NOTES ON INTRODUCTION TO BIOMEDICAL IMAGING

BY ANDREW WEBB

SANDRO NUNES
TCNICAS DE IMAGIOLOGIA Prof. Patrcia Figueiredo

1. X-Ray Imaging and Computed Tomography

1.1. General Principles of Imaging with X-Rays X-Ray Imaging transmission-based technique in which X-rays from a source pass through the patient and are detected either by film or an ionization chamber on the opposite side of the body. Contrast in the image between different tissues arises from differential attenuation of Xrays in the body.

Computed Tomography the source and detectors rotate together around the patient, producing a series of one-dimensional projections at a number of different angles. This data is reconstructed to give a two-dimensional image. The x-ray source is collimated to interrogate a thin slice through the patient. It has a very high spatial resolution (~1mm).

1.2. X-ray Production The x-ray source is the most important component in determining the image quality:

1.2.1. X-ray Source The main structure of the X-ray source (also called tube) is shown below:

Production of X-rays involves accelerating a beam of electrons to strike the surface of a metal target. The X-ray tube has 2 electrodes: a negatively charged cathode (electron source) filament of coiled tungsten wire - and a positively charged anode (metal target). An electric current passes through the cathode, heating it (~2200C) and causing electrons to move away from the metallic surface (thermionic emission). The tube potential causes the free electrons to accelerate towards the anode. Since the spatial resolution is determined by the effective focal spot size, the cathode is designed to produce a tight, uniform beam of electrons. To do this, a negatively charged focusing cup is placed around the cathode:

Moreover, the anode is beveled in order to produce a small effective focal spot size:

= ()

The electrons striking the anode lose their kinetic energy, which is converted into X-rays. The anode must be made of a metal with high melting point and good thermal conductivity (usually tungsten). 1.2.2. X-ray tube current, tube output and beam intensity Tube potential: 15-150kV (rectified alternating voltage, characterized by the maximum value, kilovolts peak or accelerating voltage - kVp). Tube current: 50-1000mA (depends on kVp) Tube output: tube current tube potential. We need a high tube output in order to decrease the exposure time. It depends on: kVp Vacuum in the tube (reduces interactions between electrons and molecules and increase electrons velocity)

Tube power rating: maximum power dissipated in an exposure of 0.1s. Limited by anode heating, which can be reduced by causing it to rotate at roughly 3000 rpm. Intensity of the X-ray beam: power incident per unit area (J/m2). It depends on the number ( current) and energy of the X-rays ( kVp2)

1.2.3. The X-Ray Energy Spectrum The output of the source is shown below:

Electrons striking the anode generate X-rays by 2 processes: 3

-Bremsstrahlung: generated when an electron is deflected by the tungsten nucleus, losing kinetic energy which is emitted as an X-ray. The Bremsstrahlung radiation has a wide range of energies, with a maximum corresponding to the case where all the electrons kinetic energy is converted into a single X-ray (with energy kVp). It is characterized by a linear decrease in X-ray intensity with increasing X-ray energy, however many low energy X-rays are absorbed within the tube (additional external filters are used because low energy X-rays would be incapable of passing through the patient, adding to the dose unnecessarily).The efficiency, , of the Bremsstrahlug is given by: = () k constant related to the target material Z atomic number of the target material -Characteristic Radiation: shown as sharp peaks. It is emitted when an accelerated electron hits an electron in the inner shell of the tungsten atom, causing it to be ejected. An electron from the outer shell fills the hole, which causes a loss in potential, emitted as an X-ray. Only happens for incoming electrons with energy > 70 keV.

1.3. Interactions of X-rays with tissue The X-ray can be classified according to the type of interaction: Primary interaction: passes through the body with no interaction. Secondary interaction: scattered radiation, whose trajectory between source and detector was altered. Caused by coherent and Compton scattering. Absorbed radiation: absorbed radiation that does not reach the detector. Caused by photoelectric interactions. 1.3.1. Coherent Scattering Also called Rayleigh scattering. The radiation is absorbed by the tissues atoms and then emitted in a random direction. Reduces the quantity of X-rays reaching the detectors and alters their trajectory. 1.3.2. Compton Scattering Refers to the interaction between an incident X-ray and a loosely bound electron in an outer shell of an atom in tissue. A fraction of the X-ray energy is transferred to the electron, the electron is ejected and the X-ray is deflected from its original path. The difference in energy is very small, which means that this radiation is detected with approximately the same efficiency as primary radiation. Also, it does not depend on atomic number, thus it is absorbed the same way in different tissues. 1.3.3. Photoelectric Effect The energy of the incident X-ray is absorbed by an atom, with a tightly bound electron being emitted from the K or L shells. A second electron with a higher energy level then fills the hole, emitting a characteristic X-ray with very low range that does not reach the detector. For an incident energy just higher than K-shell, the probability of photoelectric interactions is very high and is given by: 4

1.4. Linear and mass attenuation coefficients of X-rays in tissue Attenuation of the X-ray beam through tissue is given by:

linear attenuation coefficient The value of is given by:

Contributions of photoelectric interactions dominate at lower energies, whereas Compton scattering is more important at higher energies. X-ray attenuation is often characterized by mass attenuation coefficient which is equal to the linear attenuation coefficient divided by the density of the tissue. We can see from the graph below (right) that, at higher energies, little differentiation is possible because the number of photoelectric interactions decreases.

HVL (half value layer) thickness of tissue that attenuates half of the X-ray intensity parameter commonly used to characterize X-ray attenuation.

1.5. Instrumentation for planar X-ray Imaging The remaining components of an X-ray imaging system are: 1.5.1. Collimators Restricts the dimensions of the beam in order to match the desired field of view (FOV). The collimator consists of sheets of lead, which can be slid over one another to restrict the beam in either one or two dimensions. Beam dimensions higher than the FOV increase patient dose unnecessarily and the number of Compton-scattered X-rays.

1.5.2. Antiscatter grids Even using a collimator, secondary radiation can represent between 50% and 90% of the x-rays reaching the detector. Therefore, it is placed an antiscatter grid between the patient and the X-ray detector. This grid consists of strips of lead foil interspersed with aluminum as a support, with the strips oriented parallel to the direction of the primary radiation. Two important properties are the grid ratio and the strip line density:

There is a tradeoff between reduction of the scattered radiation and the patient dose that must be delivered to give the same amount of detected X-rays. It can be characterized by the Bucky factor F:

1.5.3. Intensifying screens These screens convert the X-rays into light, to which the film is much more sensitive. This is done by placing a phosphor layer before the film. The greater the thickness, the higher the SNR and lower the patient dose. However, it also worsens the spatial resolution as it increases the uncertainty in the position of the original X-ray.

1.5.4. X-ray film The presence of lighter regions is due to a chemical reaction of silver particles in the film to metallic silver. This reaction is reduced in the areas where light hits the film, so the degree of blackening depends on the intensity and time of the light hitting a specific area. This blackening is measured by the optical density (OD):

Ii intensity of the light incident on. It intensity of the light transmitted through the X-ray film.

1.6. X-Ray image characteristics 1.6.1. Signal to Noise Ratio SNR is proportional to the statistical variance in the number of X-rays per unit area (quantum mottle). Since this variance is characterized by a Poisson distribution, SNR is proportional to the square root of the number of detected X-rays per unit area (N). It is affected by: X-ray tube voltage: the higher the kVp, the higher high-energy X-rays produced and thus, the number of X-rays reaching the film (1) X-ray tube current () X-ray exposure time () Intensifying screen thickness () X-ray filtration () Object thickness () Antiscatter grid ratio ()

1.6.2. Spatial Resolution The main factors affecting it are: Effective focal spotsize (2) Magnification factor () Film speed () Intensifying screen thickness ()

The resultant spatial resolution is given by:

1 2

The arrows represent what happens to SNR if we increase the value of each given factor. means higher R, the minimum distance which can be resolved, this is, worse resolution.

