved, It turns out that when we cover half the lens the noise power aso halves, with the result thatthe signal-to-noise ratio amplitude in the image is reduced by 40% (ie, (2). Other problems exist, as well including those that have to do wih the increased burden on the computer by making it copy the data, Inhomogeneities in the magnetic field, and attenuation of the RF field, for example, but these are of less con sequence (af least to us). The bottom line is that if there is enough signal- to-noise ratio to spare, then half-Fou- rier technique isan excellent way to reduce image acquisition time. IE there is not enough to spare, then one could compromise—say 3/4-NEX— and accept more modest time savings with less reduction (in this case only 15%) in signal-to-noise ratio ‘We can take the symmetry argu- ment one step further by noting that if there i left-to-right syaunetry, there must also be top-to-bottom sym- metry. Thus if we had performed the experiment by covering up frst the top and then the bottom ofthe lens, ‘we would have gotten the same re- sults. In k-space, however, this would mean covering up (or more accu- rately, not acquiring) half the readout line. That is precisely what is done in tive half-echo technique. Instead of acquiring the entire echo during the readout fora line in k-space, the ac- Guisition is aborted halfway through and the acquisition for next line is started. All the considerations regard ing image quality discussed above pertain, and the only cost to this tech- nique is the same 40% loss in signal- tomoise ratio, The benefit is areduc- tion in TE, which can be important in reducing susceptibility arifacts, The ‘more general term for this approach is partial-echo technique, when other than half the echo is acquired. FAST SPIN-ECHO TECHNIQUE ‘The basic idea of this approach, first suggested by Hennig etal (11-13) and ‘Mulkern et al (14) and subsequently ‘modified by Melki and colleagues (15-18), isto fill more than one line of k-space at each 90° RF pulse. The pulse sequence (Fig 11) has a 90" pulse followed by many 180° pulses, and instead of the phase-encoding, Radiology * 305 sitfaction crating with ag single spatial frequency Figure, (a) Optical grating shows multiple sifacted Bean ordee). The angular spac inghetween each oder ic inversely propor tional othe grating spatial equeney fe line spacing) In theory an infinite numberof orders are generated.) gating paced {tor neu the Fourier tnaforn plane, nal ipl images we created one fo ach fine tionordr Asbelore the separation between each image (order is inversely proportional to the tne spacing spatial eqteney ofthe {ating With smal spaces betwoun (Gig spatial frequency) the ‘wellsparsted ahd the higher oer can be nored. The ental mage hs the highest intensity andi he ome depayed tthe con- sole pulse being incremented merely once per 90° pulse, its incremented with every 180° pulse. Since a line in k-space is created and filled with each pphase-encoding pulse, this approach fs many lines kespace with each ‘90° RF pulse. The savings in image acquisition time can be considerable, In the extreme case it is possible to fill all of k-space with one 90° pulse fol- lowed by 256 sequential 180° RF pulses, which would allow total im. ge acquisition in just afew seconds. ‘The problem with such an ap- pproach is that there is T2 decay of ex: «ited spins, Each line of k-space will be smaller than the one acquired just before it by the ratioe- ©, with the amplitudes of each successive line in k-space decreasing in monotonic fash- ion until finally they vanish below the noise level ‘Taisis very different from the situa- tion in the classic spin-echo tech- nique, in which every line of k-space (acquited at the same TE) has the same amplitude, and this difference has dramatic effects on image quality. ‘The actual effect on image quality de- pends on exactly how we decide to fil k-space. We remember that the desired con- fourier transform, plane diffraction crating trast in an MR image depends on the ‘number of protons in each tissue and on the intrinsic tissue parameters TI and 12, but we also remember that we can modify that contrast by the values we choose for TE and TR, In the spi echo pulse sequence, every line of k-space is acquired with the same TE and TR, and so any or all lines con- tribute equally to image contrast. In actual fact, as I have discussed above, ‘most contributions to image contrast ‘come from lines near the center of k-space, while data near the edges of espace determine image resolution, The important point however, is that any data line, no matter when itis, ‘acquired in the spin-echo imaging sequence, could be placed anywhere in k-space without affecting image quality. In this new sequence, each line of data is different, and it matters very much where itis placed, We could, for example, fill k-space in the manner of the classic spin-echo sequence—from one edge to the other—but then the edges (or at least ‘one edge) would have high signal and the center would have relatively low signal intensity. In the case of a 1256 echo train with a TE of 20 msec for the first echo, for a tissue with a T2 , /. ciffraction grating with low spatial frequency value of 50 msec (muscle for ex- ample), the lines at the center of k space would be much more than 1,000 trillion times smaller (actually more than 104 times smaller) than the value at the edge of k-space and would appear fo have an effective TE cof 2,560 msec (ie, 20 x 128). (By effec- tive TEI mean that the contrast ‘would be similar to that obtained with @ conventional spin-echo se- quence with a TE of 2.560 msec). This would be an exceptionally T2-weighted sequence indeed! Tt would actually be more compli- cated than that. The effect of the high signal intensity at one side of Kespa would be to create an