A primer on magnetic resonance imaging artifacts

Mri artifacts and how to overcome them

1. Primer on Commonly Occurring MRI Artifacts and How to Overcome Them Chikara Noda, PhD 1 Bharath Ambale Venkatesh, PhD 2 Jennifer D. Wagner, BS, RT 3 Yoko Kato, MD, PhD 1 Jason M. Ortman, RT 1 João A.C. Lima, MD, MBA 1

2. Author Affiliations: 1 Division of Cardiology, Johns Hopkins University School of Medicine, Baltimore, MD, USA 2 Division of Radiology, Johns Hopkins University School of Medicine, Baltimore, MD, USA 3 Canon Medical Research USA, Cleveland, OH, USA Corresponding author : J.A.C.L. (email: jlima @jhmi.edu ) Presented as an education exhibit at RSNA 2020 (HP137-ED-X). Supported by a National Institutes of Health grant (133032) to Johns Hopkins University Disclosures of conflicts of interest.—J.D.W. Employed by Canon Medical Research, USA. Part of ongoing educational collaborations with Canon Medical Systems Corporation.

3. Introduction Various artifacts can occur when acquiring MR images. MRI technologists need to know the causes of artifacts and how to avoid them in order to optimize clinical examinations. Radiologists need to be aware of these artifacts in order to perform accurate readings. This presentation describes: How artifacts relate to system conditions, patient physiology, or tissue characteristics How to identify artifacts and distinguish similar artifacts from each other How to mitigate artifacts in clinical practice An abbreviations table and artifact map have also been included to make the relationships among artifacts and the use of various terminology as clear as possible.

4. Learning Objectives Learn how to identify the artifacts presented here by taking note of the specific details associated with each example. Understand the most common causes of each artifact and how to mitigate its impact on image quality. Understand the pitfalls and trade-offs of each artifact reduction strategy.

5. Primer Abbreviation Generic Term Vendor Nomenclature Primer Abbreviation Generic Term Vendor Nomenclature B0 Main magnetic field PDW Proton-density–weighted B1 Radiofrequency field PI Parallel imaging ASSET, SENSE, SPEEDER BH Breath hold r R-factor or acceleration factor bSSFP Balanced steady-state free precession True SSFP, FIESTA, True FISP, balanced FFE RF Radiofrequency BW Bandwidth SAR Specific absorption rate CS Compressed sensing Compressed SENSE, Compressed SPEEDER SE Spin echo SE DSV Diameter spherical volume SNR Signal-to-noise ratio ESP Echo spacing ETS, ESP SS-FSE Single-shot FSE FASE, SS-FSE, HASTE ETL Echo train length ETL, TF T1-FFE RF-spoiled GRE FFE, FLASH, SPGR f0 Center frequency T1W T1-weighted (dark fluid) FS Fat saturation CHESS, Chem Sat, FS T2W T2-weighted (bright fluid) FSE Fast-spin echo FSE, TSE TOF Time of flight MRA2D/3D, TOF GRE Gradient-recalled echo FE, GRE UTE Ultrashort echo time mUTE, UTE NSA Number of signal averages NAQ, NEX, NSA VENC Velocity encoding Abbreviations Abbreviations used throughout this primer, their definition, and their correlation with common vendor-specific terms are shown here.

6. Motion Chemical Shift Dielectric effect / Standing wave Susceptibility Fat-water swapping Overview of Commonly Occurring Artifacts at Routine MRI Off-Resonance effects Patient Tissue heterogeneity System Flow / Pulsation Ghosting / Blurring Gibbs / Truncation Aliasing / Fold-Over / Wrap Section overlap / Cross-talk Hot lips & PI unfolding errors Streak Software Postprocessing Compressed Sensing Spike noise / Herringbone / Popcorn Zipper / RF Interference Moiré Fringes / Zebra stripes Hardware N/2 ghost Noise

7. Causes In-Plane Aliasing, Fold-Over, or Wrap RETURN TO INDEX Description  Phase Encoding  ◦ +90 ◦ -90 ◦ +180 ◦ -180 ◦  Total anatomy exposed to RF  Left Hip RF RF RF RF -90 ◦ +90 ◦ Left Hip -90 ◦ -90 ◦ +90 ◦ -90 ◦ Resulting Image with aliasing Right Hip A B C When aliasing occurs, anatomy outside of the FOV appears on the final image. On modern scanners, in-plane aliasing is typically only problematic in the phase direction. In the phase direction, signals within the FOV are encoded from –180 to +180 degrees (A). However, radiofrequency does not abruptly stop at the edges of the FOV. This means if anatomy exists outside of the phase FOV, it is also excited. However, since the same phase-encoding steps are simply repeated outside of the FOV, the position of any tissue outside of the FOV will match the phase encoding of value of areas within the FOV (C), and aliasing of those signals will occur (B).

8. Solutions If this is achieved by increasing FOV, the additional anatomy is visualized. If this is achieved by adding over-sampling, the extra encodings are acquired but not reconstructed. (B) Phase encode swapped from right to left and anterior to posterior. PRESAT PRESAT In-Plane Aliasing, Fold-Over, or Wrap RETURN TO INDEX (A) Phase-encode gradient is extended across all anatomy Option Trade-Off Unless you… Example Extend phase encoding across all signals by: Increasing FOV Resolution SNR matrix A Adding over-sampling Scan time SNR decrease number of averages A Swap phase or frequency direction changes direction of flow, respiration, and chemical shift artifacts B Add spatial presats (spatial saturation pulses) beyond phase FOV Scan time slightly SAR make other changes C Option Trade-Off Unless you… Example Extend phase encoding across all signals by: Increasing FOV A Adding over-sampling A Swap phase or frequency direction changes direction of flow, respiration, and chemical shift artifacts B Add spatial presats (spatial saturation pulses) beyond phase FOV make other changes C (C) Spatial presats are added These can suppress signals so wrap is less noticeable. Note.— = decrease, = increase.  

