Artifacts

8
Artifacts




After reading this chapter, you will be able to:



  • Recognize common MRI artifacts.
  • Understand why these artifacts occur.
  • Explain remedies for common MRI artifacts.
  • Analyze the mechanisms of flow and how they are used to image vessels.

INTRODUCTION


All MRI images have artifacts. Some artifacts degrade the image and may mask or even mimic pathology. It is therefore very important to understand their causes and how to compensate for them. Other artifacts are beneficial, and we deliberately create them to demonstrate flow, visualize pathology, or characterize lesions. Some artifacts are irreversible and are only reduced rather than eliminated. Others can be avoided altogether.


In this chapter, we discuss the appearances, causes, and remedy of the most common artifacts encountered in MRI. We also explore flow phenomena and how the artifacts they cause are used to image flowing vessels. A list of acronyms of the five main system manufacturers is provided at the beginning of this book. This includes some of the parameters used to compensate for artifacts and angiography techniques described in this chapter. As with other chapters, scan tips link the theory of artifacts and flow to practice.


PHASE MISMAPPING


Appearance


Phase mismapping or ghosting artifact produces replications of moving anatomy across the image in the phase encoding axis. Phase mismapping usually originates from anatomy that moves periodically throughout the scan such as the anterior abdominal wall during respiration (Figure 8.1), pulsation of vessels and CSF, swallowing, and eye movement. Ghosting is less obvious with distance from the source of motion. The physical separation between each ghost depends on the scan time parameters (see Chapters 6 and 7) and period of motion (Equation (8.1)).










Equation 8.1


therefore

Sp is the separation between ghosts in pixels

TR is the repetition time in millisecond


M(p) is the phase matrix


NSA is the number of signal averages


Tm is the period of motion of the moving object

Tm is calculated thus: heart rate = 60 beats/min or 1 beat/s Tm = 1 s. This equation calculates the separation in pixels and if multiplied by the pixel size determines the actual distance between ghosts
Image described by caption and surrounding text.

Figure 8.1 Axial image through a breathing abdomen showing phase mismapping.


Cause


Phase mismapping is produced by anatomy moving along the phase-encoding gradient during the pulse sequence. Unlike the frequency encoding and slice-select gradients that are applied at the same amplitude every TR, the phase-encoding gradient is applied at a different amplitude (see Chapter 5). As anatomy moves during the scan, its reconstructed signal is misplaced in the phase encoding direction as the gradient amplitude changes. Imagine the anterior abdominal wall moving during the scan as shown in Figure 8.2. Anatomy is located at a position along the phase-encoding gradient during a particular TR period but may move to another position during the next phase-encoding step. Signal from the abdominal wall acquires different phase values depending on its position along the gradient. Accurate reconstruction of the image relies on each line of data having a discrete incremental change in phase. Repetitive motion results in periodic perturbations of the data collected in k-space. After FFT, it is these perturbations that result in “ghosts” of moving anatomy being mismapped into incorrect spatial locations across the image. This is also sometimes described as inter-view or view-to-view mismapping.

Diagram shows timeline with repetition x and repetition y having 90 and 180 degrees with PE x and PE y having time period between phase encoding steps along with inspiration and expiration od phase mismapping.

Figure 8.2 One of the causes of phase mismapping.


Secondly, there is usually a time delay between phase encoding and readout (sometimes called intra-view mismapping), so anatomy may move between phase encoding and the echo. This time factor is very short, typically having a value in millisecond, so little motion is likely during this period. In addition, mismapping does not usually occur along the frequency axis of the image, as frequency encoding is performed as the frequencies in the echo are read and digitized.


Remedy


There are several ways to reduce phase mismapping. Equation (8.1) shows that the amount of separation between ghosts depends on the TR, phase matrix, NSA, and the period of motion. Therefore, if the TR, phase matrix, and NSA increase, the separation between ghosts increases and there are therefore fewer ghosts in the image. The same applies when imaging structures that pulsate rapidly because this reduces the period of motion [2]. There are other remedies, however, so let’s discuss these in more detail.