1.6.3. Contrast to Noise Ratio Refers to the difference in signal intensity from various regions of the body (for example the difference between the SNR of bone and soft tissue). It is affected by all the factors that affect SNR and R, in addition to: Energy of the X-ray: if high energies are used, Compton scattering dominates () FOV: for values between 10cm and 30cm, the proportion of Compton scattering reaching the detector increases linearly. After that, it is constant. Geometry of the antiscatter grid

1.7. X-ray contrast agents X-ray contrast agents are chemicals that are introduced in the body to increase image contrast. Some examples are the barium sulfate and the iodine-based X-ray contrast agents. These chemicals have a particular K-edge energy that can be used to distinguish the tissues on which they accumulate from the surroundings.

1.8. X-Ray Imaging methods The main imaging techniques that use X-rays are:

1.8.1. X-ray angiography Angiography techniques produce images that show selectively the blood vessels in the body. Iodine-based contrast agents are injected into the bloodstream to improve contrast. A related imaging technique called digital subtraction angiography consists on taking an image before the agent is administered and one after, and then compute the difference (yielding very high contrast).

1.8.2. X-ray fluoroscopy X-ray fluoroscopy is a continuous imaging technique using X-rays with very low energies (used, for example, for placement of stents and catheters). Since very low energies cause low SNR because of the quantum mottle, a fluoroscopic image intensifier (CsI:Na) is used to improve the SNR. A fluorescent screen is used to continuously monitor the area of interest.

1.8.3. Dual-Energy Imaging Technique that produces two separate images corresponding to soft tissue and bone (used for imaging the chest region). There are 2 ways of performing dual-energy imaging: Two X-ray exposures, one applied immediately after the other, with different values of kVp; Single exposure and 2 detectors. The detector placed directly beneath the patient absorbs low energy X-rays and hardens the beam detected by the second detector. Therefore, the image from the first detector corresponds to a low energy x-ray, high 9

contrast image, and that from the second, to a high energy x-ray, low contrast image. If more beam hardening is required, a copper filter can be put in front of the second detector.

1.9. Clinical applications of X-ray imaging Apart from the ones described above, there are additional applications of X-ray imaging:

1.9.1. Mammography X-ray mammography is used to detect small lesions in the breasts. It requires very high spatial resolution and CNR to detect the microcalcifications (<1mm). A low dosage is also important to avoid tissue damage (a molybdenum filter is used to remove high energies, which also improves CNR); Fast intensifying screen/film combinations are necessary to allow the use of low Kvp to optimize SNR; Large source to detector distance and small focal spot size to increase resolution.

1.10.

Computed Tomography

CT enables the acquisition of 2D thin slices, which can be obtained in order to reconstruct a 3D volume. These 2D slices are reconstructed from a series of 1 dimensional projections of the object acquired at different angles. The detectors, which are situated opposite to the x-ray source, detect the total number of x-rays transmitted through the patient, producing a one dimensional projection. The signal intensities in this projection are dictated by the two dimensional distribution of the tissue attenuation coefficients within the slice. The X-ray source and the detectors are then rotated by a certain angle and the measurements are repeated. The image is reconstructed by a process termed backprojection.

1.10.1. Scanner instrumentation The basic operation of a 1st generation scanner is shown below:

10

The image acquired this way has M X N points. The spatial resolution could be increased by using finer translational steps or the angular increments, up to a value limited by the effective x-ray focal spot size. The 2nd generation replaced the single beam by a fan beam and used multiple detectors, which reduced the scanning time. It required the development of fan-beam backprojection. The 3rd generation uses a much wider x-ray fan beam (45) and an increased number of detectors (between 512 and 768). Two collimators are used to restrict the fan beam and the slice thickness (1-5mm). The rotation covers 360. In the 4th generation detectors, a complete ring of detectors surrounds the patient. No decrease in scanning time.

1.10.2. Detectors for Computed Tomography The most common detectors in CT are xenon-filled ionization chambers (xenon has a high atomic number of 66, thus there is a high probability of photoelectric interactions). Xenon is kept at high pressure to increase the number of interactions. X-rays transmitted through the body ionize the gas in the detector, producing electron-ion pairs, which are attracted to the electrodes by the applied voltage. This creates a current proportional to the number of incident x-rays.

1.11.

Imaging processing for computed tomography

The image reconstruction process is shown in the figure below, where only 2 projections are acquired:

11

The measured intensities can be expressed by:

Simple matrix inversion is not sufficient to determine the attenuation coefficients because its computational burden is very high for large data and because the presence of noise in the projections cause instability in the inversion techniques.

1.11.1. Processing data corrections Beam hardening: there is the need to use algorithms that account for this factor because it causes the effective linear attenuation coefficients to decrease with distance from the source; Imbalances in detectors sensitivity: if not corrected, a ring or halo artifact can appear. It can be solved by calibrating the detectors: an object with a spatially uniform attenuation coefficient is measured before the actual patient.

1.11.2. The Radon Transform and the Backprojection Techniques The mathematical basis for reconstruction of an image from a series of projections is the Radon transform. For an arbitrary function f(x,y), its Radon transform is defined as the integral of f(x,y) along a line L:

12

Each X-ray projection, (, ), can therefore be expressed as a function of the Radon transform of the object:

To obtain the reconstructed image we need to compute the inverse Radon transform by using filtered backprojection. The image is displayed as a map of the tissue CT number, defined by:

0 - linear attenuation coefficient of the tissue 1.11.3. Fan-beam reconstructions The 2nd, 3rd and 4th generation scanners use fan-beams, which are not parallel to one another, thus the backfiltered algorithms need some modifications. The simplest way is to resort the acquired data in order to form sets of parallel X-ray paths, such as S1D1 and S2D3 in the following picture:

13

This way, the standard backprojection algorithms can be used.

1.12.

Spiral/helical computed tomography

To avoid the time delay and the spatial misregistrations between slices due to the patient movement, a technique called spiral or helical CT was developed. This technique acquires data as the table moves continuously through the scanner. This allows 10 times faster scan times and the acquisition of very fast 3D vascular imaging datasets just after injection of an iodinated contrast agent, resulting in significant increase in SNR of the angiograms. In terms of instrumentation, the main difficulty is that X-rays must be produced continuously, without the cooling period. Therefore, the X-ray source must be designed to have a high heat capacity and a very efficient cooling. Moreover, detectors must be very efficient in order to reduce the tube current and alleviate the anode heating (for example, scintillation crystals, made of BGO, are used since they have a high efficiency 75-85% in converting X-rays to light). The most important acquisition parameter in helical CT is the spiral pitch:

Where d is the table feed per rotation and S is the slice thickness. p typically lies between 1 and 2, thus the radiation dose is lower than in single-slice CT. For values of p less than 1, the x-ray beams of adjacent spirals overlap, resulting in high tissue radiation dose. For values of p greater than 2, gaps appear in the data sampled along the longitudinal axis and image blurring happens due to the patient movement. Due to the helical trajectory, modifications of the backprojection reconstruction must be made. The modified algorithms use linear interpolation of data 180 apart on the spiral trajectory to estimate the data that would be obtained with a stationary table.

1.13.

Multislice spiral computed tomography

Multislice spiral CT incorporates an array of detectors in the z direction (direction of table motion). It improves efficiency by allowing higher values of the table feed to be used and, thus, lower scan times. The spiral pitch is defined slightly different:

Where Ssingle is the single-slice collimated beam width In a multislice system, the focal-spot-to-isocenter and the focal-spot-to-detector distances are shortened compared to the single-slice scanner and the number of detectors in the longitudinal direction is increased from one long element to a number of shorter elements. There are 2 types of detector arrangements: Fixed consists of 16 elements with a total length of 2 cm. Signals from sets of 4 individual elements are typically combined. Only fixed values of pitch can be chosen. 14

Adaptive consists of 8 detectors with different lengths with also a total length of 2 cm. Any pitch value can be chosen from 1 to 8.