9. Cardiac short-axis (A) and long-axis (B) SSFP MR images . Wrap is evident on both data sets, with aliasing of moiré pattern (A, *) seen in addition to that of recognizable anatomy (A and B, arrowheads). After swapping phase and frequency encoding, aliasing is eliminated (C). In-Plane Aliasing, Fold-Over, or Wrap RETURN TO INDEX A B C * *  Phase Direction   Phase Direction   Phase Direction 

10. IMAGING SLAB Section Level Section-encoding Aliasing RETURN TO INDEX Three-dimensional (3D) T2* MR images targeted to the upper brain. If section-encoding oversampling is not applied, signals from just beyond the edges of the volume (dotted line) may be erroneously represented as existing within the opposite end of the volume. Related Artifact:

11. Flow Encode Aliasing RETURN TO INDEX Related Artifact: A B Phase images of main pulmonary artery with corresponding mean flow velocity curve above. (A) VENC = 100 cm/sec, (B) VENC = 150 cm/sec The dark area of image A (arrowhead) results from the aliasing of through-plane blood flow whose speed exceeded the selected VENC. Velocity analysis using this image then fails to determine the peak velocity (oval). By repeating the sequence with a wider VENC (B), it is possible to accurately depict the maximum flow velocity on both the image and associated chart.

12. Description Causes Gibbs or Truncation A  Phase Direction  RETURN TO INDEX (A) Sagittal T2W cervical MR image with FOV = 240 mm, matrix (frequency-encoding direction [f] × phase-encoding direction [p]) = 272 × 256. Gibbs (arrowheads) is common in the spine and may interfere with the depiction of spinal cord contusions or mimic a syrinx. Out-of-phase (B) and in-phase (C) axial T1W abdominal GRE MR images with truncation artifacts (arrowheads) along the margins of organs. Small lesions along the lateral liver and spleen can be overlooked because of this artifact. Gibbs results from the Fourier transform process used to reconstruct the MR signal into images. When strong signals change suddenly in a stepwise manner, they can be truncated by the Fourier process and thus inaccurately approximated in the final image. Alternating stripes at high-contrast boundaries. Gibbs can occur in any direction, but it is most com mon along the phase-encoding direction since this direction typically employs a lower matrix in the interest of time savings. B  Phase Direction  C

13. A B C Gibbs or Truncation Option Trade-Off Example Decrease pixel size by: Decreasing field of view SNR Not shown Increasing matrix Scan time (phase-encoding [PE] matrix); SNR A Apply raw data filter Image blur B Decrease echo train length Scan time C Decrease bandwidth Chemical shift; Sensitivity to motion Not shown Option Trade-Off Example Decrease pixel size by: Decreasing field of view Not shown Increasing matrix A Apply raw data filter B Decrease echo train length C Decrease bandwidth Not shown Solutions RETURN TO INDEX Baseline Axial fluid-attenuated inversion-recovery (FLAIR) MR images of the brain. FOV = 220 mm and matrix (f × p) = 256 × 128 in baseline. In this situation, Gibbs (arrowheads) may complicate the assessment of subtle cortical abnormalities, such as focal cortical hypoplasia.

14. Description When acquiring sections with varied angles within the same acquisition (eg, multi-slab multi-angle sequencing for the intervertebral disks), the signal intensity from overlapping areas is diminished (ovals). Section Overlap or Cross-talk RETURN TO INDEX

15. Section Overlap or Cross-talk Causes Solutions Avoid overlapping sections (C) A B C D Interference is lessened Use interleave method for section acquisition. This increases the time between adjacent section encodes, thus allowing spins to relax before they are excited again (D) RETURN TO INDEX If overlapping sections ( circles in A and B ) are acquired at the same time, neighboring sections contain spins that are already saturated , leading to diminished signal intensity in shared areas

16. Causes Unique to parallel imaging, these artifacts occur when the acquired FOV or calibration data are smaller than the imaged object, with artifact location depending on R-factor . This can also be seen at 3D imaging when section encode acceleration is used with insufficient section coverage. Hot Lips and Parallel Imaging Unfolding Errors (A) Axial T1W MR image of the liver with FOV (f x p) = 400 x 360 mm, R-factor = 2. Adipose tissue has been incorrectly unfolded into the abdominal cavity (arrowheads). Image (B) shows a sagittal 3D T1 magnetization-prepared rapid gradient echo (MPRAGE) with a pseudo lesion seen on the axial MPR. Review of a coronal multiplanar reformation (C) at the same level reveals that the lesion is actually a section-encoding unfolding artifact caused by the ear. C RETURN TO INDEX HOT LIPS A Description These artifacts manifest as signal or noise that overlays a somewhat centralized location on the image. This is different from normal aliasing, which is found along the edges of the volume. B

17. A B C D Unfolding artifact on FOV (f x p) = 400 x 360 mm (A) is eliminated when FOV is increased to 450 x 450 mm (B). A similar situation is seen in image (C), where too small of an FOV (180 x 180 mm) with too much acceleration (R=2) has resulted in a midline band of noise. With increased FOV (220 x 220 mm), the artifact is eliminated (D). Hot Lips and Parallel Imaging Unfolding Errors Option Trade-Off Example Expand phase FOV Resolution, SNR B, D Reduce R-factor (or increase autocalibration signal) Scan time B in Next slide Reacquire sensitivity/calibration data Total exam time due to repeat Not shown Option Trade-Off Example Expand phase FOV B, D Reduce R-factor (or increase autocalibration signal) B in Next slide Reacquire sensitivity/calibration data Not shown Solutions RETURN TO INDEX

