Protocol optimization

7
Protocol optimization




After reading this chapter, you will be able to:



  • Discuss factors that affect protocol optimization.
  • Analyze parameters that influence each of these characteristics.
  • Use what you have learned to effectively modify scan protocols.

INTRODUCTION


In this chapter, we explore how to develop and modify scan protocols. We discuss common protocol parameters and how image characteristics and the acquisition are affected when these parameters are changed. Protocol optimization enables us to maximize image quality and acquire diagnostic images in the shortest scan time. These skills are a major part of clinical MRI practice.


Although it is common to view a protocol as a way of examining a certain area or pathology, e.g. brain protocol, tumor protocol, protocols must be considered in a much broader context than this. A protocol is defined as a “set of rules,” and in MRI these rules are a variety of different parameters that we select at the imaging console. They include extrinsic contrast parameters, geometry parameters, and a variety of imaging options and data acquisition methods. Protocols are judged by how well they show anatomy and pathology, and this is based on producing images that demonstrate the following four characteristics:



  • High signal-to-noise ratio (SNR)
  • Good contrast-to-noise ratio (CNR)
  • High spatial resolution
  • Short scan time.

In an ideal world, all four of these characteristics are achieved in every image. However, due to a variety of constraints, this is not usually possible. Optimizing parameters in favor of one of the aforementioned characteristics usually means compromising another. The skill lies in making informed decisions about which is most important for each patient and pathology, and using knowledge of underpinning physics to appropriately balance protocol parameters.


Let’s investigate the four main characteristics that determine protocol optimization. Each is defined and then the factors that affect them are explored. This chapter also explains how changing any of these parameters affects another. These are called trade-offs. A list of acronyms of the five main system manufacturers is provided at the beginning of the book. This includes some of the scan parameters and imaging options described in this chapter. Scan tips are used to apply the theory of protocol optimization to practice.


SIGNAL-TO-NOISE RATIO (SNR)


The SNR is defined as the ratio of the amplitude of signal received to the average amplitude of the background noise.



  • Signal is the voltage induced in the receiver coil by the precession of coherent magnetization in the transverse plane at, or about, time TE.
  • Noise represents frequencies that exist randomly in space and time.

Signal is cumulative and predictable. It occurs at, or near, time TE and at specific frequencies at, or near, the Larmor frequency. It depends on many factors and can be altered. Noise, on the other hand, is not predictable and is detected by the whole coil volume [1]. It occurs at all frequencies and is also random in time and space. It is equivalent to the hiss on a radio when the station is not tuned in properly, and some of it is energy left over from the “Big Bang.” In the context of MRI, the main source of noise is from thermal motion in the patient [2] but it is also generated by background electrical noise of the system. Noise is constant for every patient and depends on the build of the patient, the area under examination, and the inherent noise of the system. The purpose of optimizing SNR is to make the contribution from signal larger than that from noise. As signal is predictable and noise is not, this usually means using measures that increase signal relative to noise, rather than reducing noise relative to signal.


Therefore, any factor that affects signal amplitude affects the SNR. These are as follows:



  • Magnetic field strength of the system
  • Proton density of the area under examination
  • Coil type and position
  • TR, TE, and flip angle
  • Number of signal averages (NSA)
  • Receive bandwidth
  • Voxel volume (Equation (7.1)).









Equation 7.1
SNR ∝ (voxel volume) SNR ∝ inline

M(p) is the phase matrix


M(f) is the frequency matrix


NSA is the number of signal averages


RBW is the receive bandwidth (KHz)

This equation shows some of the parameters related to SNR. The parameters relate to the SNR by a square root of the total sampling time of the slice. Therefore, another way of expressing this is that the SNR is proportional to the voxel volume and to the square root of the total sampling time

Magnetic field strength


The magnetic field strength plays an important part in determining SNR. As we discovered in Chapter 1, as field strength increases, so does the energy gap between high- and low-energy nuclei. As the energy gap increases, fewer nuclei have enough energy to align their magnetic moments in opposition to B0. Therefore, the number of spin-up nuclei increases relative to the number of spin-down nuclei. The NMV increases at higher field strengths, and there is more available magnetization to image the patient. SNR therefore increases. Although the magnetic field strength cannot be altered, when imaging with low-field systems, SNR may be compromised, and it might be necessary to alter protocol parameters that boost the SNR. This usually manifests itself in longer scan times.