edge-enhanced image, which is similar in some re: Volume 195 * Number? A ' £ 4, vs spects to my previous example of a camera or telescope with the center of the lens blocked (Fig 6b). However, the effect of the T2 decay would be to make that very edge-enhanced, very Twveighted image practically invis ible! ‘We could alternatively fil the cen- ter lines of k-space first (Fig 12). In that ease, since for the most part itis the center of k-space that determines contrast, and since we are filling that center with data acquired with a short TE, the contrast would be similar to that obtained with the spin-echo se- quence with a short TE and long TR vhat some would calla proton-den sity-weighted image”) he While contrast would be similar to that in a Figure 10. (a) With large space between igratngs (le, low spatial frequency) the higher orders will not be separated and may ‘overlap the central image. (b) The overlap of the +Tand ~T diffracted order with the cen trlimage eeroth orden) the cause of erp. Themage doe not eal ap around: tas uke that way. The ete aie to insficient spacing beivcen gating es and 3 insticen separation ofthe ted orders Increasing te nutler of ines or decreasing the spacing between lines (esdecreasing the steps between phaser coding gradients) wil seperate the orders fnrthe and eliminate the wap.) Cmca ration of "erp" Gus to Overlap of ei irncted orders. The image seen a te conte window (wounded bythe vertical Back ines) really only part ofthe Total reconstructed image data. We never see the pat ofthe ‘ge dat that als outside ofthe consae wi dow, and thus we mistakenly interpret he hose oveying the bacco the hes s being ‘wrapped atin the age, (@) Redaction nthe ie of espace (ar expe by ing the gradient srengih) spreads the at Sted oer and thus eines "wrap bat tals reduces nage se and resolvon As in evoniy tat pat of the image dita tween the wert Bac ies oseen a he ing ajem console The scjacent orders ae not normally dipayed a they ae ts sition (erated tically bys conventional image, however, the resolution would not. Since each cessive line in k-space is less intense than the one before, as we move away from the center the lines of k- space get dimmer and dimmer until, like before, they vanish below the noise. The point where they vanish becomes the effective size of k-space, and this can be very small indeed Take for example the case in which the T2 of the material being imaged is 100 msec and the TE is 20 msec, After only 15 lines the intensity would be just about at the point where it equals the noise, and all successive lines tively vanish below that fe that 15 x 20 = 300 msec, which equals3 x T2, To a good eng Radiology + 307 nu. Figures 11,12. (11) Multecho imaging (RARP) In this sequence the 90° excitation pulse is followed by many 180° echoes that form pulses. ‘The data at each successive echo are ust to fila lne in kospace. The stength of exch phase-encading gradient (one is paced | | h | | 2. trench 180" [RE pulse) determines data line position in espace. The mechanism for k-space ing ie identical fo that ofthe clase spin-echo sequence ut in the RARE technique multiple (up to 256 ines are filled with each 90° Is flled with ench 5 RF excita ‘citation pulse, whereas in classic spin-echo imaging only one ine puse. The RARE technique can be upto 256 hes faster then the clas spin-echo sequence (12) RARE problems Sequential lines in kspace acquired ater each successive echo have successively decreased amplitudes, This ls very diferent from the classic pinvecho situation, in which every line in k-space (acquired atthe samme TE) has the same intensity. The eect to decrease the ef fective width of k-space (with effective width Kite to the numberof lines will in ntensity greater than background nove), Image contrasts Yariable and depends msinly on where each lines placed. The insert atthe lefts the view of kespaces seen en face, with te grayscale prppor- tional tothe intensity ofeach line of kspace. Te larger cental view shows in perspective the the RARE technique: ~ neering approximation, the signal is Effectively Zero for times greater than 3x72) The effect on the resolution ‘would be dramatic sar to that ofa lens with avery narow aperture (only 15 of 256 ines wide’ or less than Sthapen) The resolution willbe very Bootsand the image vl be cite ure (atleast along the phase-en- coding direcnon) "This tmaging technique is unigue in that the materal being imaged deter mines the resoltion,f we had cho” Sen'a material with» longer 2, the hummer of fines in espace that would have been filed before the signal ef fectvely vanished would have been greater and the resoltion would hove een Thee the ec ton sal good for long Tematerals (ee cerebrospinal id) sha quite poor forshort T2 materials (ep, muscle) ‘One very nice feature, however, is that the contest nthe age te the amount of 12 weighting) scontol Inble by the user. Sequentit fling from the center of espace gives short TF Tong TR type image Se. squenta filling from one eg of k- space gives a very long TE ong TR Eype image. Filing from intermediate Positions can yield images with inte mediate contrast. Agsiny if we assume 2 TE of 20 msec forthe fist echo and 2 kespace tht i 256 nes wide, and if we were to star ling kespace at ine 123 (ve lines before the center the contrast would ppromimate that of indicates negative and +Gp indicates negative spatial frequencies an image acquired with a TE of 100 tne, Simple movement ofthe ine from whic the star of spaces fies proces hat Henn tal Called scroling (1), allows te use to Choose any atitearyT2 weighing in the image without changing any Ee ing parameters ‘Bre important problem with this technique at least inthe casein tthich Kespaceiflled from some place near the center, isthe large Symmetric discontinuity i crentesin ihe espace intensity pattern