mri

REVIEW ARTICLE

13 SA JOURNAL OF RADIOLOGY • August 2004

AbstractMany different artefacts can occurduring magnetic resonance imaging(MRI), some affecting the diagnosticquality, while others may be confusedwith pathology. An artefact is a fea-ture appearing in an image that is notpresent in the original object. Arte-facts can be classified as patient-related, signal processing-dependentand hardware (machine)-related. Thisarticle presents an overview of MRIartefacts and possible rectifying meth-ods.

IntroductionArtefacts remain a problematic in

magnetic resonance imaging (MRI).Some affect the quality of the exami-nation, while others may be confused

with pathology.An artefact is a feature appearing

in an image that is not present in theoriginal object. Depending on theirorigin, artefacts are typically classifiedas patient-related, signal processing-dependent and hardware (machine)-related.

It is important to recognise theseartefacts and have a basic understand-ing of their origin, especially thosemimicking pathology. In this articleemphasis is placed on recognition ofartefacts and possible rectifying meth-ods.

Patient-relatedMR artefacts

Motion artefactsMotion is one of the most com-

mon artefacts in MR imaging, causingeither ghost images or diffuse imagenoise in the phase-encoding direction.The reason for mainly affecting datasampling in the phase-encodingdirection is the significant differencein the time of acquisition in the fre-quency- and phase-encoding direc-tions. Frequency-encoding samplingin all the rows of the matrix (128, 256or 512) takes place during a singleecho (milliseconds). Phase-encodedsampling takes several seconds, oreven minutes, owing to the collectionof all the k-space lines to enableFourier analysis. Major physiologicalmovements are of millisecond to sec-onds duration and thus too slow toaffect frequency-encoded sampling,but they have a pronounced effect in

the phase-encoding direction.Periodic movements such as car-

diac movement and blood vessel orCSF pulsation cause ghost images,while non-periodic movement causesdiffuse image noise (Fig. 1). Ghostimage intensity increases with ampli-tude of movement and the signalintensity from the moving tissue.

Several methods can be used toreduce motion artefacts, includingpatient immobilisation,1 cardiac andrespiratory gating,2 signal suppressionof the tissue causing the artefact,1

choosing the shorter dimension of thematrix as the phase-encoding direc-tion, view-ordering or phase-reorder-ing methods3 and swapping phase-and frequency-encoding directions1 tomove the artefact out of the field ofinterest.

FlowFlow can manifest as either altered

intravascular signal (flow enhance-ment or flow-related signal loss), orflow-related artefacts (ghost images orspatial misregistration).

Flow enhancement, also known asinflow effect, is caused by fully magne-tised protons entering the imaged slicewhile the stationary protons have notfully regained their magnetisation.

A short overview ofMRI artefacts

L J Erasmus MB ChB

D Hurter MB ChB

M Naudé MB ChB

H G Kritzinger MB ChB

Department of Diagnostic RadiologyUniversity of the Free State

Bloemfontein

S AchoMMedSc

Department of Medical Physics University of the Free State

Bloemfontein

Fig. 1. Motion artefact (T1 coronal study of lumbarvertebrae).

The fully magnetised protons yield ahigh signal in comparison with therest of the surroundings.

High velocity flow causes the pro-tons entering the image to be removedfrom it by the time the 180-degreepulse is administered. The effect isthat these protons do not contributeto the echo and are registered as a sig-nal void or flow-related signal loss(Fig. 2).

Spatial misregistration manifestsas displacement of an intravascularsignal owing to position encoding of avoxel in the phase direction precedingfrequency encoding by time TE/2.Theintensity of the artefact is dependenton the signal intensity from the vessel,and is less apparent with increased TE.

Metal artefactsMetal artefacts occur at interfaces

of tissues with different magnetic sus-ceptibilities, which cause local mag-netic fields to distort the externalmagnetic field. This distortionchanges the precession frequency inthe tissue leading to spatial mismap-ping of information. The degree ofdistortion depends on the type ofmetal (stainless steel having a greaterdistorting effect than titanium alloy),4

the type of interface (most strikingeffect at soft tissue-metal interfaces),pulse sequence and imaging parame-ters.