Proton density


The number of protons in the area under examination determines the amplitude of received signal. Areas of low proton density in terms of those that are MR-active (such as the lungs) have low signal and therefore low SNR, whereas areas with a high proton density (such as the pelvis) have high signal and therefore high SNR. The proton density is inherent to the tissue and cannot be changed (that is why it is an intrinsic contrast parameter (see Chapter 2)). However, as the SNR is likely to be compromised when imaging areas of low proton density, steps may have to be taken to boost the SNR that are not necessary when scanning areas with a high proton density. In addition, when measures are taken to null or saturate signal from a tissue, SNR decreases because signal contribution from that tissue is removed (see Presaturation later in this chapter).


Type of coil


The type of coil affects the amount of received signal and therefore the SNR. Larger coils receive more noise in proportion to signal than smaller coils because noise is received from the entire receiving volume of the coil [1]. Coil types are discussed in Chapter 9. Quadrature coils increase SNR because several coils are used to receive signal. Phased array coils increase SNR as the data from several coils are added together. Surface coils placed close to the area under examination also increase SNR. The use of the appropriate receiver coil plays an extremely important role in optimizing SNR. In general, the size of the receiver coil should be chosen such that the volume of tissue optimally fills the sensitive volume of the coil. Large coils, however, increase the likelihood of aliasing, because tissue outside the FOV is more likely to produce signal (see Chapter 8). The position of the coil is also very important for maximizing SNR. To induce maximum signal, the coil must be positioned in the transverse plane perpendicular to B0. Angling the coil, as sometimes happens when using surface coils, results in a reduction of SNR (Figure 7.1).

Image described by caption and surrounding text.

Figure 7.1 Coil position vs SNR.


TR, TE, and flip angle


Although TR, TE, and flip angle are usually considered parameters that influence image contrast (see Chapters 2–4); they also affect SNR and therefore image quality.


Image described by caption and surrounding text.

Figure 7.2 (a) TR 700 ms, (b) TR 500 ms, (c) TR 300 ms, (d) TR 140 ms.

Image described by caption and surrounding text.

Figure 7.3 Changing TR at 3 T.

Diagram shows graph on flip angle versus SNR where flip angle is at 30 degree and 90 degree along with small traverse components.

Figure 7.4 Flip angle vs SNR.

Image described by caption and surrounding text.

Figure 7.5 Axial gradient-echo image of the brain using a flip angle of 10° at 3 T.

Image described by caption and surrounding text.

Figure 7.6 Axial gradient-echo image of the brain using a flip angle of 90° at 3 T.

Diagram shows graph with plotting for SNR versus TE where signal intensity decreases from top to bottom with first echo short TE, second echo long TE.

Figure 7.7 SNR vs TE.

Image described by caption and surrounding text.

Figure 7.8 (a) TE 11 ms, (b) TE 20 ms, (c) TE 40 ms, (d) TE 80 ms.

Image described by caption and surrounding text.

Figure 7.9 Changing TE at 3 T.


Number of signal averages (NSA or NEX)


The NSA controls the amount of data stored in each line of k-space (see Chapter 6). It is the number of times data are collected with the same amplitude of phase encoding slope and therefore how many times a line of k-space is filled with data. Doubling the NSA therefore doubles the amount of data that are stored in each line of k-space, while halving the NSA halves this amount. These data contain both signal and noise. Signal is additive over each signal average but the noise is not. It therefore increases by the factor of a square root [3]. For example, doubling the NSA only increases the SNR by √2 (=1.4) (Figure 7.10). To double the SNR, the NSA and therefore the scan time are increased by a factor of 4 [4]. To triple the SNR requires a ninefold increase in NSA and scan time. Increased scan time increases the chances of patient movement.

Diagram shows bar graph on number of signal averages versus relative SNR where when signal average is high SNR is also high and vice versa.

Figure 7.10 SNR vs NSA.


Look at Figures 7.11 and 7.12 where the NSA increases from 1 to 4. The SNR is undoubtedly greater in Figure 7.12 (exactly twice) but took four times longer to acquire than in Figure 7.11. Therefore, increasing the NSA is not necessarily the best way of increasing SNR because it results in a disproportionate increase in scan time. However, increasing the NSA also reduces motion artifact (see Chapter 8).

Image described by caption and surrounding text.

Figure 7.11 Sagittal brain using 1 NSA.

Image described by caption and surrounding text.

Figure 7.12 Sagittal brain using 4 NSA.


