Image Detail and Noise
Image Detail and Noise
|Link to Book Table of Contents||Chapter Contents Shown Below|
|Introduction And Overview||Image Detail||Noise Sources|
|Signal-To-Noise Considerations||Voxel Size||Field Strength|
|Tissue Characteristics||TR and TE||RF Coils|
|Mind Map Summary|
Introduction And Overview
Two characteristics of the MR image that reduce the visibility of anatomical structures and objects within the body are blurring and visual noise. These were introduced in Chapter 1 as image quality characteristics. Both image blurring and visual noise are undesirable characteristics that collectively reduce the overall quality of an image and the objects in the image as illustrated in Figures 1-6 and 1-7. In an image, the combined effects of blur and noise produce a “curtain of invisibility” that extends over some objects based on object characteristics. This is shown in Figure 10-1, where we see objects arranged according to two characteristics. In the horizontal direction, the objects are arranged according to size. Decreasing object size corresponds to increasing detail.
Figure 10-1. The impact of image noise and blurring on object visibility. Noise reduces visibility of low contrast objects. Blur reduces visibility of small objects.
In the vertical direction, the objects are arranged according to their contrast. The object in the lower left is both large and has a high level of contrast. This is the object that would be most visible under a variety of imaging conditions. The object that is always the most difficult to see is the small, low contrast object, which in Figure 10-1 would be located in the upper right corner.
In every imaging procedure we can assume that some potential objects within the body will not be visible because of the blurring and noise in the image. This loss of visibility is represented by the “curtain” or area of invisibility indicated in Figure 10-1. The location of the boundary between the visible and invisible objects, often referred to as a contrast-detail curve, is determined by the amount of blurring and noise associated with a specific imaging procedure. In general, blurring reduces the visibility of anatomical detail or other small objects that are located in the lower right region. Visual noise reduces the visibility of low contrast objects located in the upper left region.
The imaging protocol determines the boundary of visibility by altering the amount of blurring and noise. These two characteristics are determined by the combination of many adjustable imaging factors. It is a complex process because the factors that affect visibility of detail (blurring) also affect noise, but in the opposite direction. As we will see when a protocol is changed to improve visibility of detail, the noise is increased. Another point to consider is that several of the factors that have an effect on both image detail and noise also affect image acquisition time, which will be discussed in Chapter 11. Therefore, when formulating an imaging protocol one must consider the multiple effects of the imaging factors and then select factor values that provide an appropriate compromise and an optimized acquisition for a specific clinical study with respect to detail (blurring), noise, and acquisition speed.
We will now consider the many factors that have an effect on the characteristics of image detail and noise.
The ability of a magnetic resonance image to show detail is determined primarily by the size of the tissue voxels and corresponding image pixels. Pixel size can be changed without major tradeoffs. However, as we are about to observe, there are significant effects of changing voxel size that must be considered. The real challenge is selecting a voxel size that is optimum for a specific clinical procedure.
In principle, all structures within an individual voxel are blurred together and represented by the signal intensity representing that voxel. It is not possible to see details within a voxel, just the voxel itself. When we view an MR image, we are actually looking at an image of a matrix, or array, of the voxels. We usually do not see the individual voxels because they are so small and they might be interpolated into even smaller image pixels. However, even if we do not see the individual voxels, their size determines the anatomical detail that we can see. The amount of image blurring is determined by the dimensions of the individual voxels.
Three basic imaging factors determine the dimensions of a tissue voxel, as illustrated in Figure 10-2.
|Figure 10-2. Voxel size and detail in MR images is determined by the values selected for the three protocol factors: FOV, matrix size, and slice thickness.|
The dimension of a voxel in the plan of the image is determined by the ratio of the field of view (FOV) and the size of the matrix. Both of these factors can be used to adjust image detail. The thickness of the slice is a factor in voxel signal intensity.
The selection of the FOV is determined primarily by the size of the body part being imaged. One problem that can occur is the appearance of foldover artifacts when the FOV is smaller than the actual body section. However, there are artifact suppression techniques that can be used to reduce this foldover problem, as described in Chapter 14. The maximum useful FOV is usually limited by the dimensions and characteristics of the RF coil. The important thing to remember is that smaller image FOVs and smaller voxels produce better visibility of detail.
