The number of contiguous slices will always be a multiple of 2. Each year seems to bring a new application of MRI or a new pulse sequence which opens up new imaging opportunities with MRI. This section addresses some of these techniques. Each of the entries is not covered in depth due to space limitations. The reader is directed to the cited literature references for more detailed information.
Volume imaging is the acquisition of magnetic resonance data from a volume rather than a tomographic slice. It can be thought of as collecting several contiguous slices through a region of imaged object.
The number of contiguous slices will always be a multiple of 2.
The actual timing diagram for a volume imaging pulse sequence looks like this.
There is a volume selection RF pulse
and gradient
which rotates only those spins in the imaged volume of the object. This combination of pulses is equivalent to a slice selection combination except the slice thickness can be 10 or 20 cm. The volume selection pulses are followed by a phase encoding gradient in dimension 1
and another one in dimension 2.
Each is varied between a maximum and minimum value, just as all the phase encoding gradients have been. The two gradient pulses are applied at the same time and are cycled through all possible combinations.
The frequency encoding gradient has its dephasing negative lobe to cause the spins to be in phase at the center of the acquisition window.
The frequency encoding gradient is applied
and a signal recorded, just as it has been in all the previous sequences.
The imaging time is equal to the product of the TR value times the number of phase encoding steps in dimension 1 times the number of steps in dimension 2. Because of this large value, a gradient recalled echo sequence is typically used for volume imaging.
Angiography is the imaging of flowing blood in the arteries and veins of the body. In the past, angiography was only performed by introducing a x-ray opaque dye into the human body and making an X-ray image of the dye. This procedure produced a picture of the blood vessels in the body. It however did not produce an image which distinguished between static and flowing blood. It was therefore a less than adequate technique for imaging circulatory problems. Magnetic resonance angiography (MRA) on the other hand produces images of flowing blood. The intensity in these images is proportional to the velocity of the flow. There are three general types of MRA, time-of-flight
,
phase contrast
angiography, and contast enhanced angiography. They are described next.
Time-of-flight angiography can be performed in several ways. One method uses a spin-echo sequence where the slice selective 90o and 180o pulses have different frequencies. The 90o pulse excites spins in one plane. The 180o pulse excites spins in another plane. In the absence of flow, no signal is seen because no spins experience both the 90o and 180o pulses. In the presence of flow and the correct TE time, blood from the 90o plane flows into the 180o plane and produces an echo.
Recall the following flow artifact description from
Chapter 11.
When the blood experiencing the 90o
pulse does not experience the 180o
pulse, no echo is observed. If the slice location of the 180o
pulse is now changed to match the location of the blood which experienced the 90o
pulse only that blood will contribute to the echo signal.
Phase contrast angiography is a little more complicated. The first new concept which you need to understand is that of a bipolar magnetic field gradient (GBP) pulse. A bipolar gradient pulse is one in which the gradient is turned on in one direction for a period of time then turned on in the opposite direction for an equivalent amount of time. A positive bipolar gradient pulse has the positive lobe first
and a negative bipolar gradient pulse has the negative lobe first.
The area under the first lobe of the gradient pulse must equal that of the second.
A bipolar gradient pulse has no net effect on stationary spins. Spins which have a velocity component in the direction of the gradient will be effected by the bipolar gradient pulse.
For example, a stationary spin exposed to the first lobe of the bipolar gradient pulse will acquire a phase in radians given by
A = 2 
g 
x GBP dt B = -2
g
x GBP dt
from the second lobe. If GBP of the two lobes are equal and the positions are equal during the two pulses the phase acquired from the A lobe equals that from the B lobe.
If this bipolar gradient pulse is placed in any one of the imaging sequences, in addition to the other gradients, it will not effect the image since all we have done is imparted a phase shift to the moving spins. Since an image is a magnitude representation of the transverse magnetization there is no effect. However if two imaging sequences are performed in which the first has a positive bipolar gradient pulse and the second a negative bipolar gradient pulse, and the raw data from the two is subtracted, the signals from the stationary spins will cancel and the flowing blood add. Look at the animation to convince yourself of this. A positive bipolar gradient pulse will have this effect on stationary and flowing spins, compared to a reference spin experiencing no gradient.