Advantages: Shorter acquisition time Thinner slices: better spatial resolution Can get isotropic volumes

Disadvantages: Larger beam width Higher dose for the same quality Cone beam artifacts

15

2. Nuclear Medicine
2.1. General principles of nuclear medicine Nuclear medicine images the spatial distribution of radiopharmaceuticals introduced in the body. It can detect biochemical changes in tissue, serving as a diagnostic to pathological conditions such as formation of edema, tumor enlargement or metastasis, and changes in the tissue morphology. These radiopharmaceuticals, termed radiotracers, are compounds linked to a radioactive element, whose structure determines its distribution in the body. Radiation, usually in the form of -rays, is detected using a gamma camera. The following picture shows the basic principles and instrumentation involved:

Decay of the radioactive element produces -rays, which emanate in all directions. Attenuation of -rays occurs the same way as in X-rays. In order to determine the position of the source of the -rays, a collimator is placed between the patient and the detector so that only those components of radiation that have a trajectory at an angle close to 90 to the detector plane are recorded. A scintillation crystal is used to convert the -rays into light. These light photons are in turn converted into an electrical signal by photomultiplier tubes (PMTs). The image is formed by analyzing the spatial distribution and the magnitude of the electrical signals from each PMT. Planar nuclear images are characterized as having poor SNR and low spatial resolution (~5 mm), but extremely high specificity because there is no background radiation in the body.

16

3D nuclear images can be produced using the principle of tomography. A rotating gamma camera is used in a technique called single photon emission computed tomography (SPECT). The most recently developed technique is positron emission tomography (PET), which is based on positron-emitting radiotracers. This technique has a sensitivity advantage over SPECT of 2 to 3 orders of magnitude.

2.2. Radioactivity Radioactivity is an intrinsic property of particular isotopes that have unstable nuclei. The phenomenon of radioactivity refers to the process whereby various forms of radiation are emitted as result of spontaneous change in the composition of the nucleus. The stability of a nucleus is dictated by the relationship between Z (n protons) and A (n protons +neutrons). Strong neutron-neutron, proton-neutron and proton-proton forces are attractive over very short distances, whereas electromagnetic forces between protons are repulsive, and so the nucleus stability is determined by the balance between these forces. For A<50, the stable configuration corresponds to a 1:1 ratio, whereas for A>50, the ratio of neutrons increases:

The radioactivity, or activity, Q, of a radionuclide is defined as the n of disintegrations per unit time (in Curies, Ci=3.7X1010 N/s) where is the decay constant:

It can be solved to give:

The half-life corresponds to the time required for the radioactivity to drop one-half of its current value:

When calculating the time dependence of radioactivity within the body for a nuclear medicine scan, the biological half-life of the radionuclide must be considered: 17

2.3. Production of radionuclides There are 4 basic methods for producing radionuclides: Neutron capture Nuclear fission (uses fast high energy neutrons to create 99Mo 133Xe and 131I) Charged particle bombardment (uses a cyclotron to accelerate ionized hydrogen or deuterium gas and create 201Ti, 67Ga, 111In or 123I) Radionuclide generators (most common method, via on-site generator)

2.4. Types of radioactive decay Radioactive elements can decay via a number of mechanisms, of which the most important in nuclear medicine are: -particle decay -particle emission -ray emission Electron emission

The most useful radionuclides are the ones that emit -rays or X-rays because these forms of radiation can pass through tissue and reach the detector. A useful parameter to quantify attenuation is the HVL. An -particle consists of a helium nucleus. It has a tissue HVL of only a few mm and, thus, it is not directly related to nuclear medicine. This form of radioactive decay occurs mainly for radionuclides with Z>150. A -particle is an electron. Radioactive decay occurs via conversion of a neutron into a proton, with emission of a high energy -particle and an antineutrino. Kinetic energy is shared in a random manner between the -particle and the neutrino, and hence the electron has a continuous range of energies (e.g.: a 1MeV -particle has a HVL=0.4 mm while a 5MeV -particle has a HVL=4 mm). Due to their low HVL, radionuclides that produce and particles cannot be used for imaging. No radionuclide can decay solely by -ray emission, but certain decay schemes result in formation of an intermediate species that exists in a metastable state. This is the case of 99mTc, which is the most widely used:

The energy of the emitted -ray for the 99mTc is 140 keV. Below an energy of 100 keV most -rays are absorbed in the body via photoelectric interactions. Above 200 keV, -rays penetrate the thin collimator septa, thus the ideal energy range lies between 100 and 200 keV.

18

The final mechanism of radioactive decay is the electron capture (with subsequent -ray or X-ray emission), in which an orbital electron from the K or L shell is captured by the nucleus, leaving a hole that is occupied by an outer electron. This process produces characteristic X-rays. For clinical imaging, an ideal radionuclide should have a half-life short enough to limit the dose to the patient but long enough such that the radioactivity is not exhausted by the time the nuclide has distributed within the body. The radioactive decay should be via monochromatic ray emission to a stable nuclear state, without or -particle emission.

2.5. Technetium generator

99m

Tc is the most used radionuclide because:

Can be produced from an on-site generator Half-life of 6.02h Very minor -particle emission HVL of 4.6 cm

The generator consists of an alumina ceramic column with 99Mo on its surface. The 99mTc is obtained by flowing an eluting solution through the generator, which washes out the 99mTc, leaving behind the 99Mo. Typically, the technetium is eluded every 24h and the generator is replaced once a week.

2.6. Biodistribution of the technetium-based agents within the body The majority of radiopharmaceuticals uses ionic technetium (TC4+), which is bind to a metal ion according to the target organ.

2.7. Instrumentation: the gamma camera

19

The roles of each component of the gamma camera are listed below:

2.7.1. Collimator The collimator reduces the contribution from the -rays that have been scattered by eliminating those that do not travel at angles close to 90 to the collimator surface. The normal pattern consists of lead strips in a hexagonally based honeycomb geometry. The dimensions and arrangement of the lead strips determine the contribution to the overall spatial resolution:

An important fact to note is that the spatial resolution is worse for more internal organs. There are 2 general classes of collimators: High resolution (HR) high septa thickness High sensitivity (HS) low septa thickness

For both HS and HR, the fact that the septal length is much higher than the septal thickness means that the vast majority of the -rays are absorbed. The geometric efficiency, G, of the collimator is given by:

where k is a constant related to the collimator geometry.

20

Other types of collimators are also used:

Converging collimator used for imaging small organs close to the surface of the body. Pinhole collimator increases significantly the magnification and the spatial resolution of the image, but produces some geometric distortion particularly at the edges of the image. Diverging collimator reduces the size of the image. Used to image structures larger than the size of the detector.

2.7.2. The scintillation crystal and coupled photomultiplier tubes The most common -ray detector is based on a single crystal of thallium-activated sodium iodide, NaI(Ti). When a -ray strikes the crystal, electrons are ejected and lose energy in a short distance by ionizing and exciting the scintillation molecules. Deexcitation of these states occurs via emission of photons in the blue visible range. The intensity of the light is proportional to the energy of the incident -ray. The choice of crystal thickness in nuclear medicine involves the same tradeoff between spatial resolution and sensitivity as described for X-rays. Then, these light photons are detected by hexagonal PMTs, which are closely coupled to the scintillation crystal. This geometry gives efficient packing and also has the property that the distance from the center of one PMT to that of each neighboring PMT is the same (important for Anger position network). The PMTs amplify the current, which is digitized using an A/D converter. Each PMT should have an identical energy response. For planar nuclear medicine a 10% variation in uniformity is tolerated. However, for SPECT imaging this value should be less than

21

1%. In practice, calibration is done using samples of uniform and known radioactivity or, more recently, through a LED calibration source for each PMT in real time.

2.7.3. The Anger position network and the Pulse Height Analyzer By comparing the magnitudes of the currents from all of the PMTs, the location of individual scintillations within the crystal can be estimated. This calculation is done by using an Anger logic circuit. This network produces four output signals, X+, X-, Y+ and Y-, the relative magnitudes and signs of which define the location of the scintillation in the crystal. The summed signal, termed the z-signal, is sent to a pulse-height analyzer (PHA), which compares the z-signal to a threshold value to determine if it was originated by a scattered -ray (for a 99mTC scan, it corresponds to that produced by a -ray with energy 140 keV). The z-signal is accepted if its energy is inside a certain interval given by the full-width half-maximum (FWHM) of the photopeak: the narrower the FWHM, the better it is at discriminating between scattered and unscattered -rays. Typically, a 15% interval is used. The following picture shows, in the left, the presence of 2 additional peaks: one at 90 keV, due to Compton scattering and another at 111.5 keV, due the photoelectric interactions in the crystal. In the presence of a patient, the energy spectrum broadens due to Compton scattering in the patient.