Matrix size refers to the number of voxels in the rows or columns of the matrix. The matrix size is a protocol factor selected by the operator before the imaging procedure. Typical matrix dimensions are in the range of 128 to 512 mm.
Random RF energy can be generated by thermal activity within electrical conductors and circuit components of the receiving system. In principle, the patient’s body is a component of the RF receiving system. Because of its mass, it becomes the most significant source of image noise in most imaging procedures. The specific noise source is the tissue contained within the sensitive FOV of the RF receiver coils. Some noise might be generated within the receiver coils or other electronics, but it is usually much less than the noise from the patient’s body.
Many devices in the environment produce RF noise or signals that can interfere with MRI. These include radio and TV transmitters, electrosurgery units, fluorescent lights, and computing equipment. All MR units are installed with an RF shield, as described in Chapter 2, to reduce the interference from these external sources. External interference is not usually a problem with a properly shielded unit. When it does occur, it generally appears as an image artifact rather than the conventional random noise pattern.
Image quality is not dependent on the absolute intensity of the noise energy but rather the amount of noise energy in relation to the image signal intensity. Image quality increases in proportion to the signal-to-noise ratio. When the intensity of the RF noise is low in proportion to the intensity of the image signal, the noise has a low visibility. In situations where the signals are relatively weak, the noise becomes much more visible. The principle is essentially the same as with conventional TV reception. When a strong signal is received, image noise (snow) is generally not visible; when one attempts to tune in to a weak TV signal from a distant station, the noise (noise) becomes significant.
In MRI, the loss of visibility resulting from the noise can be reduced by either reducing the noise intensity or increasing the intensity of the signals. This is illustrated in Figure 10-3.
|Figure 10-3. Factors that affect the signal-to-noise ratios in MR images|
Let us now see how this can be achieved.
One of the major factors that affects signal strength is the volume of the individual voxels. The signal intensity is proportional to the total number of protons contained within a voxel. Large voxels, that contain more protons, emit stronger signals and result in less image noise. Unfortunately, as we have just discovered, large voxels reduce image detail. Therefore, when the factors for an imaging procedure are being selected, this compromise between signal-to-noise ratio and image detail must be considered. The major reason for imaging relatively thick slices is to increase the voxel signal intensity and it also allows shorter TE values.
The strength of the RF signal from an individual voxel generally increases in proportion to the square of the magnetic field strength. However, the amount of noise picked up from the patient’s body often increases with field strength because of adjustments in the bandwidth factor for the higher fields. This is described in Chapter 14. Because of differences in system design, no one precise relationship between signal-to-noise ratio and magnetic field strength applies to all systems. In general, MRI systems operating at relatively high field strengths produce images with higher signal-to-noise ratios than images produced at lower field strengths, when all other factors are equal.
Signal intensity, and the signal-to-noise ratio, depend to some extent on the magnetic characteristics of the tissue being imaged. For a specific set of imaging factors, the tissue characteristics that enhance the signal-to-noise relationship are high magnetic nuclei (proton) concentration, short T1, and long T2. The primary limitation in imaging nuclei other than hydrogen (protons) is the low tissue concentration and the resulting low signal intensity.
TR and TE
Repetition time (TR) and echo time (TE) are the factors used to control contrast in most imaging methods. We have observed that these two factors also control signal intensity. This must be taken into consideration when selecting the factors for a specific imaging procedure.
When a short TR is used to obtain a T1weighted image, the longitudinal magnetization does not have the opportunity to approach its maximum and produce high signal intensity. In this case, some signal strength must be sacrificed to gain a specific type of image contrast. Also, when TR is reduced to decrease image acquisition time, image noise can become the limiting factor.
When long TE values are used, the transverse magnetization and the resulting signal it produces can decay to very low values. This causes the images to display more noise.
The most direct control over the amount of noise energy picked from the patient’s body is achieved by selecting appropriate characteristics of the RF receiver coil. In principle, noise is reduced by decreasing the amount of tissue within the sensitive region of the coil. Most imaging systems are equipped with interchangeable coils. These include a body coil, a head coil, and a set of surface coils as shown in Figure 10-4.
Figure 10-4. Both the amount of noise and the intensity of the signal received depend on the RF receiving coils. The body coil picks up the most noise and the weakest signal, resulting in the highest noise level in the image.