The Situation is eminscent ofalene with 2 big blac gouge int and in foc just Ske tat ents pattern creates ort facts (consisting of sreaKing, ghost, ana strong background nese) that degrade the nage Mil and colleagues (15,16) de serbed wo modifications to this tech higue that overcame the problems with reoltion and image artifact ‘The ist was fo take fessaggres- sive stance on the numberof ies fied eth each 90 pulse which they fallen shat), Recognizing that ok. tany materia, collection of more than 16 ora lines sith each shot was Not well (either bocaus fie of Contrast or los of resolution), they proposed simiton the numberof Ties (our to 16) fled at each shot snd then use of several such se aqences (to 16 fl space: Con- erthe casein which tenes are fled seach shot which they would able amplitudes of the espace lines used in i the phase-encoding direction. call an echo train length of 16) and assume a TE of 20 msec (which they called the echo spacing). Steen shots would be needed to fil 256 ines of K-space, and each shot would acquire data with an echo at 20 msec, 40 msec, {60 msec, 80 msec. 260 msec, 280 sec, 300 msec, and 320 msec. They ‘would then collect and rearrange the data into 16 lines of kspace with a TE (0 20 msec, 16 lines with a TE of 40 see, another 16 lines with a TE of 60 msec, and $0 on until a final grouping ‘of 16 lines with a TE of 320 msec. if ‘we aisume a TR of 2 seconds (2,000 msec), the total image acquisition time woud bea quite respecable 32 see conds (ie, 16 shots at 3 seconds per shot) The second modification was to depart from the sequential approach toiling k-space and use instead a ‘more symmetric arrangement. Thus, for example, in the ease ofthe 16, shots used above, data from the frst acquisitions from each shot would be placed in the 16 lines at and near the center of k-space (in the ease ofa k- space that is 256 lines wide, the frst line of data from the frst shot would be put a line number 128 [which ‘would be done by acquiring that piece of data with no phase-encoding pulse applied]), the frst line of data from the second shot would be put at Jine number 129 (done by acquifing that piece of data with a small post- tive phase-encoding gradient-for Tera Rant eseesmnyeee etnigue Kapac elle wth data auied with several short cho teins aher han the singe ong pinot the RARE iechrique-In his example, we imagine that 1 echo tara als called sot), each containing th eee ee one SRI, as pation of kspsce fle by the fst ive echoes of each train, The center portion of spaces led ty ihe hea eh ee cain Done Se these te ete of kepace, each acquired wilh he same TE, The end el of ach tin wile place nae aa eg Diet unc egychan an tht oe ober sd: These il all haven cg wi ce the TE ft Bis essed see Sg ns tench side wil have lower amplitude than the soup of 1lines nthe center The tin echoofesch ai ees oe Fave sn me TE wns 20sec) wil be ute oil he net lines, symmetrically paced eight To cchsile ations Tats ave a i ower amplitude because of the longer TE. Data from successive echoes wl ll space ns sinter fashion wat ee a example, +1 G), the first line of data from the third shot would be put at line number 127 (accomplished by acquiring it with a small negative phase-encoding pulse—for example, =1G),and s0 on until ines 120-136 were filled. Each of these first lines ‘would have been acquired with a TE ‘of 20 msec. The second lines from each shot (those with a TE of 40 msec) would be placed symmetrically about the center, grouped with eight to one side (lines 137-124) and eight to the other side (limes 112-119). Proper placement is accomplished by using, appropriate phase-encoding gradi- ents during acquisition—for example, +4 G for the data to be placed at line 137, ~4G for the data to be placed at line 118, +5 G for the data to be placed at line 138, and so on. The pro- {ess continues in this symmetric fash- ion until all of k-space is filled (Fig 13). The large discontinuities that oc- cur when k-space is filled sequentially are averted, and attendant artifacts are reduced, As with the original rapid acquisi- tion with relaxation enhancement (RARE) technique, the lines in k-space can be scrolled to simulate any de= sired T2 weighting. ‘The use of limited echo train lengths (between four and 16) and symmetric filling of k-space removes ‘many of the limitations of the original RARE technique, with the result that image quality approaches that of the classic spin-echo technique but with ‘much greater speed. Even so, some Volume 195 + Number? pee limitations persist that have caused Some to question its use in some ap- plications. First, because of symmetric filling, the intensity of data at the edges of k-space is decreased compared with that in the center. Such shading is a commonly used “special effect” in photographic imagery, where itis Called apodization and serves to cre- ate a “softer” image with slightly re- duced resolution and smoother edges (ltis sometimes used by professionals 1n the motion picture industry, for example, for photographing older actors and actresses, because it can reduce the prominence of wrinkles.) Second, the decrease or change in. _ signal intensity from grouping to grouping (ie, from the group with a TE of 20 msec to the groups with a TE of 40 msec) is not smooth but rather is stepwise with finite discontinuities Although not as severe as in the ini tial sequential arrangement of the RARE technique, with the single large “gouge,” these discontinuities cause haze and artifacts in the image, much like scratches in a lens (or even streaks in a car's windshield) might. The regu- larity of the defects causes “ghost” images (the situation is exactly like the example of the grating discussed earlier, except the grating lines are farther apart and weaker, so the higher order are closer together and dimmer) that