Swapping phase and frequency


As ghosting only occurs along the phase encoding axis of the image, its direction can be changed so that artifact does not interfere with the area of interest. For example, imagine we select an axial stack of slices through the chest of a patient with a suspected coarctation of the aorta. Frequency encoding is usually performed right to left across the patient, as this is the longest axis of anatomy in the axial plane (Figure 8.3). The phase encoding direction is therefore performed anterior to posterior. Pulsation of the heart and great vessels along the phase encoding axis produces ghosting in this direction of the image and may obscure the aorta. Swapping phase and frequency so that frequency encoding occurs anterior to posterior, and phase encoding occurs right to left, places the artifact away from the area of interest (Figure 8.4). This strategy does not eliminate or reduce movement artifact. It repositions it so that it no longer obscures the area under investigation.

Image described by caption and surrounding text.

Figure 8.3 Axial T1-weighted image of the chest. Phase is anterior to posterior. Source: Westbrook 2014 [3]. Reproduced with permission of John Wiley & Sons.

Image described by caption and surrounding text.

Figure 8.4 Axial T1-weighted image of the chest. Phase is right to left. Source: Westbrook 2014 [3]. Reproduced with permission of John Wiley & Sons.


Presaturation pulses


Presaturation nulls signal from specified areas. If we place presaturation pulses over the source of motion, signal from this motion decreases or is nulled altogether. For example, in sagittal imaging of the cervical spine, swallowing produces ghosting artifact along the phase axis (anterior to posterior) and obscures the spinal cord. A presaturation pulse placed over the throat reduces this artifact. Saturation pulses may also be positioned outside the FOV to saturate the signal from inflowing blood. This reduces phase mismapping from pulsatile flow.


Respiratory compensation techniques


In chest and abdominal imaging, respiratory motion along the phase encoding axis is likely to obscure important structures. There are several techniques that either reduce or eliminate this artifact. In fast sequences, it is usually possible for the patient to hold their breath during the scan. In longer sequences, a method known as respiratory compensation reduces motion artifact from respiration. Manufacturers may offer two types of respiratory sensor. The first kind is a physical device such as respiratory bellows positioned around the patient’s chest, or a pad-shaped sensor that can be placed in proximity to moving anatomy. The second type uses a software-based strategy that tracks the position of the diaphragm using a navigator echo. In the first instance, a strap is placed around the patient’s chest. The strap features a corrugated hollow tube (bellows) that expands and contracts as the patient breathes (Figure 8.5). This expansion and contraction causes a change in volume and air pressure within the bellows that are connected by hollow rubber tubing to a transducer that sends data to a slave computer responsible for physiological monitoring. A transducer is a device that converts the change in pressure within the bellows to an electrical signal. The system analyzes this signal, the amplitude of which corresponds to the maximum and minimum motion of the chest wall during respiration. It then reorders the lines of k-space so that when all the lines are filled, the data in k-space do not reflect the periodic motion of breathing throughout the scan. Modern systems also provide the option to use respiratory navigator echoes that are built into the pulse sequence. These RF pulses are designed to excite a narrow strip of spins and are typically positioned over the right hemidiaphragm. A sample is taken approximately five times per second and can track the position of the diaphragm over time. The amplitude of this signal corresponds to the maximum and minimum excursions of the diaphragm during respiration. The slave computer then reorders the lines of k-space so that when all the lines are filled, the data in k-space do not reflect the periodic motion of breathing throughout the scan.

Image described by caption and surrounding text.

Figure 8.5 Placement of respiratory compensation and cardiac gating leads.

Diagram shows respiratory compensation and k-space with signal from bellows where red color represents resolution data and white color represents signal data for data placed at edge and placed near center of k-space.

Figure 8.6 Respiratory compensation and k-space.

Image described by caption and surrounding text.

Figure 8.7 Image showing respiratory motion.

Image described by caption and surrounding text.