If the geometric efficiency of the system is high and the injected dose of radiopharmaceutical is large, then the number of -rays reaching the scintillation crystal can exceed the recording capabilities of the system (because there is a finite recovery time for various components of the gamma camera). If scintillation events occur at time intervals less than the recovery times, then they cannot be recorded. The overall dead time of the system is defined by:

22

where N in the true count rate and n is the observed count rate. Standard gamma cameras typically have a 20% loss in the number of counts.

2.8. Image characteristics Nuclear medicine images are characterized by low SNR and poor spatial resolution, but an extremely high CNR. Postprocessing is used to increase the image SNR, although this further degrades the spatial resolution.

2.8.1. Signal to Noise Ratio The number of disintegrations per unit time fluctuates around an average value described by a Poisson distribution and, thus, SNR is proportional to the square root of the number of counts (). The factors that affect SNR are: Radioactive dose administered () comparing to X-rays, the number of counts is 10000 times lower Effectiveness of the radiopharmaceutical at targeting a specific organ () Time over which the image is acquired () limited by radioactive and biological halflives Sensitivity of the gamma camera () related to the collimator geometry and the thickness of the scintillation crystal Postacquisition image filtering () by applying low pass filters (however causes blur)

2.8.2. Spatial Resolution There are 4 major contributions to the spatial resolution of a nuclear medicine scan: Intrinsic spatial resolution of the gamma camera, excluding the collimator () reflects the uncertainty in the exact location at which light is produced in the scintillation crystal due to thickness of the crystal and the Anger position encoder (~3-5 mm). Geometry of the collimator Degree of Compton scattering () goes up with depth of the radiopharmaceutical within the body Postacquisition image filtering () Considering the first 3 terms, the overall spatial resolution is given by:

Typical values are approximately 1-2 cm and 5-8 mm for organs in depth and close to the surface of the collimator, respectively.

2.8.3. Contrast to Noise Ratio If theres no background signal (due to the fact that the radiopharmaceutical did not distribute out of the targeted area), then CNR ~ SNR. It is affected by: 23

Compton scattering () Partial volume effects () depends on R Postprocessing filtering

2.9. Single photon emission computed tomography SPECT applies tomographic principles to produce a series of 2D images. Its main image characteristics are: CNR: improved by up to a factor of 5 to 6 times because sources of radioactivity are not superimposed. R: not changed. Approximately Gaussian PSF with FWHM ~1 cm SNR: it needs about 5 times as many counts to obtain an equivalent SNR as in planar scintigraphy (~500,000 counts in brain and ~100,000 counts in myocardium)

It is the standard acquisition modality for myocardial and brain perfusion and for oncological investigations.

2.9.1. Instrumentation for SPECT

SPECT can be performed using either multidetector or rotating gamma camera systems. The former has higher sensitivity and spatial resolution but is very complex and expensive, making it a rare technique in clinic environment. The latter is preferred for routine clinical imaging because it can also be used for planar scintigraphy. Data is collected from multiple views as the detector rotates around the patient. An improvement to this setup is to increase the number of cameras in the system as the sensitivity per slice is proportional to the number of cameras. Thus, two and three camera systems are commonly used. A 360 rotation is generally needed in SPECT because the effects of -ray scatter and tissue attenuation and the dependence of the spatial resolution on the source to detector distance all mean that projections acquired at 180 to one another are not identical. A focused collimator is used to improve sensitivity, which requires increased complexity of data reconstruction. The data matrix acquired is usually 64x64 or 128x128 with first allowing better SNR and the second allowing better spatial resolution. Projections can be acquired in a stop-and-go mode or in a continuous rotation. 2.9.2. Scatter and attenuation correction In SPECT, scatter correction is needed because we do not have a well-defined -ray geometry due the lack of tight source collimation. Moreover, tissue attenuation has here a different role: in CT, the reconstructed image is an estimate of the spatial distribution of X-ray attenuation coefficients, while in SPECT the reconstructed image is an estimate of the spatial distribution of injected radiopharmaceutical. Thus, spatially dependent -ray attenuation gives rise to artifacts and need to be corrected. If tissue attenuation is not included, a measured SPECT projection can be represented as:

24

If we add tissue attenuation, then:

This equation is extremely difficult to solve analytically and, thus, it is an important part of SPECT data processing. The scatter correction is done by using a dual-energy window detection method: the main window contains contributions from both scattered and unscatterd -rays (fractional width of 20% around 140keV); the subwindow is set to a lower energy (7% around 121keV). The true primary -rays count can be calculated from:

The attenuation correction can be done using 2 different methods. The first assumes that -ray attenuation is uniform throughout the entire body. Thus assumption works well for the brain, but introduces artifacts in cardiac imaging, for example. The second measures the attenuation distribution using transmission of a reference activity to make a calibration.

2.10.

Clinical applications of nuclear medicine

The major clinical applications are the measurement of blood perfusion in the brain, the diagnosis of tumors in various organs and the assessment of cardiac function.

2.10.1. Brain Imaging Planar scintigraphy using DTPA or 99tTc-glucoheptonate, which do not cross the blood brain barrier (BBB), can be used to determine a rupture of the BBB as there will be a high concentration of these radiotracers in the brain. SPECT using Ceretec or Neurolite, which can cross the BBB, is used to measure blood perfusion as the concentration of these radiotracers is proportional to the regional cerebral blood flow (rCBF), indicating the presence of tumors, epilepsy, dementia, etc. An example can be seen in the picture below:

25

2.10.2. Cardiac Imaging Cardiac SPECT scans are performed to measure blood flow pattern in the heart and to detect coronary artery disease and myocardial infarcts. The most common test is the stress test, used to measure myocardial perfusion and diagnose myocardial ischemia and infarct.

2.11.

Positron Emission Tomography

PET, as SPECT, is also a tomographic technique used to measure physiology and function, rather than gross anatomy. The fundamental difference is that now the radiotracer emits positrons, which, after annihilation with an electron in the tissue, result in the formation of 2 rays. This fact makes it possible to produce images with much higher SNR and spatial resolution as we will see. PET is mainly used in oncology, cardiology and neurology. The main disadvantage is its high cost and the need to have a cyclotron on-site to produce positron-emitting nuclides.

2.11.1. General principles The radiotracers used in PET are structural analogs of a biologically active molecule, such as glucose, in which one or more atoms have been replaced by a radioactive atom. An important example is the FDG, which contains 18F and [11C]palmitate. Isotopes such as 18F undergo radioactive decay by emitting a positron, that is, a positively charged electron, and a neutrino:

The positron then annihilates with an electron, resulting in 2 -rays, each with an energy of 511 keV, which travel in opposite directions an angle of 180 to one another. Because 2 antiparallel -rays are produced and both must be detected, a PET system consists of a complete ring of scintillation crystals surrounding the patient. Since they are created simultaneously, both are detected within a certain time window with the 2 crystals that detected them defining a line along which the annihilation occurred. This process of line definition is called annihilation coincidence detection (ACD). This difference in localization method is the major reason for the much higher detection efficiency in PET than in SPECT (moreover, the fact that the -ray energy is much higher means less attenuation, increasing further the sensitivity). The spatial resolution depends upon a number of factors including the number and size of the individual crystal detectors (R~3-5 mm).

2.11.2. Radionuclides used for PET

11

All the radionuclides used in PET are produced by a cyclotron. The most common are 18F, C, O and 13N.
15

26

2.11.3. Instrumentation for PET The major differences are the scintillation crystals needed to detect the 511 keV -rays efficiently and the additional circuitry needed for coincidence detection.

2.11.3.1.

Scintillation crystals

They are in large number and are usually formed from bismuth germinate (BGO). The crystals are coupled to a smaller number PMTs (for cost reasons). Typically, each block of scintillation crystals consists of an 8X8 array cut from a single BGO crystal. Each block is coupled to 4 PMTs. Four of these blocks are arranged to form a bucket. The full detector ring may have up to 32 of such buckets. The ideal detector must have a high density and a large effective atomic number in order to increase the -rays detection efficiency by increasing Compton and photoelectric interactions. The size of the crystal is also important as it affects spatial resolution (however, crystals too small can cause scatter to adjacent crystals, which reduces resolution; width for BGO ~1 cm).