Metal artefacts are caused by exter-nal ferromagnetics such as cobalt containing make-up, or internal fer-romagnetics such as surgical clips,

spinal hardware and other ortho-paedic devices. Manifestation of theseartefacts is variable, including totalsignal loss, peripheral high signal andimage distortion (Figs 3 and 4).

Reduction of these artefacts can beattempted by orientating the long axisof an implant or device parallel to thelong axis of the external magneticfield, possible with mobile extremityimaging and an open magnet.5,6

Further methods used are choosingthe appropriate frequency encodingdirection, since metal artefacts aremost pronounced in this direction,4

using smaller voxel sizes,5 fast imagingsequences,6 increased readout band-width5,6 and avoiding gradient-echoimaging when metal is present.5 Atechnique called MARS (metal arte-fact reduction sequence) applies anadditional gradient, along the sliceselect gradient at the time the fre-quency encoding gradient is applied.5

Signal processing-dependent artefacts

The ways in which the data aresampled, processed and mapped outon the image matrix manifest theseartefacts.

Chemical shift artefactChemical shift artefact occurs at

the fat/water interface in the phaseencoding or section-select directions(Fig. 5). These artefacts arise due tothe difference in resonance of protonsas a result of their micromagneticenvironment. The protons of fat res-onate at a slightly lower frequencythan those of water. High fieldstrength magnets are particularly sus-ceptible to this artefact.7

Determination of the artefact canbe made by swapping the phase- andfrequency-encoding gradients andexamining the resultant shift (if any)of the tissues.

Partial volumePartial volume artefacts arise from

the size of the voxel over which thesignal is averaged. Objects smallerthan the voxel dimensions lose theiridentity, and loss of detail and spatialresolution occurs. Reduction of theseartifacts is accomplished by using asmaller pixel size and/or a smaller slicethickness.

REVIEW ARTICLE


Fig. 2. Flow-related signal loss in the carotid andbasillary arteries (T2 axial study of the brain).

Figs 3 and 4. Metal-related artefacts.

Fig. 5. Chemical shift artefact: bright and darkstreaks around the kidneys in an axial gradient-echo opposed-phase image (acknowledgementGE Health Care).

Wrap aroundThis artefact is a result of mismap-

ping of anatomy that lies outside thefield of view but within the slice vol-ume. The selected field of view issmaller than the size of the imagedobject. The anatomy is usually dis-placed to the opposite side of theimage (Figs 6 and 7). It can be causedby non-linear gradients or by under-sampling of the frequencies containedwithin the return signal.

The sampling rate must be twicethe maximal frequency that occurs inthe object (Nyquist sampling limit). Ifnot, the Fourier transform will assignvery low values to the frequency sig-nals greater than the Nyquist limit.These frequencies will then ‘wraparound’ to the opposite side of theimage, masquerading as low-frequen-cy signals. In the frequency encodedirection a filter can be applied to theacquired signal to eliminate frequen-cies greater than the Nyquist frequen-cy. In the phase encode direction, arte-facts can be reduced by an increasingnumber of phase encode steps(increased image time). For correc-tion, a larger field of view may be cho-sen.

Gibbs phenomenon (ring-ing artefact)

This is caused by the under-sam-pling of high spatial frequencies atsharp boundaries in the image.8,9 Lackof appropriate high-frequency com-ponents leads to an oscillation at asharp transition known as a ringingartefact. The artefact occurs near thesharp boundaries, where high con-trast transitions in the object occur. Itappears as multiple, regularly spacedparallel bands of alternating brightand dark signal that slowly fade with

distance (Fig. 8). Ringing artefacts aremore prominent in smaller digitalmatrix sizes.

Methods employed to correctGibbs artefact include filtering the K-space data prior to Fourier transform,increasing the matrix size for a givenfield of view, the Gegenbauer recon-struction and Bayesian approach.10-12

Machine/hard-ware-related

artefactsThis is a wide, still expanding sub-

ject and should be discussed separately.Only a few common artefacts thatshould be recognised are mentioned.When in doubt, however, a technicianshould be consulted.