can sometimes be seen as dim overlapping images (3) but that more often manifest as a faint haze over the image. Both of these effects serve to lessen the contrast, sometimes reducing the conspicuity of lesions. Debate contin- tues over the importance of these and other effects on clinical utility and on the best areas of applicability of this, approach to fast image acquisition G19), MOTION AND RESPIRATORY COMPENSATION Rearrangement of the onder of lines in espace cn aso be used to mnt mize tome motion artifacts (20-25) Tagine MR image acquisition ofa stowly pesiodicaly moving abject such asa chest or abdomen with re: ratory motion Each tine of espace tullbe afeted by that motion tea {eater or lesser amount, depending On the degree and rate of mation as that ines acquired {lll not dwell on thenature of tha‘ effect, forts not important tous now, What sim Portantis that theres some eect and thatitis proportional tothe motion) Inthe casein which he spin-echo Prlse sequences sed, with enh fine St espace acquired ata regular rate (say fve lines per second, ora Th of 2100 se) and with espace filed in order fromane end tothe other, ke Space willbe modulated by the effect Of the periodic motion. Figure 14 Scheraticallythstats the sitation Ot respiratory motion at rate of 12 per minute, where have assumed, for purposes ofilusration, tha the motion simpy affects the ampltate Radiology + 309 ofeach ne of k-space, In this case, espace loos somewhat ike the cor. rugated surface ofa washboard, The important entre, forex purposen is the periadcty ofthe comagations Juste the case of a grt or stock: ing) placed just infront of our eyes,or 2 grating paced tthe lens (or Fou Fer) plane of camera, the periodic Corrugations act like a ifraction grat Srqancterate mip images (higher onder difactions) that ee copies of the original images. Since the cor ations have alwer peiodiity than Each line of espace, he higher orders overlap the orginal image Ghey ave not settered out ofthe FOV) and Spear as phone of the ogi tn age In the casein which only part of the image moves such asthe beating aorta or femoral artery, nuliple cops ies willbe seen above and below the trucimage of the vee difracted along the phase-encoding direction) In the case in which the whole object moves, such as with respiratory mo. ton, ghosts ofthe entre abject wil be seen difracted apove and below the trae image (rig 13) The ect factor that gives rise to the ghonts isthe periodicity of the modulation of kxpace, which makes itact ke an optic! groing The pe- tiodicity could be reduced, the ghosts ‘would vanish. The periodic pro Portional tothe rao of TR tothe pe Fod of the motion, a relationship that an beat least intuitively apprevated Byingpection of igure 14H is thet ratio that determines the number of Gyles ofthe "corrugation anti i the “corrugation” or grating spacing that determines how widely the dif fraction orders are scattered, as well as how intense they are. The number of “corrugations” can be calculated and is actually proportional to (PR NSAYT, where NSA ip the number of signals averaged and T's the period ol the motion With more eyes (achieved by either increasing TR, increasing NSA, or decreasing 1) the orders (“ghosts”) ave diated far. ther away and wall disappear tbe dif fracted out of aur FOV} when the pe Hodicity of the “corrugation” equals TR (or when TR x NSA ='T), Ths condition i called "pseudogating,” There isa beter, more reible way to minimize the eee of motion Bailes etal 21) proposed the Foor dering oflines of espace to remove that periodicity Instead of ust ling lines of kspace at» regulate frome tone ena tothe other, hey would choose where place the lines de- pending on the motion atthe time 3 bine wes being acquired Thus, for ox Figure 1, () "ive" tho wal ghosts due t low rs (0) Widely spaced ghosts due to relatively rapid femoral artery pulsations and pation. The relatively rapid repetitive mation Df the femoral artery causes "fine" modulation in kspace, which lead to large diffraction ingles ofthe higher order (and so wide separation ofthe ghosts). The slower epetlions of respiratory motion lead to widely spaced modulation in k-space and thar close spa (less, diffetion) ofthe ghosts. The major determinant of the spaving an srengt af the "ghose ‘nthe aio of TR tothe periodicity ofthe moving sich ample ifthe motion was strongand negative (sy end expiration) they wot pat that ines the bottom ot Espace Ifthe motion was strong and poaltive while aine was acquire fhatline woud be puta he top of espace. Lines acquired while the mo- tion was moderate would be put near themiidle of space, This remrange- ment would transform the period ‘modulation of espace into slow ‘monotonic modulation (ig 16), Which would completely ebmirate the peasity of higher-order ghost images: Since all the spatial frequen. Gies ofthe orginal image are stl ac. quired and used for reconstruction, the orignal image quality il mine tained only the ghosts ate lost. This technique, which was one ofthe fst demonstrations of rearranging ines of espace, demonstntes the enormous power and fet of k-space manip Fen that stig Yo inning SEGMENTED K-SPACE IMAGING Rearrangement of lines in k-space is an adequate way to compensate for slow, regular respiratory motion, but it does not work well for irregular or complex motion (eg, regular breath- ing, sighing) or for the very fast mo- tion seen with cardiac activity. The problem is that even ifthe lines are rearranged, all of k-space must still be filled. Ifthe sequence being used has a relatively long TR, then it will take a relatively long time to fill k-space (N x TR, where Nis the number of lines in k-space). With long image ac- quisition times the probability that patient motion remains nicely per- odie decreases, and so does image quality. One could use sequences