Figure 8.8 Image without respiratory motion.


Respiratory gating and triggering


These techniques utilize signal collected by the transducer or navigator echoes. In this context, the term triggering refers to an action that is triggered by an event. This could be the application of an RF pulse that is triggered by the inspiratory movement of the chest wall. Gating is slightly different in that we select a threshold (gate) that only allows data through when a certain condition is met. This could be used to cease data acquisition at a predefined threshold of the respiratory cycle and only allow data to be collected when the diaphragm is relatively motionless at full expiration. In respiratory triggering, the RF excitation pulse is triggered by each respiratory excursion. Each echo is therefore obtained at the same phase of respiration. However, this method has two main drawbacks. Firstly, the periodicity of the patient’s respiratory cycle becomes the TR and therefore determines the image contrast. This is likely to be around 15 breaths/min and limits the TR to around 4000 ms (or multiple of such). Secondly, the patient’s rate of respiration may change during the acquisition leading to mixed contrast. Respiratory gating does not affect the TR because this technique permits us to select the normal user-defined parameters. The gate threshold is selected to only allow data through that have been collected when the diaphragm is relatively motionless and to reject data that are acquired at the extremes of the respiratory cycle. The image therefore only includes data collected within an acceptance range when the diaphragm is at a comparatively stationary neutral position between breaths. The main trade-off is that up to 60% of the data may be rejected, and this means that the acquisition time increases significantly in order to fill every line of data (Figure 8.9).

Diagram shows graph plotting for respiratory triggering and gating with respiratory excursions for approximately TR 4000 ms and excitation pulses.

Figure 8.9 (a) Respiratory triggering and (b) gating.


Multiple NSA


Increasing the NSA may be a useful strategy to reduce motion artifact. Correctly encoded signal appears at the same position and is reinforced by every signal average. The mismapped signal that creates the ghosts is somewhat randomly located on each signal average and becomes less apparent because it averages out of the image with increasing NSA. The trade-off is that scan time is increased by an entire acquisition period for every additional signal average (see Chapters 6 and 7). This trade-off is compensated by using radial k-space filling in which the central region of k-space where most of the signal is stored is filled every TR, but the edges are only filled once. As a result, the image benefits from multiple signal averaging, but with a smaller trade-off in scan time (see Chapter 6). Voluntary motion is also reduced by making the patient as comfortable as possible and immobilizing them with pads and straps. A nervous patient always benefits from a thoughtful explanation of the procedure and a constant reminder to keep still over the system intercom. A relative or friend in the room can also help in some circumstances. In extreme cases, sedation of the patient may be required. Involuntary physiological motion, such as bowel peristalsis, is controlled by administering antispasmodic agents.


Cardiac triggering


Triggering may also be used in cardiac studies. Cardiac triggering monitors cardiac motion by coordinating the RF excitation pulse with the R wave of cardiac systole. This is achieved by using an electrical signal generated by cardiac motion to trigger each RF excitation pulse. There are two forms of triggering used:



  • Electrocardiogram (ECG, EKG) gating uses electrodes and lead wires that are attached to the patient’s chest to produce an EKG (see Figure 8.9). This is used to determine the timing of each RF excitation pulse. Each slice is acquired at the same phase of the cardiac cycle, and therefore phase mismapping from cardiac motion decreases. EKG triggering should be used when imaging the chest, heart, and great vessels.
  • Peripheral pulse triggering uses a light sensor attached to the patient’s finger to detect the pulsation of blood through the capillaries. The pulsation is used to trigger RF excitation pulses so that each slice is acquired at the same phase of the cardiac cycle. Peripheral triggering is not as accurate as EKG triggering, so it is not very useful when imaging the heart itself. However, it is effective at reducing phase mismapping when imaging small vessels or the spinal cord where CSF flow may degrade the image.

Gradient moment rephasing


This reduces ghosting caused by flowing nuclei moving along gradients (see later).


Table 8.1 Things to remember – phase mismapping.