2.11.3.2.

Annihilation coincidence detection circuitry

The ACD circuitry is designed to maximize the ratio of true-to-false recorded coincidences. Considered the example in left picture:

A positron is emitted and annihilates with an electron, producing 2 anti-parallel -rays. The first is detected by the crystal number 2 and produces a number of photons. These photons are converted into an amplified electrical signal at the output of the PMT, which is fed into a PHA. If the voltage is inside a predefined range, than the PHA generates a logic pulse, which is sent to the coincidence detector. When the second -ray is detected (by the crystal number 10) 27

and produces a voltage that is accepted by the associated PHA, a second logic pulse is sent to the coincidence detector. This detector sums the 2 logic pulses and passes it through another PHA, which has a threshold set just under the sum value. If the logic pulses overlap in time, then the system accepts the 2 -rays as having evolved from one annihilation and record a line integral between the 2 crystals.

2.11.4. Image Reconstruction The process is basically the same as in SPECT, however, prior to reconstruction, corrections to attenuation and accidental coincidences must be made.

2.11.4.1.

Attenuation Correction

The 2 methods available are very similar to the ones described for SPECT.

2.11.4.2.

Correction for accidental, multiple and scattered coincidences

The main sources of noise in PET are accidental and scattered coincidences:

The accidental coincidences term is usually much higher. An event of this type is shown below:

28

The rate at which these accidental coincidences are recordeded is given by:

Where Ri and Rj are the single count rates in the individual detectors i and j. To correct the accidental coincidences, we can either measure the Ri and Rj or use an additional parallel timing circuitry. A multiple coincidence represents the combination of a true coincidence with one or more unrelated events, such as the one in the following picture:

In this case, 2 events are recorded at the same time. Since it is not clear which one should be accepted, both are discarded, which results in a significant loss of counts and in an effective deatime for the PET system. The deadtime loss (DTL) is estimated by measuring the number of triple coincidences and is corrected by the factor:

Scattered coincidences occur if one or both -rays suffer scattering before reaching the detectors. They can be corrected by measuring the amount of scatter in the image in areas outside the patient, with these values being extrapolated mathematically to estimate the amount of scatter inside the patient.

2.11.5. Image characteristics The factors that affect the SNR and CNR are identical to the ones in SPECT (however, there is no depth dependence on PET). Other factors that affect spatial resolution are:

29

Finite distance which the positron travels before annihilation; Slight deviation from a nominal angle of 180 due to the motion of the center mass of the annihilation (non-colinearity) Size of the crystal The overall spatial resolution is given by:

Where krecon is the degradation due to the reconstruction algorithm

2.11.6. Multislice and 3D PET Imaging Multislice capability can be introduced into PET as for CT by having a number of detector rings stacked alongside one another. Retractable septa are positioned between each ring: these are kept in position for multislice operation and retracted for 3D imaging. Sensitivity is about 10 times higher for 3D due to absence of the septa.

2.11.7. Clinical applications of PET PET is used to provide quantitative information on metabolic and physiological changes in 3 major areas: brain imaging, cardiac studies and tumor imaging.

2.11.7.1.

Brain Imaging

In the body, FDB is metabolized in the same way as glucose. Thus, it can cross the BBB. Inside the cell, FDG is phosphorylated and, unlike glucose, it is trapped inside the cell. The amount of FDG is therefore proportional to both the initial glucose transport and subsequent phosphorylation, providing data about the presence of tumors (high metabolic rate) or epilepsy (low metabolic rate), for example.

30

3. Ultrasonic Imaging
3.1. General principles of ultrasonic imaging Ultrasound imaging operates at frequencies between 1 and 10 MHz, producing images based on backscattering of mechanical energy from boundaries between tissues and from small structures within tissue. It can be used to obtain anatomical information (either at surface, using higher frequencies, or deep in the body, using lower frequencies) or to measure blood flow (via Doppler shift). Advantages: Noninvasive, easily portable, inexpensive diagnostic modality, allowing realtime imaging and high spatial resolution. Disadvantages: relatively poor soft-tissue contrast and the fact that gas and bone impede the passage of ultrasound waves.

Basic principle: A short pulse, typically 1-5 , of energy is transmitted into the body, using an ultrasound transducer. The transducer is focused to produce a narrow ultrasound beam, which propagates as a pressure wave. When the ultrasound wave encounters boundaries between tissues or structures within organs, a part of the energy is scattered in all directions, with a certain fraction being backscattered along the original transmission path and returning to the transducer, which converts it to a potential (later amplified and digitized). The direction of the beam is then changed to acquire a second line of data adjacent to the first one and so on.

31

3.2. Wave propagation and characteristic acoustic impedance A useful model of tissue is that of a lattice of small particles held together by elastic forces:

As the ultrasound energy passes through the tissue, the particles move very short distances (W) about a fixed mean position, whereas the ultrasonic energy propagates over much larger distances. The directions of particle vibration and wave propagation are the same (longitudinal wave). Assuming a planar wavefront and no loss of energy, the particle displacement W can be described as a function of the sound velocity by:

The value c depends on the tissue density and compressibility :

The particle velocity (much slower than c) in the z direction, , is given by

The pressure, p, of the wave is given by:

Because the transducer undergoes sinusoidal motion, () and () can be described as:

The mean intensity is given by

An important parameter in ultrasonic imaging is the characteristic acoustic impedance Z of the tissue (analog to Ohms Law):

32

3.3. Wave reflection and refraction Interaction of an ultrasound wave with a boundary between two tissues with different acoustic impedances is depicted below:

For the simplest case where the angle between the wave and the interface is 90, the pressure reflection coefficient (defined as the ratio between pressures of the reflected - and incident - - waves) and the pressure transmission coefficient (defined as the ratio between pressures of transmitted - , - and incident waves) are given by:

The same ratios can be expressed in terms of intensities of the waves:

Irrespective of whether the value of or is being considered, it is clear that the reflected signal detected by the transducer is maximized if the value of either 1 or 2 is zero. However, in this case, the ultrasound will not reach deep structures in the body. For example, more than 99% of the intensity is reflected at a gas/soft tissue interface. At the other extreme, if 1 and 2 are equal in value, then no signal is detected from the boundary. When the angle between the wave and the interface is different than 90, the equations governing the reflection and transmission angles are given by:

33

If 1 2 , then the transmitted signal is refracted, which leads to misregistration artifacts. The pressure and intensity coefficients are given by:

3.4. Energy loss mechanism in tissue In addition to reflection, ultrasound waves can be attenuated by absorption and scattering.

3.4.1. Absorption Absorption refers to the conversion of mechanical energy into heat, which can occur either by: Classical absorption: due to friction between particles as they are displaced by the passage of the ultrasound wave. Proportional to the square of the wave frequency. Relaxation: due to the time taken for a molecule to return to its original position. If the relaxation time of the tissue is of the same order of the wave period, then the relaxation mechanism can act against the compression/rarefaction cycle, which leads to loss of energy.

3.4.2. Scattering Scattering occurs when the beam encounters tissue irregularities or particles that are the same size or smaller than the ultrasound wavelength. If the size of the scattering body is small compared to the wavelength, then scattering is relatively uniform in direction with slightly more energy being scattered towards the transducer (Rayleigh scattering). It is characterized in terms of scattering cross section , which depends on the fourth power of wave frequency.

3.4.3. Attenuation Attenuation is the sum of the scattering and absorption processes. It is characterized by an exponential decrease in both pressure and intensity of the ultrasound as a function of propagation distance z:

34

Where the intensity attenuation coefficient and the pressure attenuation coefficient depend linearly on the wave frequency.