with a very short TR (<7 msec, for example, as used in the fast low-angle shot technique)(27), which would en: able filling a 128-line k-space in less, than 1 second; this approach, how- ever, would still be toa slow for car diac imaging and can give relatively. poor contrast even for stationary o slowly moving structures. The limited contrast is due to the fact that, by it self, a sequence with very short TR has little TI or T2 weighting. Prepara- tory pulses that control contrast (eg, a 180" inversion pulse for TI contrast or a driven equilibrium sequence for T2 contrast) can be added just before the short TR sequence, but because of T1 and T2 decay these lose their effect even with acquisition times as short as 1 second. Rearrangement of lines in k-space so that the early acquired lines are put near the center of espace helps maintain contrast, but then relaxation decay decreases the amplitude at the edges of k-space, which compromises resolution. Atkinson and Edelman (28), Chien et al (29), and Edelman et al (30) pro- posed a means of overcoming some of these problems that involved use of a method that did not fill k-space all at ‘once but rather filled ita few lines at a time. If, for example, we filed eight lines ata time, then it would take 16 acquisitions (each acquiring eight lines) to fill a 128-line k-space. We have many choices as to where to put these lines, but the simplest is simply to fill every 16th line with each acqui- sition: Thus (as illustrated in Fig 14), li fl al Figure 16,” Respratory-ondered phase encoding, Kspace is modulated by the respiratory motion. The lines in kspace are rearranged so that lines where the modulation i nepatve” (ca, ines 4 8, and 9) are placed to the let of ke five" (eg lines 1,6 and 10) ate pleced to the: Figue 17 Segeonted kapace. ahi tration space iid eight ner ata te, andall eight line are acqned nearly sal taneously (ype nat So vec oo) Thetines ar speedo san espace hwo take 16auch eh aa tne neuons to Completely fils akin apace The a ‘eta ine onthe "ont of espace na {ate he positions of the net sh aust Ina ypiel cardiac appicazon each pup Gteigh would be ace ance por Rent beat atthe sme elatve ime afer te ORS comple, that a complet image cou be sequen jot een eats lines 1, 17, 33, 49, 65, 81, 97, and 113 would be filled first then lines 2, 18, | 34,50, 66, 82, 98, and 114 would be filled with the second acquisition, and 50 on until with the 16th acquisition, all the lines of k-space would be filed. Volume 195 + Number2 ace and lines where the modulation fps tof Kspace. Intermediate lines eg, ines, 2, fect i fo replace the peviodiallysnoduated the ghosts. The linear mod wil have minimal effet onthe image (3 most will cause sight edge enhancement that in practice will usually not be notiestle A particularly useful application of this technique is cardiac imaging, in which each acquisition is synchro- nized to the cardiac rhythm (ie, tig gered by the electrocardiogram 30. that each acquisition occurs once per heart beat, at the same point in the cardiac cycle). K-space would be filled in just 16 heart beats, and with a TR (of 6.2 msec, each acquisition would take just 50 msec (8 x 6.2), The acquic - sition time is short enough to “stop” heart motion, and the total time re- quired to fill ll of k-space (16 heart beats) is short enough that most people can hold their breath for the entire sequence. As with fast low- angle shot imaging, preparatory pulses are needed to set image con- ‘rast (eg, a 180° RF pulse for TI con- ‘rasi), but this pulse is delivered with each eight-line acquisition, and in the very short tirhe required for each ac- guisition there is no remarkable Tl or T2 decay. This technique completely elim- nates all motion artifact while pre- serving image quality and shows such ‘exquisite detail in the heart that itis even possible to acquire remarkably good images of the coronary arteries 1,32) attic eeecee ee KEYHOLE IMAGING While a present itisnot wide used "keyhole imaging” (33) = lever technigue for rapid imesing and nicely summarizes some of ae methods Ihave been talking about ‘The technique was designed for cne- ‘ype maging in which images ane acquired in rpid sequence my order to ilstate dynamic changes of some siructuresin the tnage field. ‘he usual problem with sch rapid mage ingis schievement of high speed, 00d contrast and high resolution. To achieve high resolution one needs 2 large k-space, butt takes time to fil kespace ith many lines Deering fhe sizeof espace decreases the tine required forthe study but loo de- Greases the resolution Use of very Shoet TRs can decrease the time aquired but may also decrease the con: fast One approach that Thave just discusses the segmenting of k-pace— filing all of espace but only par ta time: Another he keyhole technique, fils o pace once, bat then ups dates onlya smal cena part oF Space to create the cine acquisition ‘The method takes advantage of two pPhenomens, one physical and the ther psychological Fist, the method takes advantage of the fact that a very rapidly changing structare will nearly alvaysbe Blurred to some extent less acquisition is performed in eve time, which isefectively imposible There will always be some compos mise between resolution and speed. Second, hen we lok a fell with 2 stationary background anda change ing foreground, # is aot actualy nee essary 0 updatall the background to appreciate the changes, Furthersi the background has high resolution and overall contrasts maintained the eye will usally not notice that the changing srucfares have lower resolution. Thuy, ia detailed image is btoined initially (pethaps before the change tar by ng al of espace, and if subsequently only a relatively few lines inthe center of espace are acquired at rapid rate and used to replace just the previously acquired Central