Radiofrequency (RF) quad-rature13

RF detection circuit failure arisesfrom improper detector channeloperation. Fourier-transformed datadisplay a bright spot in the centre ofthe image. If one channel of the detec-tor has a higher gain than the other itwill result in object ghosting in theimage. This is the result of a hardwarefailure and must be addressed by aservice representative.

External magnetic field (B0)inhomogeneity14

B0 inhomogeneity leads to mis-mapping of tissues. Inhomogeneousexternal magnetic field causes eitherspatial, intensity, or both distortions.Intensity distortion occurs when thefield in a location is greater or less thanin the rest of the imaged object(Fig. 9). Spatial distortion resultsfrom long-range field gradients,

REVIEW ARTICLE


Figs 6 and 7. Wrap around artefacts.

Fig. 8. Gibbs artefact (T1 sagittal study of thebrain). Fig. 9. B0 inhomogeneity: intensity distortion

across a T1 axial study of the lumbar vertebrae.

which remain constant in the inho-mogeneous field.

Gradient field artefacts (B1 inhomogeneity)15

Magnetic field gradients are usedto spatially encode the location of sig-nals from excited protons within thevolume being imaged. The slice selectgradient defines the volume (slice).Phase- and frequency-encoding gra-dients provide the information in theother two dimensions. Any deviationin the gradient would be representedas a distortion.

As the distance increases from thecentre of the applied gradient, loss offield strength occurs at the periphery.Anatomical compression occurs andis especially pronounced on coronaland sagittal imaging.

When the phase-encoding gradi-ent is different, the width or height ofthe voxel are different, resulting in dis-tortion. Anatomical proportions arecompressed along one or the otheraxis. Square pixels (and voxels) shouldbe obtained.

Ideally the phase gradient shouldbe assigned to the smaller dimensionof the object and the frequency gradi-ent to the larger dimension. In prac-tice this is not always possible becauseof the necessity of displacing motionartefacts.

This may be corrected by reducingthe field of view, by lowering the gra-dient field strength or by decreasingthe frequency bandwidth of radio sig-nal. If correction is not achieved, thecause might be either a damaged gra-dient coil or an abnormal currentpassing through the gradient coil.

RF inhomogeneity15

Variation in intensity across theimage may be due to the failure of the

RF coil, non-uniform B1 field, non-uniform sensitivity of the receive onlycoil (spaces between wire in the coil,uneven distribution of wire), or pres-ence of non-ferromagnetic material inthe imaged object.

Asymmetrical brightness14

There is a uniform decrease in sig-nal intensity along the frequency-encoding axis. Signal drop-off is dueto filters that are too tight about thesignal band. Some of the signal gener-ated by the imaged section is, thereby,inappropriately rejected. A similarartefact may be caused by non-uni-formity in slice thickness.

RF noise14,15

RF pulses and precessional fre-quencies of MRI instruments occu-py the same frequency bandwidthas common sources such as TV,radio, fluorescent lights and com-puters.

Stray RF signals can cause vari-ous artefacts. Narrow-band noise isprojected perpendicular to the fre-quency-encoding direction. Broad-band noise disrupts the image overa much larger area. Appropriate siteplanning, proper installation andRF shielding (Faraday cage) elimi-nate stray RF interference.

Zero line and star artefacts. Abright linear signal in a dashed pat-tern that decreases in intensityacross the screen and can occur as aline or star pattern, depending onthe position of the patient in the‘phase-frequency space’.

Zero line and star artefacts aredue to system noise or any cause ofRF pollution within the room(Faraday cage). If this pattern per-sists, check for sources of systemnoise such as bad electronics or

alternating current line noise, looseconnections to surface coils, or anysource of RF pollution. If a star pat-tern is encountered, the manufac-turer should readjust the systemsoftware so that the image is movedoff the zero point.

Stars/zippers. Although lesscommon, stars/zippers are bandsthrough the image centre due to animperfect Faraday cage, with RFpollution in, but originating fromoutside, the cage.

Residual free induction decaystimulated echo also causesstars/zippers.

RF tip angle inhomogene-ity15

These are patchy areas ofincreased or decreased signal inten-sity. This artefact is produced byvariations in RF energy required totip protons 90 or 180 degrees with-in the selected slice volume.