3.5. Instrumentation The instrumentation for ultrasounds consists of: Transducer (single-crystal or, more commonly, an array of crystals): converts oscillating voltage into mechanical vibrations and vice-versa Detection electronics: modules for time-gain compensation and beam forming; Computer: data processing, image display and data storage

3.5.1. Single-crystal transducers A schematic diagram of a transducer is shown below:

The main components of a transducer are: Piezoelectric crystal: oscillates at the same frequency as the applied alternating voltage with a change in thickness proportional to the magnitude of this voltage. Usually made of lead zirconate titanate (PZT). It has a natural resonant frequency of:

Where is the velocity of sound in the crystal and d is the thickness. If the diameter of the crystal is much larger than its thickness, a longitudinal wave is transmitted. Damping material: absorbs energy from the vibrating transducer, which shortens the ring-down time produced after a voltage pulse. A shorter pulse (with a broad 35

bandwidth BW in the frequency domain) gives a better spatial resolution. It is often specified the quality factor Q which, in a well damped material, is between 1 and 2.

Since the acoustic impedance of PZT is about 15 times that of skin, there is a huge amount of energy reflected. Thus, a layer of material with an acoustic impedance is placed to maximize energy transmission:

3.5.1.1. The beam geometry of a single transducer Considering a plane-piston (flat face) transducer, made up of a large number of point sources, the total pressure wave emitted is the superposition of the spherical waves emitted by these point sources:

36

The near-field boundary (NFB) corresponds to the last maximum and separates the nearfield (where the wavefront if not well defined), or Fresnel zone, from the far-field, or Fraunhofer zone (where the wavefront is well approximated as planar):

Where a is the diameter of the transducer and is the wavelength of the ultrasound. Beyond this point, the beam diverges and its intensity decreases smoothly. In addition to the main beam, side lobes may be present due to the transducer acting as a diffraction grating, which is undesirable since they remove energy from the main beam and can introduce artifacts. The greater the ratio wavelength/transducer diameter, the fewer the number of side lobes but the closer the NFB lies to the face of the transducer.

3.5.1.2. Lateral resolution and depth of focus In the far-field region, the lateral beamwidth can be well approximated by a Gaussian function, which has a FWHM:

Where is the standard deviation. Since the diameter of the single-crystal is typically between 1 and 5 cm, the lateral resolution is usually very poor and a concave lens is used to focus the beam. Lowering the lens curvature (R) lowers the focal distance F (distance from the face of the transducer at which the lateral beam width is narrowest). The plane perpendicular to the beam axis at F distance is called focal plane. 37

For a spherical focusing lens, the FWHM at the focal point is given by

Therefore, decreasing the lens curvature (R) and increasing the diameter of the crystal (a) improves lateral resolution. The wavelength () dependence arises from the appearance of side lobes as discussed previously. However, there is a tradeoff: the better the lateral resolution (small FWHM), the lower the depth of focus as the beam diverges much more at locations away from the focal plane.

3.5.1.3. Axial resolution Axial resolution is defined as the closest separation, in the direction of the propagating wave, of two scatterers that result in two backscattered signals. This distance can be expressed as:

Where PD is the pulse duration. Therefore, axial resolution can be improved by reducing PD (either by using a higher frequency ultrasound which also increases attenuation or improving the transducer damping).

38

3.5.2. Transducer arrays There are several problems regarding single crystal transducers: Require manual or mechanical steering of the beam; Tradeoff between lateral resolution and DOF Large distance between the face of the transducer and the NFB

One way of avoiding these problems is to use an array of small piezoelectric crystals. There are three types of arrays:

3.5.2.1. Linear sequential arrays A linear sequential array consist of a large number (64-512) of rectangular piezoelectric crystals, each having a width of the order of the ultrasound wavelength. The width of the ultrasound is determined by the number of elements that are excited simultaneously. A planar wavefront is produced by exciting a number of elements (3 for example), which corresponds to the first line. The process is repeated for the adjacent set (2nd, 3rd and 4th elements), displacing laterally the focus point, and so on. Additional lines can be acquired by performing again the same process but with an even number of elements (4 for example).

It is particularly used when a large FOV is required close to the surface of the array.

3.5.2.2. Linear phased arrays The layout of a linear phased array is very similar to that of a linear sequential array, but operates in a different way. A much larger number of elements is excited for each line with the voltage pulses exciting each one being delayed in time in order to produce a curved wavefront similar to that produced by a focused single crystal transducer. The elements are smaller than

39

the ultrasound wavelength and, thus, the focal point lies well beyond NFB where the geometry of the wavefront is well characterized.

A process called dynamic focusing can be used to optimize the lateral resolution: as we go deeper in the body, the number of elements required to produce a focal point at that distance increases and thus the number of elements is increased dynamically during transmission of the ultrasound.

3.5.2.3. Multidimensional arrays To allow for focusing in the elevation dimension, a number of rows can be added up this dimension, forming a multidimensional array.

3.5.2.4. Annular arrays Annular arrays allow for 2D lateral focusing, but the steering of the beam must be done manually or mechanically.

40

3.5.3. Beam forming and time-gain compensation Time-gate compensation (TGC): signal amplification is dependent on the time that the signal takes to reach the transducer. Signals arising from structures close to the transducer (faster to arrive) are amplified by a smaller factor than those from greater depths (which suffered more attenuation). Various linear or nonlinear functions can be used and adjusted online by the operator.

3.6. Diagnostic scanning modes There are three basic modes of diagnostic anatomical imaging: A-mode, M-mode and Bmode. Recent technical advances include the use of compound and 3D imaging.

3.6.1. A-mode, M-mode and B-mode scans Amplitude (A)-mode: refers to the acquisition of a 1D scan, that is, a plot of the amplitude of the backscattered echo versus the time after transmission of the ultrasound pulse. Used for measuring distances for example in ophthalmology (assuming constant velocity of the sound). Motion (M)-mode: consists of a series of A-mode scans to detect motion of a moving structure. The brightness of the displayed signal is proportional to the amplitude of the backscattered echo.

41

Brightness (B)-mode: produces a 2D image through a cross section of tissue. Each line in the image consists of an A-mode scan, with the brightness of the signal being proportional to the amplitude of the backscattered echo. Both stationary and moving structures can be scanned.

3.6.2. Three-dimensional imaging In 3D imaging, ultrasound waves are sent at different angles and the returning echoes are processed to reconstruct a 3D image. There are two main approaches: Using a 2D US probe for acquiring images by manually moving the probe in a direction perpendicular to the plane of each B-mode scan; Using a dedicated 3D US probe: a 2D array acquires 3D volumes directly.

It allows, for example, better estimation of tumor or cardiac valves dimensions and detection of fetal malformations.

3.7. Artifacts in ultrasonic imaging Artifacts can result from a number of different effects: Reverberations: occur if there is a very strong reflector (bone or air) close to the transducer surface. Appear as a series of repeated lines; Acoustic shadowing: occur if either a very strong reflector (gas/tissue boundary) or highly attenuating medium shadows a deeper-lying organ. Appears as a dark area or hole; Acoustic enhancement: opposite effect of acoustic shadowing. Occurs if region of low attenuation is present, making the areas appearing behind with higher than expected intensity. Useful for differentiating fluid-filled cysts from solid masses; Refraction: occur when the signal is refracted between two tissues with different characteristic acoustic impedances (these can result in 20 deviations). 42

3.8. Image characteristics The factors that affect the image SNR, spatial resolution and CNR have already been mentioned so only a brief summary is presented below:

3.8.1. Signal-to-noise ratio The noise arises from three components: Electronics in the detection system: can be minimized if the backscattered signal has a high enough amplitude and is amplified by a sufficient factor; Speckle: coherent wave interference in tissue. It gives a granular appearance to what should appear as homogeneous tissue (evidenced in the picture above, on the right); Clutter: arises from side and grating lobes, tissue motion and other acoustic phenomena. Can be minimized by using harmonic imaging methods. The signal intensity is affected by: Intensity of the ultrasound pulse; Frequency of the transducer: the greater the frequency, the higher the attenuation; Type of focusing used: the stronger the focusing at a particular point, the higher is the energy per unit area and the higher the SNR at that point, but very low outside of it; Degree of damping: the lower the amount, the higher the intensity at the fundamental frequency.