nes, the changes willbe seen and the entre image will appear to the viewer to have high resolution, By adjusting josthow many ines near the center of espace are acquired with the rapid sequence the operator Ean adjust the cosnpromise between Speed and tre recottion ‘The changes Thave been referring tocan ether be changes in contrast as would be seen alter gadolinium or other contrast agent was injected, Radiology + 311 * ag! ¥ be changes in position, such as might be seen with images of a moving joint. Most applications to date have involved imaging changes following admininstration of contrast material 4-36), SPIRAL IMAGING Allthe methods Ihave described so farutlzea rectangular k-space that is filed one line ats tne. This a direct consequence of he fact tat the phase ana frequency-encoding gradentsare Separated in ee. Fish the phase. encoding gradient ispalsedon to set a spatial requency (aline in kspace) and then the frequency encoctyy pulse comes on fil hat line, There Isno aw of nature, however, that specifies it must be this way. Al that ature (ora leat the MR radials) Fequires is that espace hele. No one rally cares too much about ex actly how it filled, for once its filed we will ust go ahead and per forma two-dimensional Fourier tans: form on kespace and ignore the de. ils about hove the data got tere The pulse sequence designer as con- siderable freedom in choosing when and how to apply the encoding pra ents, and some interesting tinge hap pen ifone decides to et the phsse fd frequeney-encoding gradients occur atthe same time Imagine fist thatthe two gradients were just turned on atthe same time tnd forthe same duration In that ase we would again fil nein F- Space, but that line would beat ahd would cut a diagonal through space (with every increment fa the frequency-encoding direction there would be an equal iherement the hase-encoding direction tracing + ine tha bisects the two anes, ie 45), This world be interesting and we could indeed al space wits collection of 45" kines, butt would be complicatedthey would have ta have tobe of unequal length for one thing, and there would beno paricu- Inradvantage Imagine next that we apply grad ents with sinusoidal amplitudes (No- hese sit written that the gradionts, snust have constant amplitaes They an just as well have amplitudes that vaty with de, and the most gener vatying amplitade isthe sine rave) Te that cas, it toms out, we wil ly circle in space, and the diameter of the ctce wil be proportional to the amplitude ofthe gratientif we Ss suse the amplitudes are equal (i they are not equal then we will ach ally flan elipne) By succesively 312 * Radiology varying the amplitudes, we would sll concentric ig of kespace a bulls ee pater). This would work and be tore interesting than fing keopace ‘th collection of lines at some {gl, but too woul be compli cated (for example the duration of the gradients wouid have fo vary 33 Wwe filed eles farther from the cen ter) and would not ofr many advan- tages imagine now (a did Macovsky (37 Moyer etl 33} and Shenberg and Macowsi 38) that we again tse grac dient with sinusoidal amplitudes at thistime we let the amplitudes in trease with ime fg 1), This time we will fi k-space witha spiral patern, ating atthe center (bath geadients small) nd spsaling farther and far, ther away a the apie continue ously increase. This netully does have some advantages paiclarly for very fas imaging For one thing we ould fe wished fll allofke Space in one spiral (by using gredient- echo pulses vith very shor fhe and TEs) More important, ovever,s the fact that even weil espace quickly, ee steain woul be put on the ciety generating the prion, forinstead of changing fromone value to another quickly as they might in fine ata time scheme, they need only amp up gently as te sng waves aradually build in amplitude Thisisno smal advantage and hes stimulated considerable interest nthe application ofthe spiral imaging ap: proach to cardiac, neurofuncional tnd other high-speed MR imaging, techniques ‘THREE-DIMENSIONAL FOURIER IMAGING “There i one important ference between optical imaging and ME im aging that Have ignored tought ths discussion’ Optical systems mage plane, whereas MR imaging systems Ga, and do, mage a volume ave avoided that diference by focusing attention onto one plane or section in the tree-dimensional volsne and by urposely disregarding how that sec ton came tobe Setecte. is now time to address, briely, how setions preselected. Ta the so-called two-dimensional Fourier transform technique a section is selected by applying the RF encts tion pulse (@ SF pulse inthe ease of spirvecho imaging a smalic angle pulse nthe case of gradient cho imaging) inthe presence of Bradiont™the sectionrseoct gradient hs gradien a small magnetic ela that i larger on one side or along one ani of the volume than the other) changes the resonant (e,precessing) frequency ofthe sping along that ais in direet proportion tothe strength of the gradient—higherat one side than the other. The RF excitation pulse, whieh is applied tothe entire volume, will excite only one plane thats or ented perpendicata othe gradient direction The location ofthe section fies how close to the center oc an tcige) is determined by the frequency of that RF puis. The width ofthe se. lected (or excited) section proper. tional othe gradient strength and the bandit ofthe RF pase, Nar- ree ection requ ra fan ‘Widths and large gradient, Limitations fon the narrowness of the bandwidth in practical systems, imposed in par by the desire for faster acquisitions Duta premium on the ability ofa System to deliver lange gradients Once a section is selecteh al the dis cussions on image quality and pace that we have just gone through apply. Th the theee-dimensional Fourier transform technique, the RF exo. tion pulse is agnin applied in the