A negative bipolar gradient pulse will have this effect on the same stationary and flowing spins.
If the vectors (and hence signals) from the positive and negative bipolar gradient pulses are subtracted, the vectors from the stationary spins cancel and the moving spins have a net magnitude.
The net effect is an image of the flowing spins. From this animation it is easy to see that for optimum signal, you would like the vectors from the fastest flowing blood to acquire 90o of phase from each bipolar gradient pulse. Spins with lesser flow rates will acquire lesser phase shifts. The direction of the bipolar gradient yields signal from only those spins with a component in that direction.
A pulse sequence for one phase encoding gradient step of a phase contrast angiography sequence looks like this. Signals from the two parts are subtracted and used to produce that phase encoding line of the raw data.
Here are two examples of MRA images. The first is coronal projection of the flow in the head.
The second is an axial projection through the brain.
Contast enhanced angiography is based on the difference in the T1 relaxation time of blood and the surrounding tissue when a paramagnetic contrast agent is injected into the blood. This agent reduces the T1 relaxation times of the fluid in the blood vessels relative to surrounding tissues. When the data is collected with a short TR value, the signal from the tissues surrounding the blood vessels is very small due to its long T1 and the short TR. Images are recorded of the region of interest with rapid volume imaging sequences. The high quality of images from contrast enhanced MR angiography has made MRI the modality of choice for angiography.
Diffusion imaging can be performed in a manner identical to the phase contrast angiography sequence. The major difference is that the gradients must be increased in amplitude so as to image the much slower motions of molecular diffusion in the body. The reader is directed to the second half of the previous section (phase contrast angiography) for details of this sequence.
There several imaging techniques that may be used to shorten the image acquisition time without shortening TR. Two techniques, fractional Nex imaging and fractional echo imaging, are presented in this section, while others will be presented in subsequent sections in this chapter. Before fractional Nex or fractional echo imaging can be understood, it is first necessary to examine a relationship between the data in different halves of k-space.
K-space data is made up of complex data representing the Mx and My components of magnetization. The complex data in the right half of k-space is the complex conjugate of the data in the left half of k-space. Similarly, the data in the top half of k-space is the complex conjugate of the data in the bottom half of k-space.
Fractional Nex ImagingFractional Nex imaging takes advantage of this complex conjugate relationship between the top and bottom halves of k-space to reduce the number of phase encoding steps. In half-Nex imaging, phase encoding steps +8 through -128 of +128 to -128 are recorded. Steps -128 through 0 are generated from the complex conjugate relationship between the halves of k-space. Phase encoding steps -8 through 0 are recorded to assure the center of k-space is at 0 and there is a smooth transition between the halves. Fractional Nex imaging sequences use Nex values between Nex=1 and Nex=1/2. Because fewer data points are collected in fractional Nex imaging, the signal-to-noise ratio becomes poorer as Nex is decreased. The advantage of fractional Nex imaging is that an image can be recorded faster than with Nex=1 but with the same contrast between the tissues as in the Nex=1 case.
Fractional Echo ImagingFractional echo imaging is similar to fractional Nex imaging in that the complex conjugate relationship between the left and right halves of k-space is used to shorten the echo time in an imaging sequence. If the entire echo does not need to be recorded, the minimum TE value is decreased and more signal may be achieved. In some instances, shorter acquisitions times may result.
A fast spin echo imaging sequence is a multi-echo spin-echo sequence where different parts of k-space are recorded by different spin-echoes. For example we might have a four echo spin-echo sequence with a TE of 15 ms. The k-space will be divided into four sections. The first echo is used to fill the central part, lines 96-160, of k-space. The second echo is used for lines 64-96 and 160-192. The third echo fills lines 32-64 and 192-224. The last echo fills lines 1-32 and 224-256 of k-space.
There are some problems with the steps between the parts of k-space, but since there is little data in these regions the steps can be corrected for. The benefit of the technique is that a complete image can now, as was shown in this example, be recorded in one fourth of the time.
Chemical shift imaging is the production of an image from just one chemical shift component in a sample.
For example if the object being imaged is composed of water and fat hydrogens, each with a different chemical shift, a chemical shift image would be an image of either the fat or water in the object. Since most routine chemical shift imaging is done to suppress the fat signal it is often times referred to as fat suppression imaging. There are several methods of performing chemical shift imaging, the two which are covered here the inversion recovery method and the saturation method.