18. Cardiac short-axis MR images obtained with modified look-locker inversion recovery (MOLLI) sequence. (A) was performed with R-factor = 3, which resulted in unfolding artifact (arrowheads) . Th is artifact jeopardizes quantification of the myocardial values on the resulting T1 map (arrows) . Therefore, a repeat image was obtained (B) with R-factor = 2. With this change, the artifact is no longer problematic (C). A B Hot Lips and Parallel Imaging Unfolding Errors RETURN TO INDEX

19. A B C * * Causes Description Effects of regularization in sparse sampling are shown on FS PDW coronal MR images of the knee. With over-regularization, fine detail is lost (A, *), and under-regularization, aliasing, and noise are evident (C, arrowheads). Thus, a balance of regularization must be found (B). Compressed Sensing Artifacts If parameters are not properly optimized, blurring of fine structures or textured noise may be observed on protocols that use sparse sampling as a method of acceleration. RETURN TO INDEX When compressed sensing is applied, if k-space has insufficient phase encodes to support a given acceleration factor, blur results. Additionally, regularization, which is used to threshold out noise and aliases, will also blur the image if set too high. If set too low, excessive noise may be seen.

20. Solutions Compressed Sensing (Optimizing Encodings or Acceleration) Increase PE matrix or add oversampling to increase the number of encodings in k-space. If averaging is not specifically needed, convert NSA to oversampling, as oversampling supports random sparsity more effectively. Decrease acceleration factor. PD FS MR images of the knee obtained with compressed sensing (scan duration = ~45 seconds). Image (A) is filling k-space with 250 encodings and accelerated by a factor of 4. Image (B) is filling k-space with 375 encodings and is accelerated by a factor of 3.3. The ratio of encodings to acceleration factor is higher in (B), resulting in superior image quality. Coronal T2W MR images through the hippocampus with 2 NSA (C) and 2 oversampling (D). Note that even though both sequences have exactly the same matrices (320 x 320) and scan time, cortical delineations are slightly clearer on image (D). A B C D C

21. Solutions Compressed Sensing (Optimizing Regularization) C Set regularization in-line with vendor-specified recommendations: exceeding recommendations can cause blur, and dropping below the recommendation can result in the retention of aliasing, which will appear as textured noise, in the final image. Sagittal PD MR images of the knee acquired with sparse sampling. Images (A) and (C) have been highly over-regularized and demonstrate fine linear patterns (ovals) as well as striations in the anterior aspect of the ACL (C, arrowheads). These artifacts are not created when normal regularization values are applied to (B) and (D). A C B D RETURN TO INDEX

22. Streak RETURN TO INDEX (A) Standard FSE MR demonstrating motion that is diminished (C) after being reacquired by filling k-space radially (B). Note that small streaks (D, arrowheads) are visible along some interfaces in the final image. A B C D Description High signal intensity is seen on the edges of the reconstructed MR image. The pattern of this artifact usually occurs in a diagonal direction that mimics the radial sampling pattern. Causes Radial scans collect data while rotating through the center of k-space. Since this acquisition method also rotates the phase-encoding direction at the same time, motion artifacts such as aliasing, chemical shift, and motion are radially distributed. This is the cause of streak artifacts.

23. Solutions Streak A B Axial UTE contrast-enhanced MR images of the lung i n a patient with a history of COVID-19. Image (A) was acquired with a 20-second BH using 5 220 trajectories. Image (B) was performed with 44 730 trajectories with respiratory gating across 5 minutes . Streak artifacts (arrowheads) interfere with evaluation of the lung parenchyma. Reduction of streak artifact (as well as increased clarity) is appreciable. UTE cardiac short axis. (C) FOV = 320 mm, (D) FOV = 450 mm. Streak artifacts that interfere with visualization of the mediastinum (C, arrowheads) are reduced by increasing FOV (D). However, it should be noted that increasing the FOV does not change the streak artifacts around the edge of tissues (D, arrowheads). The impact of streak artifacts can be reduced by increasing the number of trajectories used for acquisition. This will result in finer streaking, and with high enough trajectory values, streaks may become nearly imperceptible. However, increasing trajectories will increase scan time. Since streak artifact is often exacerbated by radially encoded aliasing, increasing FOV such that all anatomy is properly encoded can also reduce the impact of this artifact. RETURN TO INDEX C D

24. C D A B Filtering or Reconstruction Traditional k-space filtering often causes blur on images, as is seen in the axial T2W MR image of the lumbar spine in (A); the texture of bone and muscle appears more natural when the filter level is reduced (B). M odern artificial intelligence (AI)–based denoising techniques often perform better (C) but can still result in blur if applied too aggressively, especially if recommended settings are vastly exceeded (D). Description Unnaturally smooth appearance, plasticized look, smearing of detail, or inhomogeneous distribution of signal intensity across anatomy. Causes Excessive use of filters or incorrect application of intensity correction. RETURN TO INDEX

25. Solutions Axial postcontrast T 1W SE MR images of the brain with a denoising algorithm (A) compared with a k-space filter (B). While both SNR and contrast-to-noise ratio (CNR) are elevated in (A), some small hollows are not clearly visualized (oval). In addition, Gibbs ringing is emphasized (arrowhead). On the other hand, (B) shows less Gibbs artifacts and better displays fine details but is a bit noisy. Filtering or Reconstruction RETURN TO INDEX A B Familiarize yourself with the postprocessing options of your vendor. Image feel can vary greatly depending the postprocessing that is applied.