pres ence of a gradient, but the gradient strength and RE bandwith are cho- sen to excite a thick la (the slab can bbe as thick a the patient or smaller The slab is sudvided nto naval sections by applying phase-encoding gradients inthe section select direc: tion Justa the phase-encoding ga dhients deserted shove were applied torilllines ina two-dimensional kx space, these new phase-encoding gra dents are applied to fil prions tha three-dimensional space, space now isa "cube" made up of tinesin three dimensions one se aiong the frequency-encoding direction and the ther two along orthogonal phase- encoding directions (which we can Eallx andy) Image space is also a cube, and image data ae along three orthogonal directions Ge, height, width and depth). ast a with the two-dimensional tase, the image space is reconstructed by means of Fourier transfoim, but now a three- dimensional Fourier transform is used. This simply means that three arate Fourier transforms ate per formed, one after the other andcach works on the data transformed by the One before (excep thatthe fst ne fperateson the faw data) The fst toro transforms construct the data in a section just as before, with the same considerations regarding resolution, {quality artifacts and soon. The third transform erates the sections fil tl tI tl ez S1 ez TL. @z SL ox 1 Gx Gx CE $ mem ae sad phase I ‘encoding frequency encoding oblique linear scan (a) Figure 18.” Simultaneous application of phase-and frequency-encoding gradicnta a spiral scan (c) long an oblique line through k-space. lr both gradients are equal, the ling Sat a6" tone de, data wi 3B acl in kespace, A change in the amplitudes wath in kspace will be elipical, ¢ Sinusoidal and diced) in any orientation: the choices can be made to ft the viewing habits of the viewer andar idesiy optimized to tthe ortentation ofthe afatom aly iti worth epenting that regardless of whether wer dimensional tr three-dimensional techniques re "sed, the considerations of thage Suality and resolution in any sage Plane are those that have discussed Eetore CONCLUSION {A major strength of MR imaging Fesin the ability 'o manipulate espace. Unlike operator of any’ other metical imaging modality, the MR imaging operator can manipulate data to contol contrast reschution, FOV, and imaging time by simply changing the order with wich Espace filed one can contd art- facts or trade off contrast and resolu: Radiology + 313 tion. By shaping k-space, the operator can tailor the imaging sequence to the structure being imaged (for example, bby using a rectangular FOV with dif ferent resolutions in x and y to image the thoracic spine) to optimize both speed and resolution, K-space can be so easily cut and rearranged to fash- jon an image that it the operator is, almost like a tailor, cutting cloth to create a new suit As designers and radiologists gain experience with the methods and power of k-space manipulation, the capabilities and applications of MR imaging are being extended at an ever-increasing pace, There is no rea- son to expect that an ultimate limit has been reached. The zeal limits now seem to be the limits of the imagina- tion of those who build and use the systems, The purpose of this review is to serve a8 an aid to that imagination and to give structure and body to, what may sometimes be abstract no- tions. If we remember the close rela- tionship between a lens and k-space and then consider what might hap- pen ina simple camera, we may be able to give form to a hazy notion and help ease the path from thought to deed, APPENDIX: HOW K-SPACE GOT ITS NAME Consider a collection of hydrogen atoms, such as might be found ina bucket of water ora piece of fat, that happened to be placed in an MR im- aging system. The MR signal that would emanate from that collection in response to a pulse excitation (eg, by. 90° RF pulse) would be given by the equation § = e-"/T-cos(yBi}, where f = time, T2is the spin-spin relaxation time, y is the gyromagnetic ratio, and Bis the magnetic feld. If we note that Sis areal function and if we will con- sider only the real part of the follow ing equation, this can be written in ‘more general terms as S = 6-H! p-pet 8, When a gradient (6) is applied, the ‘magnetic field can be writen a5 B = B, + Gx, where use of bold charac- ters indicates that the variable isa vector with components along or thogonal projections, for example the Siysand 2 directions, In genera, G can be time varying, but will assume, for the purposes of this discussion, that itis constant (as isthe case with zmuch of MR imaging—spiral imagin being one notable exception). Ween then write S = eMC). Now, in the more interesting case in which we image a person and not justa bucket of fat or water, the mage netization will vary from point to point in the field, in part because of differences in the number of protons and in part because of differences in the relaxation times. In that case, the ‘equation for the total MR signal be- comes S= ff ferme mnie fixje" BPG dy where f(s the spatially varying Imagnetzaton dx a smal (iferen- ta) volume element in the patient and /ff indicates ntegrtion (summa: ton) over the volume o the patent The term ee" Et0 does not vary with location Ge, x) and sos Constant over the patient Ieean be replaced, for pumposes ofthis discs: sion, bya constants A, placed in iron of the integral signs The goncral equation forthe MR signal emanating froma patient after an RF pulse oxo tation Becomes 6) S=AL LS torer206 a, ‘The equation can be made even more compact by defining the relationship. = YGt, so that the general equation now becomes SAS Sf foermn te ‘This equation is identical in form and function to the equation that defines the Fourier transform of a fanction {(9), To illustrate, examine the usual form of the Fourier transform of some funetion of time f(t). This