In the inversion recovery method an inversion recovery imaging sequence is used and the TI time is set to T1ln2 where T1 is the spin-lattice relaxation time of the component one wishes to suppress. For fat suppression that component is fat, for water suppression it is water. This technique only works when the T1 values for the two components are different.
In the saturation method a frequency selective saturation pulse is applied before the standard imaging pulses of a sequence, for example a spin-echo sequence. The saturation pulse sets to zero the magnetization from the component we wish to suppress. When the standard imaging sequence follows it detects no signal from the suppressed component. The accompanying animation shows an RF timing diagram for the sequence.
The saturation pulse consists of the frequency selective pulse which causes the Z magnetization for a specific chemical shift to be zero.
In the case of a fat saturation sequence, this chemical shift compound is fat. This pulse is followed by a dephasing gradient to force the transverse magnetization from this chemical shift component to zero.
The saturation pulse is followed by, in this example, a spin-echo sequence.
This technique works best when the T1 for the suppressed sequence is long compared to the time between the saturation pulse and the spin echo sequence.
Echo planar imaging is a rapid magnetic resonance imaging technique which is capable of producing tomographic images at video rates. The technique records an entire image in a TR period.
To understand echo planar imaging it is helpful to understand the concept of k-space. A magnetic resonance image is referred to as image space. Its Fourier transform is referred to as being k-space. In magnetic resonance imaging, k-space is equivalent to the space defined by the frequency and phase encoding directions. Conventional imaging sequences record one line of k-space each phase encoding step. Since one phase encoding step occurs each TR seconds the time required to produce an image is determined by the product of TR and the number of phase encoding steps. Echo planar imaging measures all lines of k-space in a single TR period.
A timing diagram for an echo planar imaging sequence looks as follows.
There is a 90o slice selective RF pulse
which is applied in conjunction with a slice selection gradient.
There is an initial phase encoding gradient pulse
and an initial frequency encoding gradient pulse
to position the spins at the corner of k-space. Next there is a 180o pulse.
Since the echo planar sequence is typically a single slice sequence, the 180o pulse need not be a slice selective pulse. The phase and frequency encoding directions are next cycled so as to traverse k-space. This is equivalent to putting 128 or 256 phase and frequency encoding gradients in the usual period when the echo is recorded. If we zoom into this region of the timing diagram it will be clearer.
You can see that there is a phase encoding gradient, followed by a frequency encoding gradient, during which time a signal is recorded. Next there is another phase encoding gradient followed by the reverse polarity frequency encoding gradient during which time a signal is recorded.
Looking at the k-space trajectory map at the same time as we are zoomed into the phase and frequency encoding gradient area we can see how the gradients trace out k-space.
The rate at which k-space is traversed is so rapid that it is possible, depending on the image matrix, to obtain 15 to 30 images a second. This is video rate acquisition.
When echo planar imaging was first developed, it was thought that echo planar imaging would have its greatest impact in providing real time magnetic resonance images. Its greatest application appears to be in the area of functional MRI of the brain. Functional imaging is imaging which relates body function or thought to specific locations in the brain.
During brain activity there is a rapid momentary increase in the blood flow to the specific thought center in the brain. For example when you move your right index finger there is a rapid momentary increase in the circulation of the specific part of the brain controlling that movement of the finger. The increase in circulation means an increase in oxygen which is paramagnetic which affects the T1 and T2 of the local brain tissues. The difference in T1 and T2 relative to surrounding tissues causes a contrast between the tissues.
Nuclear magnetic resonance (NMR) spectroscopy in a clinical setting is the study of the specific resonance frequencies absorbed by a sample or tissue.
These frequencies are related to the specific molecules present and can therefore be used to assess the disease state of a tissue.
It is relatively easy to obtain an NMR spectrum of a bulk sample, but this is not very useful clinically.
To improve the clinical utility of a spectral information, the spectrum must be from a small known location.
Several techniques have been reported in the literature to obtain spectra from small regions in a sample.
It is more difficult, and more clinically useful, to obtain an NMR spectrum from every voxel in an imaged object.
Several methods have also been proposed for this form of spectroscopy.