Receive bandwidth


This is the range of frequencies that are accurately sampled during the sampling window (see Chapter 6). Reducing the receive bandwidth results in less noise sampled relative to signal. By applying a filter, noise frequencies much higher and lower than signal frequencies are filtered out. Look at Figure 7.13. The areas shaded green and red represent the ratio of signal to noise, respectively (the squares are shaded orange where signal frequencies are the same as noise frequencies). In the left-hand diagram (which represents a broad receive bandwidth), there are 15 green signal squares and 7 red noise squares. Therefore, the SNR is approximately 2 : 1. In the right-hand diagram (which represents a narrow receive bandwidth), there are still 15 green signal squares but only 5 red noise squares. Therefore, the SNR increases to 3 : 1. Although the height of the signal curve is lower in the left-hand diagram, the area under each curve is the same (i.e. 15 green squares). The height of the signal curve in the left-hand diagram is lower because signal frequencies are spread over a wider frequency range. Therefore, as the receive bandwidth decreases, SNR increases as less noise is sampled as a proportion of signal. Halving the bandwidth increases the SNR by about 40%, but increases the sampling window [5]. As a result, reducing the bandwidth increases the minimum TE (see Chapter 6). Reducing the bandwidth also increases chemical shift artifact (see Chapter 8).

Diagram shows two plotting for frequency versus amplitude with signal and noise plotted with less noise and more noise for same signal.

Figure 7.13 SNR vs receive bandwidth.


Voxel volume


The building unit of a digital image is a pixel. The brightness of the pixel represents the strength of the MRI signal generated by a unit volume of patient tissue (voxel) (see Chapter 6). Voxel dimensions are determined by the pixel area and the slice thickness (Figure 7.14). The pixel area is determined by the field of view (FOV) and the number of pixels in the FOV or image matrix.

Image described by caption and surrounding text.

Figure 7.14 The voxel. The large blue square is the FOV.


Large voxels contain more spins or nuclei than small voxels and therefore have more nuclei to contribute toward signal. Large voxels consequently have a higher SNR than small voxels (Figure 7.15). SNR is therefore proportional to the voxel volume, and any parameter that alters the size of the voxel changes the SNR. Any selection that decreases the size of the voxel decreases the SNR and vice versa. There are three ways to achieve this:



  • Changing the slice thickness. In Figure 7.16, voxel size is altered by halving the slice thickness from 10 to 5 mm. Doing so halves the voxel volume from 1000 to 500 mm3 and hence halves the SNR.
  • Changing the image matrix. The image matrix is the number of pixels in the image. It is identified by two numbers: one denotes the number of pixels in the frequency direction (usually the long axis of the image), the other one the number of phase pixels (usually the short axis of the image). Look at Figures 7.17 and 7.18 where the phase matrix increases from 128 (Figure 7.17) to 256 (Figure 7.18). As the FOV remains unchanged, there are smaller pixels and therefore voxels in Figure 7.18 than in Figure 7.17. Therefore, as the voxel volume halves, the SNR also halves.
  • Changing the FOV. Look at Figures 7.197.21. The FOV halves, which halves the pixel dimension along both axes. Therefore, the voxel volume and SNR decrease to one quarter of the original value (from 1000 to 250 mm3). Comparing Figure 7.20 with Figure 7.21, it is evident that SNR significantly decreases in Figure 7.21. Depending on the area under investigation and the receiver coil, it is sometimes necessary to take steps to increase SNR when using a small FOV especially in conjunction with a large coil.
Image described by caption and surrounding text.

Figure 7.15 Voxel volume and SNR (spin numbers are not representative).

Diagram shows 3D representation of image matrix 4 x 4, slice thickness is 10 mm, and FOV as 40 mm and each slice having dimension as 10 mm on all sides and in another for 5 mm width.

Figure 7.16 SNR vs slice thickness.

Image described by caption and surrounding text.

Figure 7.17 Sagittal brain using 128 phase matrix.

Image described by caption and surrounding text.

Figure 7.18 Sagittal brain using 256 phase matrix.

Diagram shows 3D representation of image matrix 4 x 4, slice thickness is 10 mm, and FOV as 40 mm and each slice having dimension as 10 mm on all sides and in another for 5 mm length and breadth.

Figure 7.19 SNR vs FOV.

Image described by caption and surrounding text.

Figure 7.20 Sagittal brain using a square FOV of 240 mm.

Image described by caption and surrounding text.

Figure 7.21 Sagittal brain using a square FOV of 120 mm.

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