equation, ‘which is very common in electrical engineering work (40), is Stem as, where w indicates frequency and £ indicates time. Except for the fact that time (#) and frequency (w) in this equation are one-dimensional scalars, ‘while k and x in the previous equa- tion are three-dimensional vectors, the equations are identical. By comt- parison then, § represents the Fourier transform of the spatial magnetization in the patient f(x) and isa function of spatial frequency, and k can be un- derstood to describe that spatial fre- tency. the vector khas three components, which for the case of constant gradi- Flo) ents can be defined ask, = yGut, ky yGyt, and k, = yGzl. These three come ponents are in fact the coordinates that define the domain of S, or, as we usually cal it, the k-space, a Acknowledgments: | gratelly acknowledge the assitance of Michadl Cooper in ecaing, many of the excllent ustatons and Mary Grace Zethi for her help wi the Fouriet leansform images, References 1. Elelatin WA, Hetchion JM Johnson G, Redpai Sp wrp NM aging snd pplication te uman woke body ig ing Pay hed ak 96,25 81 86, ‘Mls SW, Hackney DE, Liter. Fst Spitecho imaging the bats an pe Magn Reson 1585 91-83 3, Ltr Eiaten 9, QuwaierE, Kresel ~ HY. is principe fat api cho ‘Magn Reson @ 109 918-304 Went FW Fastscon magnetic reso- ~ ance! panes nd appeatone Magn ewon Que 16820 § Twleg DB. The ltaecor formulation = ofthe nisi aging oes with applic tions in analysand Eyes of npn methods, Med Phys 198, 10si0-6 6 eters J, Chratoferien JO, Gelman ‘Nit smutaton using he Lapa ism. Mag Reson naing 19611 S37- \n 2. Roemer PB, Edelstein WA, Hayes CE, ~ Souzase, Atveer Okt the SR phased aay Magn Reson Mea 19m, 16198 25, 8 Bom, Wolf E Hatori tredacton In: Trineples of opten Sie Condon, England: Pergamon, 964 scnrh 9. Axel Doughtersy. Reduction of ais ingin 2D FE MRI by pseudoterig SE. Magn Reson Imaging 1987, 88 10, KucihareN, Kamo O, Umeda eta Appletions mone dimensional and two stintensional NMR a peuofiter by. tered time averaging] Mg Reson 908 Sse, VL Hennig] Nawenh 8 Paedburg iL RARE imaging fst imagtg method for dea! ‘MR Magn Reson Med 1985, 25-333" Hennig FiedburgH, Strobel B. Rapid sonoma pour oR mg raphy without contact agents | Comput ‘Ais Tomogr 1966 1035 38. 13, Hennig); Tedburg Cina phic tons and methodological developments of {he RARE technique Magn Reson imaging, tose, 6361395." 1M. Malern RV, Melk PS, Joka P, Higuchi N, Jolese FA. Phase-eneode order ands fee on conta ad statin angle shot RARE sequencen Med Phys 1; 16 100 1007 415. Melk PS, Matkern RY, Panych LP Joss FA Comparing the FAISE method with ‘conventional duthesho sequences] Magn Reson Imaging 191; 1315 336 16. Meth PS Jolre FA Mulhern RV._ Paral RF echo-planar imaging withthe FASE seth. I, centasteqivalenc wth spin- cho sequences Mag Reson Med 905 geass 17. Melk Ps folese FA, Muller RV._Partiat Echo planar maging with the FASE method 1 Expesimentel and theres sessment of arface Mags esos Med {ssa dee ot. 18 Melk Malkern RY, Magnetization tetris in mulisice RARE ae Suences Mage Renn Med 192; 2459- is ae! 1, a, Condit’, Anderson A, Znong J, Gore Factorsedencing contract in ae spin ‘cho MR imaging Magn Reson imaging 1393, 1097-51, ‘Axo L, Summers R, Kresel H, Charles C Respiratory effects in two-dimensional. Fourier Transform Imaging. Radiology 19 160-795-201. Bales DR, Gilderdsle Dy, Bydder GM, Cok lins AG, Flemin ON. Respiratory ordered phase encoding (ROPE): 2 method for acing respiratory motion atiacs in MR linaging J Comput Ast Teotogt 85, ie ase Lauzon ML, Rutt BK. Generalized ‘espace analysis and corecton of ation ‘fects in MR imaging. Maga Reson Med 195; 30s. Schieber ML, Lister J. Common asi facts encountered in thoraci magnetic. resonance imaging: cognition, desiv- tion, and solutions Top Nagn Reson Tag Ing 1903; 1-17, Wedeen V}, Wendt REd,Jerosch-Herold M. “Motional phase artifacts in Furie tran form MRI Magn Reson Med! 1965, 118d 1. Xiang QS, Henkelman RM, _Kespace de sexton for MR imaging of dynamic ob Jets Magn Reson Med 1985; 39422428 Volume 195 + Number 2 », ». ‘Young L_ Speci use sequences and fechnigues fe Stak DD, Bradley Woe Magnetic resonance maging St Lous No Mosby oe 106106 Haase A, Frohm J Matta, eta FLASH agg opi NOIR aging sing low fp angle pul) Mage Rese 58 Srassiee ‘Atkinson D, Edelman RR. Cnean phy ofthe hextin angle breath lg nth a segmented abo LAST sequence. Radilogy 191, 178357-200 Chien Atkinson D)Eleiman RR. Seat tet prove contrast in tba PLAST [is ered pe cai “space epmentton | Magn Reso ag Ing 991; Fea, cman i, Walle B, Sing A, Atkinson Dy Sunt. Segmented tore AStL method for breath ha MR aging ofthe tke with Rube contast Radic 190, isis oa Manning Wi, Li W, Boyle NG, Edelan Ri" Fatsuppresed bests magnetic ‘resonance cofanaryangegrphy. Cale fon 995 e790 Edelman RR Manning W)Peariman} Li W._ Human coronary ates: projecon angiogram reconstructed rom ores fol tordinennional MR images Rat 049 198 Ie a. Joes RA Harkseth O, Miler TB, Rick PA.Obsendal AN. Kspocesabstons a novel dynamictimaping tesinigue Magn Reson Med Toa 268600 2. ». . VaalsV, Brummer M, Dison We at. Keyhole method for accelerating nin fcontast gent aptake J Maga Ree Imaging B58 27a SpragineT. Simulation of pai and contrast dtontonsin keyhole maging Magn Reson Med 1954 3305-33 Kacharcay W, shop | Pees Keller MoGeorge 8, Detection of pita {ondenomascomparon of dyna oy hoe fast pi-echornenhanced and co ‘ventional conantenfonced MK inagiog, AIR 1994 Loa7l-or, Macovslt A Voianesc NMR imaging vith time-ayingpradints Magn Resen Med 1985, 238-48, = Meyer CH. u 8, Nibimara DG, Mi covak A. Fastoptl coronary ete png Magn Reson Med 990 23.5 5, Shenbergh Macovsi A. Appatons of time-varying arene in ong het eamahte imaging syste ed Tryst B60 Ie, PapoulsA. The Foie ntgral and ts applentons New York NY: Mia, we Radiology + 315