3.8.2. Spatial resolution Spatial resolution is affected by: Degree of focusing: the stronger the focusing, the higher is the lateral resolution at the focal spot; Length of the transmitted ultrasound pulse: the longer the pulse, the poorer the axial resolution (controlled by damping and operating frequency). 43

3.8.3. Contrast-to-noise ratio Affected by the same factors as SNR. It can be greatly improved by using contrast agents and pulse inversion techniques, discussed in the next sections.

3.9. Compound Imaging Also called sonoCT, it consists in acquiring multiple coplanar B-mode scans and combining them into a single image. Although it presents blurring due to movement across multiple acquisitions, it has the following advantages: Improved SNR (especially in the center, where the lines overlap) Reduced speckle and clutter Reduced acoustic shadowing

3.10.

Blood velocity measurements using ultrasound

Two techniques are used to estimate blood velocity: Doppler shift based techniques (used in either continuous CW , pulsed or duplex modes); Time-domain signal correlation techniques.

3.10.1. The Doppler Effect Blood flow, either toward or away from the transducer alters the frequency of the backscattered echoes due to scattering with red blood cells (RBC):

Since the wavelength of the ultrasound is much greater than the dimensions of the red blood cells, the wave is scattered in all directions, which means the Doppler-shifted signals have very low intensities. Signal intensity is proportional to the fourth power of the US frequency and,

44

thus, higher operating frequencies are used (however, maximum measuring depth decreases with higher frequencies because of the beam attenuation). The overall Doppler shift is given by:

3.10.2. Continuous wave Doppler measurements CW Doppler uses a continuous pulse transmitted by one transducer and the backscattered signal is detected by the second one, with the region of overlap of the sensitive regions of the transducers defining the area in which blood flow is detected (it is often large and the existence of more than one blood vessel can lead to misinterpretations). It is used when there is no need to localize exactly the source of Doppler shifts. Its advantages are: No maximum depth limitation; No maximum measurable velocity limitation.

3.10.3. Pulsed-mode Doppler measurements Only one transducer is used to transmit and receive backscattered signals. It allows to measure Doppler shifts in a specific region, which are measured by calculating the difference in transmit-to-receive time from the first pulse to the second (delay decreases if blood is flowing toward the transducer and vice-versa). The processing steps of this technique are depicted below:

45

Contrary to CW Doppler, there is a limit to the highest velocity which can be measured imposed by the Nyquist theorem (sampling frequency in this case, the pulse repetition rate PRR must be at least twice the measured frequency). It is given by:

It is also limited to a maximum depth : = 2

3.10.4. Color Doppler/B-mode Duplex Imaging Duplex imaging consists in interlacing Doppler flow measurements with B-mode imaging in order to impose the flow maps onto high-resolution anatomical images. Only the mean value of the velocity and not the full velocity distribution is determined. The mean velocity, its sign and its variance are represented by the hue, saturation and luminance respectively (red and blue represents flow towards and away from the transducer, respectively). When a vessel lies parallel to the face of the transducer array, directly below the center of the transducer array there is a signal void, which can be solved by using power Doppler mode. It consists in integrating the area under the frequency vs amplitude plot, allowing to: Remove angle dependence (power depends only on the number of RBC scatters); Reduce aliasing artifacts at high flow rates.

The main disadvantage of Doppler power is the loss of directional information.

3.11.

Ultrasound contrast agents, harmonic imaging and pulse inversion techniques

Contrast agents are used to increase the intensity of backscattered signals and consist of gas-filled microspheres or microbubbles injected into the bloodstream. They act by two mechanisms: Large increase of the difference in acoustic properties (higher compressibility and lower density) between gas-filled particles and the surroundings; Resonance: gas-filled microspheres expand and contract when the US travels through them, acting as harmonic oscillators. These mechanisms results in a much larger effective scattering cross section. Harmonic Imaging: the backscattered signal consists of a fundamental frequency and its harmonics. Although these harmonics have lower intensities, they can have higher SNR because of the very low contributions from clutter and tissue motion. The most common implementation uses the second harmonic using a pulse inversion technique, which combines two scans. In the first scan, both the returning fundamental and its harmonic are stored. In the second scan, the fundamental signal is inverted, but the harmonic has the same phase. Summation of both scans thus results in the elimination of the fundamental component.

46

4. Magnetic Resonance Imaging

4.1. General principles of Magnetic Resonance Imaging Magnetic Resonance Imaging (MRI) nonionizing technique with full 3D capabilities, excellent soft-tissue contrast and high spatial resolution (~1mm), although expensive and subject to patient motion due to high scanning times (typically 3-10 mins).

Brief description of the technique MRI signal arises from protons in the body (mainly water), which act as small magnets. After the patient is placed inside the scanner, a strong magnet causes the protons to precess either in a parallel or antiparallel configuration regarding the direction of the static magnetic field. The frequency of precession is proportional to the strength of the field. Application of a weak radiofrequency (RF) field causes protons to precess coherently and the sum of their number is detected as an induced voltage in a tuned detector coil. Spatial information is encoded into the image using magnetic field gradients (one in each direction). These gradients cause variations in the magnetic field, which, in turn, causes the precessional frequencies to vary depending on their spatial location. Frequency and phase is measured by the RF coil and the signal is then digitized. Finally, an inverse 2D Fourier transform is performed to convert the signal into the spatial domain.

4.2. Nuclear magnetism MRI arises from the interaction between the magnetic field and the hydrogen nuclei, which can be described from a quantum mechanical or classical approach.

4.2.1. Quantum mechanical description The spin of a proton can be viewed as a rotation around an internal axis, giving the proton a certain angular momentum P. Since a proton is a charged particle, it gives the proton a magnetic moment , which, in turn, produces a magnetic field. In the absence of an external magnetic field, the orientation of the individual magnetic moments is random.

47

The magnitude of P is quantized:

For protons, the spin quantum number l is equal to , and thus:

The magnetic momentum is related with P by the gyromagnetic ratio :

In the presence of a strong magnetic field 0 , the z component (along 0 ) of can only have the values given by:

The nuclear magnetic quantum number takes values l, l-1,,-l so, in the case of a proton, takes +1/2 and -1/2, yielding: = /4 The magnetic field only interacts with the z component, thus:

The two possible interaction energies correspond to the parallel (negative E) and antiparallel (positive E) configurations. The energy difference between the 2 levels is given by:

Using the Boltzmann distribution, we find the relative number of nuclei in each configuration:

The magnitude of the MRI signal is proportional to the difference in populations between the 2 energy levels:

Where is the total number of protons in the body.

48

4.2.2. Classical description By attempting to align the proton magnetic moment parallel to the direction of 0 , the magnetic field creates a torque. The result of this torque is that the proton precesses around the axis of 0 , keeping a constant angle between 0 and .

The precession frequency, termed Larmor frequency, can be calculated, yielding

4.2.3. Radiofrequency pulses and the rotating reference frame In order to obtain an MRI signal, transitions must be induced between the protons in both energy levels. The energy required to do this is supplied by an oscillating electromagnetic field at a specific frequency f (resonance frequency): 49

If we refer to the expression in the previous section, we notice that this resonance frequency is the Larmor frequency. The electromagnetic energy is provided as a single or series of radiofrequency (RF) pulses. To analyse the effect of a given sequence of pulses, we will consider the net effect of all of the protons in the body, thus we define the net magnetization as:

Since the distribution of magnetic momenta in the transverse plane is random, the net magnetization in x and y directions is zero:

As we will see, a detectable signal can only be produced by and , so it is necessary to rotate the net magnetization from z to xy plane by applying a second magnetic field 1 aligned along x at the Larmor frequency (much slower than the frequency of 0 ). The nuclei are now said to be phase coherent as all the vectors are pointing in the same direction. The tip angle is defined as the angle through which the net magnetization is rotated. It depends on the strength and time during which the RF is applied:

To simplify the visualization, we will use a rotating reference frame in which xy plane rotates around z at the Larmor frequency:

The signal is detected by an RF coil in the form of a voltage E across the ends of the coil loop, created by the change of the magnetic flux (Faradays law):

50

This is the reason why only the xy plane gives rise to a nuclear magnetic resonance (there is no time change in z). The higher 0 , the higher is the signal detected.