26. Phased-array coil sensitivity is high at the coil itself, but it weakens for anatomy deep within the body; true sensitivity is limited to the radius of each individual element (B). This phenomenon leads to bright subcutaneous fat and darkened inner structures (A). However, the phenomenon can be overcome if proper surface coil intensity correction is applied, as is evidenced on this properly reconstructed in-phase Dixon T1 GRE image through the liver (C). A C Filtering or Reconstruction Phased Array element radius area of low sensitivity RETURN TO INDEX B Solutions Ensure proper surface coil intensity correction is activated, especially on imaging targeted to thicker regions such as the head and trunk.

27. A C B Description Causes Zipper or RF Interference RETURN TO INDEX Dis played as a high-signal pattern that runs parallel to the phase-encoding direction. Zipper artifacts arise when electromagnetic interference reaches the scanner. This can occur in MR environments where there is an open scan- room door, compromised RF shield, unapproved or malfunctioning equipment, or RF coil connection failure. (B) Interference from injector. Ancillary devices must be designed for the MR environment and properly installed and maintained, or they can be the source of unwanted frequencies. (A) Strong zipper from open door . The scan-room door must be fully sealed or external frequencies may be detected by the receiver coils. (C) Interference from lighting. Fluorescent fixtures produce light by discharging in low-pressure gas. The process emits a small but perceivable signal.

28. Close the MRI room door completely. Turn off external electrical equipment in the MRI room or remove it from the imaging suite. Recheck the RF coil connection. If problems persist despite verifying the above item s, contact the manufacturer’s service representative. Solutions 　 Door open 　　 Door closed Zipper or RF Interference RETURN TO INDEX Sagittal T1W MR images of the lumbar spine with zipper artifact (arrowheads) overlaying anatomy of interest.  Phase Direction 

29. Annefact (A) Sagittal T2W MR images of the thoracic spine. Artifact runs head to foot but is not from radiofrequency; it is annefact. Annefact (arrowheads) results from the capture of frequencies outside of the homogeneous magnetic field. Preventive measures include limiting the number of receive elements (C) and properly centering the patient to the magnetic field. Area of magnetic homogeneity 1 4 3 2 Active Elements 4 3 2 1 Active Elements Area of magnetic homogeneity  Phase Direction  RETURN TO INDEX A C B Similar to Zipper:

30. Description Spike Noise, Herringbone, and Popcorn This artifact manifests as high signals (arrowheads) across the entire reconstructed MR image, often in a lattice or diamond pattern. The artifact usually run s in an oblique direction. B Spike artifact most often manifests on echo-planar sequencing owing to the stress that echo-planar imaging (EPI) places on system components. In (A) and (B), we see two separate occurrences captured on standard clinical DWI ( b value = 1000) A RETURN TO INDEX

31. Causes Spike Noise, Herringbone, and Popcorn A C spike K-space center After Fourier transform, herringbone pattern runs 90 degrees perpendicular to line drawn between spike and k-space center. B (A) Axial T1W FSE MR image of the thigh demonstrates a herringbone pattern. Actual k-space for image is shown at left (B). Note the noise spike outside of the center of k-space; the orientation and distance that this signal has in relation to the center of k-space will govern both the angulation and width of the herringbone stripes (C). A potential difference is generated between system parts owing to vibration from the gradient coil. The resulting discharge creates a spike in k-space, which is similar to a peak in the raw data signal . This is then manifested as periodic artifact on the reconstructed images. RETURN TO INDEX

32. Solutions Spike Noise, Herringbone, and Popcorn Unplug and replug coil to ensure firm connection. If reconnecting the coil does not improve the images, ensure that the humidity and lighting both meet specifications. If problems persist despite verifying the coil connection and environmental conditions, contact the manufacturer’s service representative. A B C T1W MR images obtained through pig heart phantom before (A) and after (B) coil reconnection. (C) Example of coil connection RETURN TO INDEX

33. Free-Induction Decay * RETURN TO INDEX Similar to herringbone: (A) Coronal short inversion-time inversion-recovery (STIR) MRI of the pelvis, (B) s agittal T2 FS MRI of the brain utilizing 3D FSE with VFA. Although similar to herringbone, FID artifacts can be differentiated because the lines are often wavy (arrows), whereas herringbone is straight. Also, FID artifacts stop abruptly in some places (*), whereas herringbone will carry across the entire image. Causes Solutions A B FID can sometimes be resolved by applying full 90° or 180° RF pulses and increasing echo time (TE) or echo spacing. Additionally, use of ≥2 averages will almost always eliminate the artifact. In theory, spin-echoes utilize a 90° or 180° pulse to excite and refocus spins. In practice, however, if some spins are not fully exposed to both pulses, errant signal results and manifests as a free-induction decay (FID) artifact. This is most common when modified refocusing pulse schemes (eg, variable flip angle [VFA]) are used and in areas where there is localized tissue inhomogeneity.

34. N/2 Ghost A B Description RETURN TO INDEX Causes N/2 occurs owing to eddy currents, incomplete gradient magnetic fields, magnetic field nonuniformity, and odd-numbered and even-numbered echo timing imbalances. Phase encode replication of tissue with echo-planar sequencing. Axial DWI of the brain, (A) b = 1000, (B) ADC. Acquisition angle is shown in (C). Excessive N/2 ghosting (arrowheads) renders both the DWI and ADC images nondiagnostic for much of the brain.