The body coil is the largest coil and usually contains a major part of the patient’s tissue within its sensitive region. Therefore, body coils pick up the greatest amount of noise. Also, the distance between the coil and the tissue voxels is greater than in other types of coils. This reduces the intensity of the signals actually received by the coil. Because of this combination of low signal intensity and higher noise pickup, body coils generally produce a lower signal-to-noise ratio than the other coil types.
In comparison to body coils, head coils are both closer to the imaged tissue and generally contain a smaller total volume of tissue within their sensitive region. Because of the increased signal-to-noise characteristic of head coils, relatively small voxels can be used to obtain a better image detail.
The surface coil provides the highest signal-to-noise ratio of the three coil types. Because of its small size, it has a limited sensitive region and picks up less noise from the tissue. When it is placed on or near the surface of the patient, it is usually quite close to the voxels and picks up a stronger signal than the other coil types. The compromise with surface coils is that their limited sensitive region restricts the useful FOV, and the sensitivity of the coil is not uniform within the imaged area. This non-uniformity results in very intense signals from tissue near the surface and a significant decrease in signal intensity with increasing depth. The relatively high signal-to-noise ratio obtained with surface coils can be traded for increased image detail by using smaller voxels.
Bandwidth is the range of frequencies (RF) that the receiver is set to receive and is an adjustable protocol factor. It has a significant effect on the amount of noise picked up. This is because the noise is distributed over a wide range of frequencies, whereas the signal is confined to a relatively narrow frequency range. Therefore, when the bandwidth is increased, more noise enters the receiver. The obvious question is: Why increase bandwidth? One reason is that a wider bandwidth reduces the chemical shift artifact that will be described in Chapter 14. Also, wider bandwidths are the result of short signal sampling, or “picture snapping” times that are useful in some applications.
One of the most direct methods used to control the signal-to-noise characteristics of MR images is the process of averaging two or more signal acquisitions. In principle, each basic imaging cycle (phase-encoding step) is repeated several times and the resulting signals are averaged to form the final image as illustrated in Figure 10-5.
Figure 10-5. an mage with reduced noise is created by averaging
the signals from four acquisitions.
The averaging process tends to reduce the noise level because of its statistical fluctuation nature, from one cycle to another. You can think of it as acquiring four images by repeating the basic acquisition four times. Then the signal intensities in each pixel position in the four images are averaged to produce an intensity value for the new averaged image.
The disadvantage of averaging is that it increases the total image acquisition time in proportion to the number of cycle repetitions or number of signals averaged (NSA). The NSA is one of the protocol factors set by the operator. Typical values are 1 (no averaging), 2, or 4, depending on the amount of noise reduction required. The general relationship is that the NSA must be increased by a factor of 4 to improve the signal-to-noise ratio by a factor of 2. The signal-to-noise ratio is proportional to the square root of the NSA. Sometimes the noise contribution from independent acquisitions adds; sometimes it cancels. Since the signals always add, adding or averaging independently acquired images improves the signal-to-noise ratio.
Parallel imaging is a technique that is useful for reducing acquisition time. The signals are acquired with a set of coils in a phased-array configuration. Making use of the geometric sensitivity of each coil within the array some degree of anatomical spatial information is obtained and used in the image reconstruction process. This makes it possible to reduce the number of phase-encoding gradient cycles and the associated acquisition time.
Mind Map Summary
Image Detail and Noise
The level of noise that appears in an image depends on the relationship (ratio) of the signal strength from the individual voxels and the noise strength coming from a region of the patient’s body. The visible noise is reduced by increasing signal strength. This can be done by increasing the magnetic field strength, increasing voxel size, increasing TR, and decreasing TE. The field strength is a design characteristic and cannot be changed by the operator. Increasing voxel size to decrease noise has the adverse effect of also increasing blurring. Voxel sizes must be chosen to provide an appropriate balance between blurring and noise.
The noise strength picked up from the patient’s body is determined by the mass of tissue contained within the sensitive pickup region of the RF coils. Surface coils that cover a relatively small anatomical region and are also close to the signal source (voxels) produce a high signal-to-noise relationship that results in lower image noise. The RF receiver bandwidth can be adjusted to block some of the noise energy from being received. However, decreasing the bandwidth to reduce noise has the adverse effect of increasing the chemical shift artifact.
Signal averaging is a useful technique for reducing noise but has the adverse effect of increasing acquisition time.