Phase mismapping, ghosting, or motion artifact is caused by periodic motion mainly as a result of spins moving between each phase encode
It mainly originates from breathing and pulsatile motion of vessels and CSF
Respiratory compensation, gating, presaturation, and gradient moment nulling are the main techniques used to reduce this artifact
Artifacts and their remedies are summarized in Table 8.8

ALIASING


Appearance


Aliasing or wrap is an artifact where anatomy that exists outside the FOV is folded onto the top of anatomy inside the FOV. In Figure 8.10, the FOV in the phase direction is smaller than the anterior-to-posterior dimensions of the head. Therefore, signal outside the FOV in the phase direction is wrapped into the image.

Image described by caption and surrounding text.

Figure 8.10 Sagittal image of the brain showing aliasing or wrap around.


Cause


Anatomy outside the FOV still experiences the effects of the gradients and produces a signal if it is within the receiving volume of the receiver coil. The signal from this anatomy has frequencies that are higher or lower than those within the FOV because nuclei are positioned on parts of the gradient that extend beyond the FOV. If the frequency exceeds the Nyquist frequency, it is not accurately digitized and is represented as a lower frequency [5].


Frequency wrap


Aliasing along the frequency encoding axis is known as frequency wrap. When the FOV is smaller than the anatomy in the frequency direction of the image, frequencies outside the FOV are higher than the Nyquist frequency and are mapped to a lower frequency. This is called high- frequency aliasing [6] (Figure 8.11, bottom image).

Diagram shows graph with plotting for aliasing and under sampling which are sampled twice per cycle, waveform interpreted accurately, misinterpreted as straight lines, and misinterpreted as wrong frequency.

Figure 8.11 Aliasing and undersampling.


Phase wrap


Aliasing along the phase axis of the image is known as phase wrap. This is caused by undersampling of data along the phase axis of the image. Signal originating outside the FOV in the phase direction is allocated a phase value and therefore a pseudo-frequency that has already been given to signal originating from inside the FOV.


Look at Figure 8.12 where the FOV in the right-to-left phase encoding axis of the image is smaller than the dimensions of the abdomen. The phase-encoding gradient is applied in this direction and produces a change of phase across the bore of the magnet. This gradient is applied numerous times at different amplitudes during the acquisition and incrementally changes the phase position of the magnetic moments of each spin every TR (see Chapter 6). This produces a certain pseudo-frequency (i.e. the magnetic moments of each spin appear to have completed a number of cycles over the duration of the scan), and, using the gradient slope shown in Figure 8.12, this is the same as those inside the FOV (red and blue areas in the diagram). As they have the same pseudo-frequency, these red and blue areas are wrapped inside the FOV.

Image described by caption and surrounding text.

Figure 8.12 Phase wrap.










Equation 8.2
fp is the perceived frequency in KHz

ft is the actual frequency in KHz

This equation calculates the degree of aliasing

Remedy


Aliasing along both the frequency and phase axes can degrade an image and should be compensated for. This is achieved by enlarging the FOV to incorporate all signal-producing anatomy. However, in the case of the phase FOV, this strategy increases the scan time. Another option is to use presaturation bands on areas outside the FOV that may wrap into the image. These can sometimes null signal from these areas and reduce aliasing. There are, however, two antialiasing software methods that compensate for wrap.


Antialiasing along the frequency axis


Aliasing in the frequency direction is eliminated by increasing the digital sampling rate so that all frequencies are sufficiently digitized. This is achieved by decreasing the sampling interval while maintaining the same sampling window. The number of data points increases, but only those related to the central frequencies are displayed [6]. This is called a low-pass frequency filter. It essentially eliminates the frequencies that exceed the bandwidth [7] (Figure 8.13) and is like filtering out the bass and treble on a music system with a graphic equalizer.

Diagram shows antialiasing along with frequency axis where analogous to Graphic equalizer with bass and treble is turned down with digital filter on top and between frequencies where frequencies outside— FOV are not sampled.