35. A B r = 2.0 r = 3.0 r = 3.0 C Solutions N/2 Ghost Minimize eddy currents by scanning with a plane perpendicular to B0 (don’t tilt or rotate the plane) and placing target close to isocenter Shim appropriately Apply FS – unsaturated fat often creates ghosts on echo-planar sequencing Minimize overlap of N/2 ghost and anatomy by increasing phase FOV or decreasing R-factor (these measures will also decrease resolution and increase distortion) RETURN TO INDEX N/2 ghosting of unsaturated scalp on SE-EPI acquired without FS (A and B). Distribution of ghosts is affected by R-factor, with increasing values (B) bringing the ghosts closer together. Note the resolution of the artifact when the sequence is repeated with FS enabled (C).

36. Description Moiré Fringes or Zebra Stripes Moiré fringes are curved bands that alternate with increasing frequency in areas of very low field homogeneity (eg, periphery of field). These are commonly seen on gradient- based acquisitions. A B (A) Coronal T1W GRE MR image of the chest with insufficient phase oversampling to prevent the signal from the arms from wrapping into the anatomy. However, since the wrapped signals originate from the periphery of the field, the wrapped signals (arrowheads) resemble moiré more than normal anatomy. (B) Cardiac short-axis SSFP MR image. SSFP images are specifically susceptible to Moiré along the periphery of larger FOVs. RETURN TO INDEX  Phase Direction 

37. Solutions Coronal T1W FS MR images of the abdomen obtained with GRE and full FOV. Moiré is seen along periphery of the field (arrowheads, A). When FOV is decreased, insufficient anti-aliasing protection exists to prevent the upper arms from wrapping into the anatomy (arrowhead, B). If coil-based parallel imaging is applied, the artifacts unfold even further into the anatomy of interest (arrowheads, C). Moiré Fringes or Zebra Stripes To avoid Moiré, set the imaging area as close to isocenter as possible. Also, avoid imaging anatomy along the edges of the FOV. When Moiré appears midline as the result of fold-over artifact, it can be lessened by adjusting the FOV or acceleration factors. A A B C RETURN TO INDEX

38. Description Causes Noise Noise typically appears as a textured pattern that is distributed evenly across the image . When sufficient noise is present, reduction of details and contrast may also be perceived, as the noise layer clouds these critical features. RETURN TO INDEX Image noise originates from both the MR environment and patient tissue. When noise is out of balance with signal (ie, low SNR), it can be problematic. This typically arises when parameters are too aggressive (eg, scanning too fast, setting resolution too high) or coils are improperly selected or set up. A B C A B C (B) Coronal T2*W FS 3D MR image of the wrist. Voxel size = 0.44 x 0.44 x 0.7 mm. Fractures are observed (arrowhead), but noise makes evaluation difficult. (C) Coronal 3D T1-fast field echo (FFE) MR image of the chest with BH. The back of the chest is noisy (oval) because of a failure to activate the posterior receive coils. (A) Coronal T2W FSE MR image of the brain obtained in 38 seconds. Insufficient time was spent encoding the signal to support this high of a resolution.

39. A B C Baseline How to Increase SNR C D Option Trade-Off Example Increase FOV Resolution, Voxel size Not shown Increase section thickness (slab in 3D) Resolution, Voxel size, Partial volume effect (oval) A Decrease matrices Resolution, Scan time (PE Matrix) B Increase NSA or oversampling Scan time Not shown Apply denoising filter Applying too high of a factor will blur the image C Option Trade-Off Example Increase FOV Not shown Increase section thickness (slab in 3D) A Decrease matrices B Increase NSA or oversampling Not shown Apply denoising filter Applying too high of a factor will blur the image C Solutions RETURN TO INDEX

40. Description Causes Axial dataset obtained through calf without contrast agent by T1W 2-point Dixon . ( A) In-phase, ( B) out-of-phase, (C) w ater image, ( D ) fat image. Swapped fat and water signals are seen in (C) and (D). Note that this correlates with a less obvious edge artifact (*) in the same area of the opposed-phase image . Fat-Water Swapping A D C B * * * RETURN TO INDEX Images obtained by using the Dixon method may swap the intensities of fat and water after reconstruction. If the out-of-phase image is also generated, subtle line artifacts may be visible. In Dixon-based techniques specifically, magnetic field inhomogeneity or phase errors in the sampled area can cause iterative calculation errors, which may result in a false determination of voxel contents.

41. Fat-Water Swapping A B Solutions x-axis ; y-axis ; z-axis DSV gantry C patient couch RETURN TO INDEX Acquire a shim. Adjust and repeat if necessary. Move target region closer to isocenter. Ensure coils are properly placed and SNR is sufficient: if SNR is degraded, the water-fat separation method may contain errors. Reduce the empty space within the scanning area. For example, use something like a liquid fluorocarbon pad to fill the empty space. Axial T1W 2-point Dixon MR images through the liver. (A) Water image. (B) Fat image. Major fat-water swap is seen (arrowheads) in right liver lobe. Ensure target is close to isocenter in all axes (C). For examinations through the trunk, only the z-axis can be easily changed. However, for inspection of orthopedic areas such as limbs, anatomy can often be brought close to DSV in all three axes with creative positioning.

42.  Frequency Direction  B C A  Frequency Direction  Description Causes Chemical Shift In areas where tissue containing fat borders a source of water signal (eg, aqueous humor, cerebrospinal fluid [CSF], etc.), an image shift occurs along the frequency-encoding direction, and white or black borders are observed at the tissue interfaces. (A) Axial T2W MRI of the kidney with chemical shift artifact seen on either side of renal cortex. Sagittal T2W cervical i n-phase (B) and water (C) MR images from the Dixon dataset. Note that chemical shift (arrowheads) is completely removed when signals from fat are suppressed on (C). RETURN TO INDEX Protons of different molecules precess at different frequencies; water protons rotate slightly faster (3.5 ppm) than fat protons. Therefore, the fat and water components of a voxel are encoded at different locations along the frequency direction.