Bounce point artefact15

Absence of signal from tissues ofa particular T1 value is a conse-quence of magnitude sensitivereconstruction in inversion recov-ery imaging. When the chosen T1equals 69% of the T1 value of aparticular tissue, a bounce pointartefact occurs.

Use phase-sensitive reconstruc-tion inversion recovery techniques.

Surface coil artefacts(attenuation of signal)15

Close to the surface coil the sig-nals are very strong resulting in avery intense image signal (Fig. 10).Further from the coil the signalstrength drops rapidly due to theattenuation with a loss of imagebrightness and significant shading

REVIEW ARTICLE


REVIEW ARTICLE

to the uniformity. Surface coil sen-sitivity intensifies problems relatedto RF attenuation and RF mis-matching.

Slice-to-slice interference14

Non-uniform RF energy receivedby adjacent slices during a multi-sliceacquisition is due to cross-excitationof adjacent slices with contrast loss inreconstructed images (Fig. 11).

To overcome these interferenceartefacts, include the acquisition oftwo independent sets of gappedmulti-slice images, subsequentlyreordered during display of the full

image set.

References1. Elster AD, Burdette JH. Questions & Answers in

Magnetic Resonance Imaging. 2nd ed. St. Louis:Mosby, 2001.

2. Huda W, Slone RM. Review of RadiologicPhysics. Philadelphia: Lippincott, Williams &Wilkins, 1995.

3. Nguyen TD, Ding G, Watts R, Wang Yi.Optimization of view ordering for motion arti-fact suppression. Magn Reson Imaging 2001; 19:951-957.

4. Suh JS, Jeong EK, Shin KH, et al. Minimizingartifacts caused by metallic implants at MRimaging: experimental and clinical studies. AmJ Roentgenol 1998; 171: 1207-1213.

5. Chang SD, Lee MJ, Munk PL, Janzen DL,MacKay A, Xiang QS. MRI of spinal hardware:comparison of conventional T1-weightedsequence with a new metal artifact reduction

sequence. Skeletal Radiol 2001; 30: 213-218.

6. Viano AM, Gronemeyer SA, Haliloglu M,Hoffer FA. Improved MR imaging for patientswith metallic implants. Magn Reson Imaging2000; 18: 287-295.

7. Smith RC, Lange RC, McCarthy SM. Chemicalshift artifact: dependence on shape and orienta-tion of the lipid-water interface. Radiology 1991;181: 225-229.

8. Elgavish RA, Twieg DB. Improved depiction ofsmall anatomic structures in MR images usingGaussian-weighted spirals and zero-filled inter-polation. Magn Reson Imaging 2003; 21: 103-112.

9. Lee SJ. An improved method for reduction oftruncation artifact in magnetic resonance imag-ing. Proceedings of SPIE: Applications of DigitalImage Processing 1998; 4360: 587-598.

10. Archibald R, Gelb A. A method to reduce theGibbs ringing artifact in MRI scans while keep-ing tissue boundary integrity. IEEE Trans MedImaging 2002; 21: 305-319.

11. Bakir T, Reeves SJ. A filter design method forminimizing ringing in a region of interest inMR spectroscopic images. IEEE Trans MedImaging 2000; 19: 585-600.

12. Schultz RR, Stevenson RL.A Bayesian approachto image expansion for improved definition.IEEE Trans Med Imaging 1994; 3: 233-242.

13. Holden M, Breeuwer MM, McLeish K, HawkesDJ,Deevil SF,Hill DL.Sources and correction ofhigher-order geometrical distortion for serialMR brain imaging. Proceedings of SPIE2001; 4322: 69-78.

14. Bushberg JT, Seibert JA, Leidholdt jun. EM,Boone JM. The Essential Physics of MedicalImaging. Baltimore: Williams and Wilkins, 1994.

15. Pusey E, Lufkin RB, Brown RK, et al. Magneticresonance imaging artifacts: mechanism andclinical significance. Radiographics 1986; 6: 891-911.

Fig. 10. Surface coil artefact: high signal at anteri-or thoracic wall adjacent to surface coil.

Fig. 11. Slice-to-slice interference (T1 axial studyof lumbar vertebrae).

mri

Documents