Figure 8.13 Antialiasing along the frequency axis.


Antialiasing along the phase axis


This is termed no phase wrap, phase oversampling, or fold-over suppression. Antialiasing software oversamples by increasing the number of phase encodings steps. This increases the number of k-space lines, and more data are stored so that there is no duplication of spatial frequencies. However, increasing the number of phase encoding increases the scan time. This is automatically compensated for by some manufacturers by reducing the NSA.

Diagram shows antialiasing along phase axis with original FOV, phase curve extended to cover new FOV, and central portion of FOV displayed where k-space fillings remain same.

Figure 8.14 Antialiasing along the phase axis.

Image described by caption and surrounding text.

Figure 8.15 With wrap.

Image described by caption and surrounding text.

Figure 8.16 Without wrap.


Aliasing can also occur in the slice-select direction in volume imaging. This is because the slice-select gradient acts like a second phase-encoding gradient and is used to locate each slice within the imaging volume (see Chapter 6). Aliasing occurs because of undersampling during this process. The effect is reduced by ensuring the volume includes all of the anatomy lying outside the region under investigation. For example, signal from the patient’s nose may otherwise wrap into the occipital area of the brain on the image.


Table 8.2 Things to remember – aliasing.













Aliasing is caused by undersampling of frequencies
If frequencies are not sampled often enough, the system cannot accurately represent those frequencies in the image
In the frequency-encoding direction of the image, this is avoided by ensuring that the digital sampling frequency is at least twice that of the highest frequency present, and additional low-pass filters are used
In the phase encoding direction of the image, aliasing is remedied by using antialiasing software. This increases the phase FOV, thereby oversampling during the phase encoding process. As a result, there is less likelihood that anatomy exists outside the larger FOV
Artifacts and their remedies are summarized in Table 8.8

CHEMICAL SHIFT ARTIFACT


Appearance


Chemical shift artifact causes the misplacement of signal from fat in the image (Figure 8.17). It can also create both signal voids and signal superimposition (high signal) in areas where fat and water interface. The renal area is a good example of this, as fluid-filled kidneys are surrounded by perirenal fat. The signal loss represents a boundary where the signal from fat has shifted by a certain number of voxels, leaving an area devoid of signal. The hyperintense area on the opposite side of the kidney is caused by the shifted fat signal being superimposed upon the signal from the underlying anatomy. The overall appearance gives an embossed effect where the anatomical structures appear to have been “lit” from one direction, casting a “shadow” in the opposite aspect. For this reason, the artifact is often referred to as the bas-relief artifact.

Image described by caption and surrounding text.

Figure 8.17 Chemical shift. In this image, the signal from the fatty bone marrow in the talus has been shifted inferiorly in the frequency-encoding direction. This is more apparent on the 16 KHz image where the signal void mimics a thickening of cortical bone.


Cause


Chemical shift artifact is caused by different chemical environments of fat and water. Although fat and water both contain hydrogen atoms, fat consists of hydrogen arranged with a chain of carbon atoms, while in water, hydrogen is arranged just with a single oxygen atom (see Chapter 2). Fat is therefore a much larger molecule than water and the hydrogen atoms within each molecule are surrounded by many other atoms. Triglyceride fat molecules are sometimes described as a “self-shielding” because the electron clouds shield the hydrogen atoms from the static field B0. This self-shielding is much more pronounced in fat than in water and results in a lowering of the Larmor frequency of the magnetic moments of hydrogen nucle in fat [9].


The difference in the precessional frequencies between magnetic moments in fat and water is called chemical shift, or sometimes fat/water shift. The amount of chemical shift is often expressed in arbitrary units known as parts per million (ppm) of the main magnetic field strength. Its value is always independent of the main field strength and equals 3.5 ppm. From this, the chemical shift between fat and water is calculated at different field strengths (Equation (8.3)). For example, at 1.5 T the difference in precessional frequency is approximately 220 Hz. The magnetic moments of fat nuclei precess 220 Hz lower than the magnetic moments of water nuclei. At 1.0 T, this difference is 147 Hz, and at lower field strengths (0.5 T or less), it is even lower and usually insignificant.