43. Chemical Shift Option Trade-Off Example Increase bandwidth SNR A Increase frequency matrix SNR B Use fat saturation # of sections that can be acquired C Swap phase or frequency direction to shift artifact appearance to different side of structure (this does not reduce the artifact itself; it simply changes its location) May impact phase or flow artifact distribution Not shown Option Trade-Off Example Increase bandwidth A Increase frequency matrix B Use fat saturation C Swap phase or frequency direction to shift artifact appearance to different side of structure (this does not reduce the artifact itself; it simply changes its location) May impact phase or flow artifact distribution Not shown Solutions RETURN TO INDEX Axial T2W MRI of the orbit at 3 T. 0.6 mm 2 pixel and bandwidth = 140 Hz/pixel in baseline. Despite the small pixel size, chemical shift (arrowhead) is distracting owing to low BW. A B Baseline  Frequency Direction  C

44. (A) Axial T2W MRI of the cervical spine with narrow bandwidth ( 195 Hz/pixel) and obvious chemical shift along lateral canal (arrowheads). After increasing BW to 390 Hz/pixel (B), the artifacts are reduced; however, SNR is also lower ed. Chemical Shift Artifact in the Spine Changing the gradient polarity can flip the location of chemical shift. In spines, the frequency gradient should be oriented to place the black aspect of chemical shift posterior to the cauda equina (D) rather than against the vertebral body where it exaggerates the thickness of the posterior longitudinal ligament (C). B A  Frequency direction  Frequency Anterior  Posterior Frequency Posterior  Anterior C D RETURN TO INDEX Focus on:

45. Description Causes Off-Resonance Bands of signal loss (arrowheads) occur in areas of increased B0 nonuniformity, such as along the boundaries of dissimilar tissues, air and tissue, and along the periphery of the magnetic field. RETURN TO INDEX Cardiac cine using bSSFP sequencing in short-axis (A) and two-chamber (B) views. Banding artifacts (bands) overlap on the anterior wall, making it difficult to trace the contour of the myocardium during functional analysis. A B Balanced SSFP sequence is particularly sensitive to the effects of off-resonance due to B0 nonuniformity that causes phase shift and phase accumulation during acquisition. This sensitivity to phase error causes banding artifacts in areas where B0 nonuniformity has increased.

46. Off-Resonance Cardiac 4-chamber view with banding artifacts visible across the heart (arrowheads, A). These are reduced after adjusting the shimming (B). Sagittal PDW FS FSE MR images acquired with patient’s forearm by patient’s side. Image (C) shows off-resonance in FS due to exceeding usable FOV of scanner. In (D), the arm is brought closer to isocenter and the artifact is resolved. A B Solutions Minimize TR for bSSFP (for the shortest TR, you might have to sacrifice spatial resolution). Readjust the shimming and scan again. Improving shimming can mitigate the appearance of banding artifacts. Ensure target region is as close to isocenter as possible. RETURN TO INDEX C D

47. Mitigating Off-Resonance Artifact at bSSFP Imaging Cardiac 4-chamber view with different center frequency offset : (A) base image at 0 ppm; (B) -0.5 ppm shift; (C) -1.0 ppm shift; (D) -1.5 ppm shift. In this case, banding artifacts were shifted outside of the volume of interest by using applying the -1.0 ppm f0 offset. A B C D Solutions The location where the banding artifact appears can be shifted by shifting the center frequency (f0). While this does not directly minimize the artifact, it can move it outside of the area of interest. f0 offset = 0 f0 offset = -1.0 f0 offset = -1.5 - + - + f0 offset = -0.5 - + - + RETURN TO INDEX Focus on: Water Fat Water Fat Water Fat Water Fat

48. Description Causes A B Dielectric Effect, Standing Wave, and B1 (A) T2W FSE MRI and (B) T1W fast low-angle shot (FLASH) MRI in a patient with multiple liver cysts. This MRI was performed at 3 T. Note that the dielectric effect (oval) is more prominent on the FSE-based sequence than it is on the FLASH, which is a GRE-based sequence. (C) Sagittal T2W with SS-FSE at 3-T MRI. Dielectric effect (oval) limits evaluation of potential placenta previa adhesion in this pregnant patient. RETURN TO INDEX Images have uneven intensity, often with decreased intensity or focal signal loss near the center of the image. This is especially problematic at higher field strengths (≥3T) and with FSE-based techniques. If anatomic diameter is similar to RF wavelength, a standing wave may form. This can cause interference in the RF distribution. It is difficult to predict when exactly this will occur, but ascites, pregnancy, and obesity can all increase its likelihood. C

49. A B C Solutions Use dielectric pads. When this phenomenon is more likely to occur (ascites, obesity, pregnancy, etc.), consider scheduling the patient on a lower field system. Triage patients to newer high-field scanners (recent 3-T MRI systems are equipped with a multi-transmission that improves signal nonuniformity compared to the conventional method) Dielectric Effect, Standing Wave, and B1 RETURN TO INDEX At 3 T , the presence of severe ascites can attenuate the RF signals and create localized shading, as evidenced by this axial T1W GRE (A), axial T2W SS-FSE (B), and coronal T2W SS-FSE (C) MR images.