Examples of both types of techniques are summarized below.
For additional techniques, the reader is directed to a good review by Matson and Weiner
.
Deconvolution techniques
are used to produce images of specific chemical components when some knowledge of composition of the sample and the NMR spectra of the components is known.
In general, the techniques have limited utility.
To understand them, consider the following one-dimensional imaging example.
A sample is composed of two components, A and B, with concentrations CA and CB along x.
The NMR spectrum of A has two absorption peak, and that of B has one.
The image of CA+CB as a function of x is I(x).
Notice that the image of the right spectral peak of A overlaps with that of B.
Because the image from the left spectral peak is clearly defined, the overlap my be eliminated by subtraction.
An images of A and B remain.
Surface coil techniques
are limited to regions near the surface of the imaged object.
These techniques use the surface coil to produce the B1 magnetic field as well as detect the signal from tissues adjacent to the coil.
The B1 magnetic field near a surface coil drops off as the distance from the coil increases.
Therefore, the rotation angle of the spins decreases as the distance from the surface coil increases.
If a high power RF pulse is applied to a surface coil, a large range of rotation angles will be obtained.
Those regions experiencing an integer multiple of a 180 degree rotation will contribute no signal, while those experiencing odd multiples of a 90 degree rotation will see a maximum rotation and contribute most to the signal.
The following is an example of the intensity variation seen in a sphere of water imaged with a coil placed off to one side.
Therefore, some localization of a spectrum can be obtained, but unfortunately the rotations of 270
Bo field spoiling in this case can be accomplished by ferromagnetic materials
or a grid of small electromagnetics
.
The variation in the Bo magnetic field over a region where a spectrum is to be recorded needs to be appreciably less than the line width of the spectral lines if no spectral distortion is to be minimized.
When the variation in the Bo is much greater than the line width, spectral lines are broadened. In extreme cases this broadening can prevent a line from being visible.
The key to sensitive point techniques is to broaden the spectral lines from those regions where signal is not desired, and not distort lines from regions where they are wanted.
The magnetic field distribution in the animation window will achieve this goal.
In the presence of this one-dimensional gradient, only those spins experiencing the uniform magnetic field will produce a signal.
This uniform Bo region can be moved around by changing currents through the coils producing the Bo function.
To record the NMR spectrum from a specific region in a three-dimensional object, a Bo(y) and Bo(z) similar to the (x) shown will be needed.
This form of spectroscopic technique is also known as topical magnetic resonance (TMR)
.
Slice Selective Techniques
Let's examine one technique using a multi-echo sequence.
A slice selective RF pulse is applied in conjunction with a X magnetic field gradient. This excites spins in an YZ plane.
A 180o slice selective RF pulse is applied in conjunction with a Y magnetic field gradient. This rotates spins located in an XZ plane.
A second 180o slice selective RF pulse is applied in conjunction with a Z magnetic field gradient. The second 180o pulse excites spins in a XY plane.
The second echo is recorded as the signal. This echo represents the signal from those spins in the intersection of the three planes.
Fourier transforming the echo produces an NMR spectrum of the spins located at the intersection of the three planes.
By prudent choice of the X, Y, and Z gradients the signal voxel can be positioned anywhere in the imaged object.
This technique is referred to as point-resolved spectroscopy and given the acronym PRESS.
Several other approaches have appeared in the literature.
The stimulated echo acquisition mode (STEAM)
and selected volume excitation using stimulated echoes (VEST)
techniques are similar to PRESS except a 90-90-90 pulse sequence is used instead of a 90-180-180 sequence.
Image-selected in vivo spectroscopy (ISIS)
uses three orthogonal slice-selective 180 degree pulses followed by a 90 degree pulse.
The FID is collected after the 90 degree pulse.
The three slice-selective 180 degree pulses are applied in specific combinations and the FIDs added or subtracted to produce a spectrum.
The final technique in this category is depth-resolved surface spectroscopy (DRESS)
.
It consists of a single slice-selective 90 degree pulse followed by a rapidly applied gradient reversal pulse.
Spectroscopic imaging techniques are those that allow the scientist to record an NMR spectrum for each voxel in an image.