Equation 8.3
ωcsf = ω0 × Cs ωcsf is the chemical shift frequency difference between fat and water (Hz)

ω0 is the precessional frequency (Hz)


Cs is the chemical shift (3.5 ppm or 3.5 × 10−6)

At 1.5 T, for example, the precessional frequency is 63.86 MHz (63.86 × 106). Therefore, the chemical shift frequency difference between fat and water at 1.5 T is 220 Hz

Chemical shift artifact occurs because the difference in frequency between the magnetic moments of fat and water causes them to be placed into different pixels in the image. The receive bandwidth determines the range of frequencies that are accurately sampled and displayed across the FOV in the frequency direction of the image. The receive bandwidth and the number of frequency samples (or data points in each line of k-space) determine the bandwidth of each pixel. For example, if the receive bandwidth is ± 16 KHz, 32 000 Hz is mapped across the FOV. If 256 data points are collected, the FOV is divided into 256 frequency pixels. Each pixel therefore has an individual frequency range of 125 Hz/pixel (32 000 ÷ 256 Hz) (Figure 8.18). At a field strength of 1.5 T, the precessional frequency difference between the magnetic moments of fat and water nuclei is 220 Hz, and therefore using the above example, fat and water protons that exist adjacent to one another in the patient are mapped 1.76 pixels apart (220 ÷ 125) (Figure 8.18, middle diagram). The actual dimensions of this artifact depend on the size of the FOV, as this determines the size of each pixel. For example, an FOV of 240 mm and 256 frequency columns result in a pixel size of 0.93 mm. A pixel shift of 1.76 therefore results in an actual chemical shift between fat and water of 1.63 mm (0.93 × 1.76 mm) (Equation (8.4)).










Equation 8.4
CSp is the pixel shift (mm)

Cs is the chemical shift (3.5 ppm or 3.5 × 10−6)


γ is the gyromagnetic ratio (MHz/T)


B0 is the main magnetic field strength (T)


FOV is the field of view (cm)


RBW is the receive bandwidth (Hz)


M(f ) is the frequency matrix

This equation calculates the pixel shift in millimeters caused by the chemical shift between fat and water. To calculate the actual number of pixels that fat and water are shifted, remove the FOV function from this equation
Diagram shows timeline where chemical and pixel shifts are seen at 125 Hz to 256 Hz frequency direction for fat and water.

Figure 8.18 Chemical shift and pixel shift.


Remedy


Chemical shift artifact is reduced by scanning at a lower field strength and by minimizing the FOV. At high field strengths, chemical shift is limited by increasing the receive bandwidth. As the receive bandwidth increases, a larger frequency range is mapped across a certain number of frequency pixels (depending on the frequency matrix). The individual frequency range of each pixel (its bandwidth) therefore increases, so the 220 Hz difference in precessional frequency between fat and water is translated into a small pixel shift. For example, if the receive bandwidth is ± 32 KHz, 64 000 Hz is mapped across 256 frequency pixels. Each pixel has a bandwidth of 250 Hz (64 000 ÷ 256 Hz). The 220 Hz precessional frequency difference between the adjacent fat and water protons at 1.5 T is translated into a pixel shift of less than 1 (220 ÷ 250). If the receive bandwidth decreases to ± 8 KHz, 16 000 Hz is mapped across 256 frequency pixels. Each pixel has a bandwidth of 62.5 Hz (16 000 ÷ 256 Hz). The 220 Hz precessional frequency difference between adjacent fat and water protons at 1.5 T is now translated into a pixel shift of 3.52 pixels (220 ÷ 62.5) (Figure 8.18, lower diagram).

Mar 9, 2019 | Posted by in MAGNETIC RESONANCE IMAGING | Comments Off on Artifacts
Premium Wordpress Themes by UFO Themes