50. Description Causes Susceptibility A B C * D Signal defects and distortions occur around metallic substances and implants, as well as in the vicinity of air-tissue interfaces on sequences. RETURN TO INDEX Metallic items disturb the magnetic field and cause mismatch between broadcast RF and local tissue signal. Also, fast switching of gradients can induce eddy currents, which cause local distortion around conductive implants. The result is signal variance and image loss near the implant. Axial T1W SE MR image of the brain in a patient with an MR-compatible ventriculoperitoneal (VP) shunt (A). Source image for brain MRA in a patient with an MR-compatible aneurysm clip (B). Coronal PDW FS MR image of the knee after surgical intervention (C). SSFP localizer image (D) in a patient with an MR-compatible pacemaker. Distortion and signal loss (arrowheads) are seen on all images. Additionally on (C), high-intensity signal surrounds the implant (*) owing to fat saturation failure and pile-up of incorrectly encoded signals.

51. A B Susceptibility Sagittal FS PDW MR images of the knee acquired with bandwidth = 195 Hz/pixel (A) and bandwidth = 488 Hz/pixel (B). On the wide-bandwidth sequence, the distortion around the implant (oval) is reduced. Solutions The impact of susceptibility artifacts can be reduced by increasing receiver bandwidth or frequency matrix. However, increasing bandwidth or matrix will decrease SNR. Swapping the phase-encoding direction may change the range of the artifact’s influence. RETURN TO INDEX

52. Susceptibility Solutions Shortening the TE is another effective way to reduce susceptibility artifacts. However, it is not always possible to employ this technique because the tissue contrast depends on TE and may require an additional change to BW or resolution. It is particularly useful, however, in patients undergoing MRA who have an MR-compatible clip or coil. RETURN TO INDEX C D B A MR images of the left calf with an implant in the tibia, obtained with different TEs. (A) TE = 5.3 msec, (B) TE = 3 msec. Signal defect is obvious around implant (arrowheads) but becomes smaller with shorter TEs. MR angiography of the circle of Willis, obtained with TE = 7.2 msec (C) and TE = 3.6 msec (D). This patient has an MR-compatible implantable clip in the left internal carotid artery (arrowheads in C, D); its resulting artifact is reduced with the shorter TE, allowing improved evaluation of this vascular segment.

53. Susceptibility Artifact in Body Imaging A D C B RETURN TO INDEX Focus on: Cardiac 2-chamber SSFP (A, C) and short-axis late gadolinium enhancement (B, D). Black banding artifacts (arrowhead) suggestive of metal are seen along the inferior cardiac w all in (A), but the patient had no history of surgical intervention. O n further investigation, he explained that he had taken iron supplements just before the examination, making it likely that this artifact arose from the iron pill found in the nearby small bowel (arrow in A). Unfortunately, it was impossible to assess inferior wall motion or potential fibrosis (oval in B) owing to these artifact s, so MRI was repeated after 2 weeks (C and D), and no artifacts were seen.

54. Susceptibility Artifact at DWI Abdominal section at the level of the pancreas was obtained with a 3-T scanner. (A) T1W (opposed phase), (B) DWI ( b = 1500). Gas in the stomach, small intestine, and colon may reduce the signal of surrounding tissues or cause distortion on the diffusion-weighted image. In this case, the signal around the pancreas body (arrowheads in B) was lost owing to gas in the transverse colon (oval in A). I t is recommended to prescribe fasting and an enema before the examination. Distortion can be reduced by decreasing the phase FOV or the number of frequency matrix. A B RETURN TO INDEX Focus on:

55. Description Causes Motion Motion can manifest in many ways, including replication along the PE direction (A), generalized blur, or signal loss (C). (A) Axial T1W GRE MR image through the abdomen. G hosting due to pulsation from the descending aorta (arrowheads) hinders evaluation of the pancreatic body. (B) Axial T2W FSE MR image of the orbit with eye motion seen as side-to-side ghosting (oval). For the cardiac T2 map seen in (C), a dark edge is seen in p laces where through-plane wall motion caused voxels to shift location between excitation and acquisition (arrowheads) RETURN TO INDEX Ghosts result from periodic motion (eg, respiration , heart beat, blood flow, and cerebrospinal fluid movement ). Blurring is caused by random motion such as physical movement, swallowing, peristalsis, eye motion, etc. Signal drop-out is caused by through-plane movement. C A B  Phase direction   Phase direction   Phase direction 

56. Baseline C D A B Motion Solutions RETURN TO INDEX  Phase direction  (A) Axial T2W MR images of the neck obtained with free breathing. Abnormal findings in the left thyroid gland (circle ) are unclear owing to ghosting (arrowheads, A). After application of respiratory gating, abnormalities in the gland are clearly visualized (circle, B). Coronal T2W FS FSE MR images of the shoulder with motion artifacts (arrowheads, C) complicate evaluation of the joint cavity. These a rtifacts are diminished after re-scan using radial sampling (D), improving visualization of local structures (oval). Option Trade-Off Example Apply fixation or immobilization to limit physical movement of parts. Examination set-up time Not shown Use respiratory gating or cardiac gating. Scan time, Examination set-up time B Acquire the images using a sequence constructed for breath-holding. Resolution Not shown Increase NSA Scan time Not shown Use a technique that incorporates radial k-space fill. Scan time, introduce the risk of streak artifacts. D Option Trade-Off Example Apply fixation or immobilization to limit physical movement of parts. Not shown Use respiratory gating or cardiac gating. B Acquire the images using a sequence constructed for breath-holding. Not shown Increase NSA Not shown Use a technique that incorporates radial k-space fill. D