The data from these techniques is generally three-dimensional (spatial-spatial-spectral) and therefore may be displayed as spectra for individual voxels or as images of a specific chemical shift.
The easiest Spectroscopic Imaging Technique to understand is based on the 3-D or volume imaging technique, described earlier in
this chapter,
with a few modifications.
The RF pulse is volume selective and the readout gradient (Gf) is turned off.
The gradients labeled Gs and Gf are cycled through their range of values to record spectra from all points in the spatial-spatial domain.
One additional spectroscopic imaging technique worth mentioning, especially because of its educational value, is spiatial-spatial-spectral imaging based on the backprojection.
Consider the following example using a one-dimensional sample to produce a spatial-spectral image.
Assume the one-dimensional sample
of length D has an NMR spectrum of width W with the following chemical shift components
.
Therefore, a spatial-spectral domain can be defined with the following data.
The distribution of signal along the x axis can be imaged using a large one dimensional magnetic field gradient applied along x.
This is equivalent to taking the projection of the data in the spatial-spectral domain onto the frequency axis.
A spectrum of the signals present in all three samples can be recorded by applying a homogeneous Bo field and recording the NMR signal.
This is equivalent to taking the projection of the data onto the x-axis.
The projection of the data in this spatial-spectral domain onto an axis located at an angle q
with respect to the frequency axis can be obtained by applying a magnetic field gradient, Gp, defined by the following equation.
The following pulse sequence is applied to obtain these projections.
An RF pulse is applied to rotate magnetization by 90 degrees.
The sequence is repeated i times and on each repetition the amplitude of the projection gradient, Gp, takes on a different value
such that qi in the above equation takes on evenly spaced values between 0 and 2p.
The projection data at the i different angles is backprojected to obtain an image of the NMR signal in the spatial-spectral plane,
as was done in
Chapter 6
for spatial-spatial imaging.
This technique has been implemented on a clinical imager to produce 1H spatial-spatial-spectral images.
The images in the animation window
are of an axial slice through the medial tibia.
The q = 90o image is a TR/TE=1000/35 ms spin-echo image.
The remaining images represent those chemical shift components in the indicated ranges.
The most interesting NMR spectral components in the body are those of metabolites. The concentration of most metabolites is typically orders of magnitude less than that of the water or fat signal in tissues. Therefore, the 1H NMR signals from water and fat must be suppressed when performing 1H spectroscopy of metabolites.
A chemical contrast medium is a substance which is introduced into the body to change the contrast between the tissues.
The contrast is changed by varying the T1 and T2 of the tissues. The T1 and T2 values are changed by changing the number of fluctuating magnetic fields near a nucleus. A typical chemical contrast media is a complex of a paramagnetic metal ion such as gadolinium (Gd). The paramagnetic field creates many oscillating fields as it tumbles through a water environment. Unfortunately, many paramagnetic metal ions are toxic. To lessen their toxicity, these metal ions are typically complexed with other molecules or ions to prevent them from complexing with molecules in the body. Some typical contrast agents can be found in the table below.
| Contrast Agent | Application |
|---|---|
| Gd-EDTA |
|
| Gd-DTPA |
central nervous system |
| Gd-DOTA |
|
| feric ammonium citrate | stomach and upper small intestine |
| mangafodipar trisodium |
lesions of the liver |
| gadodiamide |
central nervous system |
After the injection of a contrast agent into a tissue, the concentration of the contrast agent increases then starts to decrease as it is eliminated from the tissues. In general, a contrast enhancement is obtained by one tissue having a higher affinity or vascularity than another. Most tumors for example have a greater Gd uptake than the surrounding tissues, causing a shorter T1 and a larger signal.
Another class of contrast agents is one that is activated by the physiology of the body or a property of a tumor. Two such properties are pH and calcium ions. The animation window displays a contrast agent that is sensitivie to calcium ions.
The -COO- groups have a higher selectivity for Ca2+ than Gd3+. As the calcium ion concentration increases, the -COO- groups preferentially bind to Ca2+ and expose more water molecules to the Gd3+ and their fluctuating magnetic fields.
Magnetization transfer contrast is a new method of increasing the contrast between tissues by physical rather than chemical means.
For this technique to be effective, there must be at least two spin systems in the imaged anatomy which are capable of exchanging energy between themselves and one of the systems must have a much shorter T2 than the other system.
The pulse sequence looks very similar to the fat saturation imaging sequence described earlier in this chapter.
A saturation pulse
is applied with a frequency approximately 1 kHz from the center frequency. The saturation pulse is followed by a gradient-echo or spin-echo sequence.
The two spin systems could be protein and water. The protein has a very short
T2 relative to the water T2. Because of the inverse relationship between
T2 and the spectral linewidth, the NMR spectra of these two spin systems would have a very broad peak from the protein and a very narrow peak from the water.
The signal from the protein will therefore not be visible in the image due to its broad linewidth which causes its signal to be spread out over the entire image. Applying the saturation pulse
1 kHz away from the center of these peaks could directly saturate the protein spin system and not the water. Any water molecules in contact with the protein might be capable of exchanging magnetization with the protein. Therefore saturating the protein will affect the signal of the water and the contrast between water in contact with the protein and not.
One way to picture magnetization transfer is to think of the water and protein spin systems as being energy reservoirs.
In this picture there is a protein reservoir which is connected to a water reservoir,
and another unconnected water reservoir.
Energy can be placed in any one of the energy reservoirs and it will return to the lattice or surrounding molecules via spin-lattice relaxation.
Energy which is placed only in the protein reservoir by the frequency selective saturation pulse will influence the energy of the water spin system which is connected to it.
If a pulse sequence is used to probe the magnetization of the two water spin systems while there is still energy in the protein connected water system, the protein connected water system will produce image intensity as if a short TR was used. The unconnected water system will produce image intensity as if a long TR was used. There will now be contrast between the two types of water, even if the
T1
values for the two types of water are equivalent.
The amount of noise in an image is related to the sampling frequency of the FID or echo. The higher the sampling rate the more noise in the image. Similarly, lowering the sampling frequency allows less noise in the image. In the interest of improving the signal-to-noise ratio in an image, it is advantageous to use the smallest possible sampling rate. Since the sampling rate, fs, is related to the field of view (FOV), as seen in Chapter 7, the frequency encoding gradient, Gf, must be lowered proportionately to the sampling frequency in order to keep the FOV constant.
Here is what the timing diagram would look like for a spin-echo sequence using a fast sampling frequency
,
and a slow sampling frequency
.
There are three disadvantages associated with the use of a slower sampling frequency.
1. An increase in the chemical shift artifact. (See Chapter 11.)
2. A loss of contrast.
3. A restricted range of echo times, TE.
Here are two axial images through the human head at the level of the orbits. One image was recorded with a 16 kHz bandwidth and the other a 3 kHz bandwidth.
.
Notice that in the 3 kHz image there is a shift in the fat signals toward the anterior direction and a loss of contrast. The range of usable TE times in a spin-echo sequence is restricted with variable bandwidth imaging because, as the sampling rate decreases, the sampling window increases.
.
In imaging applications where these three disadvantages do not matter, and an improved signal-to-noise ratio is needed, variable bandwidth imaging can be very beneficial.
The spin-lattice relaxation time (T1), spin-spin relaxation time (T2), and the spin density (r) are properties of the of the spins in a tissues. The value of these quantities change from one normal tissue to the next, and from one diseased tissue to the next. They are therefore responsible for contrast between tissues in the various image types described in Chapter 7 and Chapter 8.
There are several methods of calculating T1, T2, and r values. These methods are applied to individual pixels to produce a calculated T1, T2, or r image. The smaller the voxel corresponding to a pixel, the more likely the T1, T2, and r values are to represent values for a single tissue. The larger the voxel, the more likely the calculated values are to represent that of a combination of tissue components.
The calculation of T1, T2, or r starts with the collection of a series of images. For example, if you wish to produce a
T2 image, a spin-echo pulse sequence is used and a series of images are recorded with varying TE.
The signal for a given pixel can be plotted for each TE value and the best fit line from the spin-echo equation drawn through the data to find T2.
A T1 image can be created from the same pulse sequence using a series of images with varying TR.
The signal for a given pixel can be plotted for each TR value and the best fit line from the spin-echo equation drawn through the data to find T1.
The spin density can be calculated once T1 and
T2 are found using the spin echo signal equation and any spin echo signal.