57. Short-axis bSSFP cine imaging with varying numbers of phases (A = 7; B = 25, C =51). Seven phases is too few to properly characterize the motion of the heart; thus blur and replication are evident. Conversely, 51 phases makes for a very clean image, but requires high segmentation and thus a long scan time. Using approximately 24 phases is quite common in cardiac MRI. Motion Artifacts at Cine Imaging A B C Description Motion artifacts are not limited to static imaging; insufficient temporal resolution can lead to blur, noise, and other artifacts when cine or dynamic imaging is performed. Solutions Decrease the acquisition period by decreasing PE matrix or increasing acceleration factor or segment data so that a smaller portion is acquired during each cycle. RETURN TO INDEX Focus on:

58. Description Causes Flow and Pulsatile (Also a Type of Motion) Moving fluid (eg, blood and CSF) can replicate along the phase direction. When the source is strongly pulsatile, the resulting ghosts may spread out with diminishing intensities as they move away from their source (C). Sagittal PDW (A) and FS T2W (B) MR images of the knee. F low artifact (arrowheads) is seen from the popliteal artery (*) that is seen overlapping the lateral meniscus. (C) Pulsatile artifact from the basilar artery (arrowheads) can be mistaken for a lesion if pulsatile artifacts are not understood. RETURN TO INDEX Phase encoding assumes that differences in phase are due to differences in spatial location. However, when spins from fluid enter the section plane, they often have phase differences that result from their own intrinsic motion, causing the signal to be encoded as ghosts across the phase FOV. A  Phase direction  * *  Phase direction   Phase direction  B C

59. A B D C Solutions Swapping the frequency and the phase-encoding directions can minimize the impact of motion artifacts on the region of interest (however, this will also affect where aliasing occurs!). Flow and Pulsatile (Also a Type of Motion) (A, B) Axial T1W FS MR images through the calf after contrast agent administration. Pulsatile artifact (arrowheads, A) overlies a lesion (oval). After swapping the phase-encoding direction, the artifact is less obtrusive (B). (C, D) Axial T1W MR images without contrast agent. Flow artifact (arrowhead, C) from the carotid artery overlaps the larynx (circle in C). The artifacts can be shifted away from the larynx after swapping the phase-encoding direction, thus making evaluation easier (D). RETURN TO INDEX

60. Effects of Presaturation Bands in Reducing Flow Artifact A  Phase direction  B PRESATURATION BAND Focus on: Solutions In-plane flow artifact can also be mitigated by adding spatial presaturation over the source of the flow (this has the trade-off of increasing SAR) or making adjustments to sequence parameters (such as shorter ETS, shorter TE, etc.) (A) Sagittal T1W MR images of the lumbar spine. Strong flow artifact (arrowheads) from the descending aorta obfuscates much of T12 and L1 and portions of the cauda equina. R epeat MRI performed with a presaturation band on the abdominal aorta (B) diminishes these flow artifacts.

61. Conclusion There are numerous artifacts that can arise at MRI. Take careful note of the details inherent in each artifact’s style of manifestation; this will aid in identification and allow proper countermeasures to be applied. We hope that this material will not only help learners to better their knowledge on the topic, but also improve the quality of the clinical images and dictations that they provide.

62. Eilenberg SS, Tartar VM, Mattrey RF. Reducing Magnetic Susceptibility Differences Using Liquid Fluorocarbon Pads (Sat Pad™): Results with Spectral Presaturation of Fat. Artificial Cells, Blood Substitutes, and Biotechnology 1994;22(4):1477-1483. Hirokawa Y, Isoda H, Maetani YS, Arizono S, Shimada K, Togashi K. MRI artifact reduction and quality improvement in the upper abdomen with PROPELLER and prospective acquisition correction (PACE) technique. AJR Am J Roentgenol 2008;191(4):1154-1158. Huang SY, Seethamraju RT, Patel P, Hahn PF, Kirsch JE, Guimaraes AR. Body MR Imaging: Artifacts, k-Space, and Solutions. Radiographics 2015;35(5):1439-1460. Maehara M, Ikeda K, Kurokawa H, Omura N, Ikeda S, Hirokawa Y, Maehara S, Utsunomiya K, Tanigawa N, Sawada S. Diffusion-weighted echo-planar imaging of the head and neck using 3-T MRI: Investigation into the usefulness of liquid perfluorocarbon pads and choice of optimal fat suppression method. Magnetic Resonance Imaging 2014;32. Runge M, Ibrahim E-SH, Bogun F, Attili A, Mahani MG, Pang Y, Horwood L, Chenevert TL, Stojanovska J. Metal Artifact Reduction in Cardiovascular MRI for Accurate Myocardial Scar Assessment in Patients With Cardiac Implantable Electronic Devices. American Journal of Roentgenology 2019;213(3):555-561. Stadler A, Schima W, Ba- Ssalamah A, Kettenbach J, Eisenhuber E. Artifacts in body MR imaging: their appearance and how to eliminate them. European Radiology 2007;17(5):1242-1255. Triche BL, Nelson JT, McGill NS, Porter KK, Sanyal R, Tessler FN, McConathy JE, Gauntt DM, Yester MV, Singh SP. Recognizing and Minimizing Artifacts at CT, MRI, US, and Molecular Imaging. RadioGraphics 2019;39(4):1017-1018. Tsuchihashi T. Artifact of MRI(MR Series). Japanese Journal of Radiological Technology. 2003;59(11):1370-1377. Yu H, Reeder SB, Shimakawa A, McKenzie CA, Brittain JH. Robust multipoint water-fat separation using fat likelihood analysis. Magn Reson Med 2012;67(4):1065-1076. Suggested Readings