The procedures just outlined will produce
T1, T2, or r images, but are not the most efficient or accurate. The reader is directed to the scientific literature for more appropriate methods.
,
Tissue classification, or image segmentation as it is also called, is the identification of tissues in a magnetic resonance image. The classification is based on a property of the tissue in the image. For example, in this spin-echo image
,
where cerebral spinal fluid (CSF) and gray matter are bright compared to other tissues, the intensity of the pixel could be used to classify the tissue as CSF and gray matter or other tissues. The histogram and look up table for this image looks like this.
Typically we would use a linear relationship between data value and pixel intensity. Furthermore the red, green, and blue content of each pixel would always be the same so as to get shades of grey for the pixels. We can segment the CSF and gray matter from other tissues in this example by modifying the lookup table such that the green and blue components of a pixel are turned off for data values greater than 865.
This procedure will create red CSF and gray matter pixels.
The image is therefore segmented into tow classes of tissues: (1) gray matter and CSF, and (2) tissues other than gray matter and CSF.
The segmentation process is done with the aid of computer algorithums. These algorithms can segment with more advanced logic than the simple greater than a given data value example given above. Many different types of images, or spectral regions, can be used to segment the tissues. Some of the possible spectral regions include: T1, T2, and r weighted; pure T1, T2, and r; angeography; diffusion; chemical shift; and functional images. Some of these images are more difficult to work with. Images which show variations in sensitivity of an imaging coil can not be used because segmentation algorithms can not destinguish between an intensity variation caused by the imaging coil sensitivity and the tissue. Calculated T1, T2, and r images are easier to work with because they do not show the variation in intensity caused by variations in imaging coil sensitivity.
In the above example it was not possible to segment gray matter from CSF because the two tissues have similar intensities in the spin-echo image. The more independent spectral regions we work with the easier it is to segment tissues. For example, segmenttion of the tissues in the brain can be accomplished with calcualted T1
,
T2
,
and r
images of the human brain.
These images are used to create a three-dimensional histogram. Similar tissue types display clusters in the histogram.
We can assign pixels in a given range of T1, T2, and r values a particular color. The resultant image displays the segmented tissues.
One additional segmentation example, based on morphology or texture in an image, is presented here.
High resolution magnetic resonance images of the wrist are taken with a 0.7 mm slice thickness, an 8 cm FOV, and a 256x256 matrix. These images show the trabecular structure in the wrist bones.
These images are used to train an algorithm to identify the different types of bone diseases by comparing their morphology. The resultant algorithm characterizes the trabecular structure and classify it based on known properties of diseased bones. The classified image
depicts normal as red and those regions which have deminished trabecular or osteoporotic (green), cystic (blue), and sclerotic (light blue) properties.
Hperpolarized noble gas imaging is the imaging of the NMR signal from a noble gas, such as 129Xe or 3He.
Xenon is used as an anesthetic, so much is known about the physiological effects of Xe, which in turn makes the imaging of hyperpolarized 129Xe easier. 129Xe is a spin 1/2 nucleus with a natural abundance of 26.44% and a gyromagnetic ratio of 11.8 MHz/T.
Hyperpolarized 129Xe is produced by a multistep process. First, Rb metal vapor is optically exciting with a 795 nm circularly polarized laser beam.
The excited Rb electron looses energy by a spin exchange transfer to the Xe nucleus during Xe-Rb collisions.
This process results in a net 129Xe nuclear magnetization of approximately 105 times that at equilibrium. The larger net magnetization means a larger signal is possible, and hence imaging of a gas is possible. The T1 of the 129Xe is approximately 15 to 40 s in tissues. Because we are dealing with a hyperpolarized gas with a long T1 value, all signal would be lost after the application of a 90o pulse. Therefore a gradient recalled echo with a 5o rotation angle is typically used to preserve signal throughout the acquisition period.
Here is an example of a spin-echo image of a rat brain.
Superimposed on this image is a 32x32 pixel false color image of the 129Xe NMR signal in the brain from breathing hyperpolarized 129Xe for approximately 40 s.
This study indicated that the 129Xe signal arose from within the brain and that there was a reduced Xe concentration in the cerebellum.
Copyright © 1996-2000 J.P. Hornak.
All Rights Reserved.