4.2.4. Spin-Lattice and Spin-Spin Relaxation After the application of an RF pulse, the magnetization components must return to their equilibrium value over time. The time evolution of each component is given by the Bloch equations. The return of to its equilibrium value 0 is governed by the spin-lattice relaxation time 1 (arises from the loss of proton energy to the surrounding lattice) and it is given by the expression:

Different tissues have different values of 1 , therefore 1 is one source of contrast in MRI imaging. and components relax back to their equilibrium values with a time constant termed spin-spin 2 relaxation time (arises from the loss of phase coherence, caused by interactions with neighboring nuclei):

The relation between phase decoherence and component is depicted below:

51

Aside from the nuclei-nuclei interaction, phase decoherence is also affected by spatial variations of the magnetic field (due to non-uniformities in the magnet design or different magnetic susceptibilities of different tissues). The overall relaxation of transverse magnetization is given by 2 , a combination of both effects:

Different tissues have different 2 and, thus, 2 is used as a contrast source as well.

4.2.5. Measurements of and : inversion recovery and spin-echo sequences The value of 1 is measured using an inversion recovery sequence, which consists of a 180 pulse, a variable delay and a 90 pulse, followed immediately by data acquisition. This sequence is repeated n times, each with a different . The detected signal is given by:

A plot of ln(( )) versus gives a straight line with a slope of 1. The measurement of 2 involves a spin-echo setup, where a 90 is applied, followed by a variable delay , a 180 pulse (in the xy plane), an identical and then the signal acquisition. To see how the spin-echo sequence works, consider a single proton which, due to spatial inhomogeneities in the magnetic field, resonates at a frequency , less than the nominal Larmor frequency. At after the 90 pulse, the precessing magnetization has accumulated a phase = . By applying the 180 pulse, we convert to and + to . During the second interval, the precessing magnetization accumulates a further phase +. Thus, at 2, + the precessing magnetization has 0 phase, which eliminates the 2 contribution, yielding 2 = 2 :

52

A plot of ln(( )) versus 2 gives a straight line with a slope of 1/2.

4.3. Magnetic Resonance Imaging MRI uses a magnetic field gradient, that is, a spatial variation in the magnetic field across a sample to produce a range of proton resonant frequencies, each dependent upon the position of a particular proton in the body. The creation of such magnetic field gradient requires 3 gradient coils to encode each spatial dimension. To simplify the image reconstruction, we assume the magnetic gradients are linear:

The gradient coils are designed such that there is no additional contribution to the magnetic field at the isocenter (z=0, y=0, x=0). The following picture shows a plot of magnetic field vs spatial position for a gradient applied along z axis:

The magnetic field experienced by a proton at a given z coordinate is therefore given by:

The corresponding precessional frequency is:

In the rotating reference frame, it can be expressed as:

Analogous expressions can be obtained for and . The process of image formation can be broken down into 3 components: slice selection, phase-encoding and frequency-encoding. 53

4.3.1. Slice Selection Slice selection is accomplished using a frequency-selective RF pulse applied simultaneously with one of the magnetic field gradients (depending on the orientation of the slices), denoted by :

If the RF pulse is applied at a frequency with an excitation bandwidth of , the protons precessing with a frequency between + and are rotated into the transverse plane, while all the others remain unaffected. The thickness T of the slice depends on the frequency bandwidth and the gradient value :

The ideal frequency profile of the RF pulse if a rectangular shape, which corresponds to a sinc shape in time domain. From the properties of the Fourier transform, a longer RF pulse results in a narrower frequency spectrum and therefore a thinner slice (typical duration ~1-5ms). Due to the fact that the RF pulse is relatively long, nuclei accumulate different phases depending on their position within the slice. Therefore, a rephasing gradient of the opposite polarity is applied afterwards.

4.3.2. Phase-Encoding Having selected a slice, the other 2 dimensions must be encoded to produce a 2D image. After the slice selection pulse, the phase-encoding gradient is applied for a period and then switched off before data acquisition begins. If the is applied in y axis, during , the protons precess at = , which introduces a spatially dependent phase shift:

54

4.3.3. Frequency-Encoding The frequency-encoding is encoded by the nuclei precessing at different frequencies under the influence of a gradient , which is applied during data acquisition. The time between successive acquisitions is referred to as dwell time , being the reciprocal of the acquisition bandwidth. A graphical representation of the slice, frequency and phase encoding processes is depicted below:

4.3.4. The k-Space formalism To understand how the acquired data matrix is transformed into the final image, we will look into the k-space formalism. First, we define two variables:

With x and y being the frequency and phase encoding directions respectively. The signal can then be expressed as:

The acquisition points can be visualized as a 2D data set in k-space. Consider the points collected when the minimum value of the phase encoding is applied. Referring to 55

the expression, all these have the minimum value. When the frequency-encoding gradient is turned on, the first data point collected corresponds to a small positive value of , the second data point to a slightly more positive value of and so forth, and so these points correspond to the first (bottom) line in k-space. The second line corresponds to the next value of the phase encoding gradient and so on. The spacing between k-space points is dictated by the required FOV of the image:

A 2D inverse Fourier transform of the k-space data ( , ) gives an estimation of (, ), that is, an image corresponding to the spatial variation in proton density. In the previous example only the positive half of the was acquired. In full k-space coverage, the SNR and spatial resolution (determined by the maximum values of and ) would be increased. This can be achieved by a gradient-echo imaging sequence, which adds a negative gradient before data acquisition.

56

4.4. Instrumentation Three basic components make up the MRI scanner: Magnet: polarizes the protons; Three gradient coils: impose a linear variation on the proton Larmor frequency as a function of position; RF coil: produces the oscillating magnetic field necessary to maintain phase coherence between protons and receives the MRI signal by Farady induction.

4.4.1. Magnet Design The magnet must produce a magnetic field that is: Strong: increases the amplitude of the MRI signal; Homogeneous: increases 2 and avoids distorted images; Temporally stable: avoids artifacts.

There are three basic types of magnets: Permanent: used for 0 < 0.35, made of rare earth alloys.

Advantages: low cost, does not require cooling, reduced susceptibility to patient claustrophobia (open). Disadvantages: large magnet weight, field homogeneity and stability are highly temperature dependent.

57

Resistive: used for 0 < 0.35, magnetic field strength is proportional to a constant current passing through a conductor such as copper. Power dissipated as heat is proportional to the resistance of the conductor and the square of the current. Disadvantages: requires cooling, maximum magnetic field strength is limited by the maximum current, homogeneity and stability highly dependent on temperature.

Superconducting: used for 0 > 0.35 (typically 1.5T or 3T). It minimizes the problem of cooling by using superconductors which have 0 resistance at very low temperatures and allows stronger magnetic fields to be used. The superconductor is housed in a can containing liquid helium at 4.2K, surrounded by radiation shields, vacuum vessels and a liquid nitrogen chamber. Because heat losses are not avoidable, liquid nitrogen and helium must be replenished on a regular basis. Additional correction and shim coils are added to correct uniformity of the magnetic field.

4.4.2. Magnetic field gradient coils Gradient coils have three basic requirements: Production of linear gradients over the imaged region: their geometry is optimized to produce a linear gradient, rather than a uniform field. Since the value of the gradient is relatively small, chilled water cooling is sufficient. The z coil consists of a Maxwell pair (two separate loops, with current flowing in opposed directions). The x and y coils have a saddle coil arrangement (four arcs each). The gradient strength is proportional to the number of turns.

Production of high gradient strengths per unit current: achieved by minimizing resistance; Fast switching times: achieved by minimizing inductance; important to reduce the time which must be allowed for gradients to stabilize in imaging techniques.

58

4.4.3. Radiofrequency coil The RF coil is designed to store as much of its magnetic energy as possible in the nearfield region, that is, within the patient. It should also detect efficiently the precessing nuclear magnetization, resulting in a high image SNR. There are RF coils designed for volume (ex: birdcage) and surface (ex: circular loop) imaging. 4.5. Imaging sequences Depending on the clinical application, either spin-echo or gradient-echo sequences are used. 4.5.1. Spin-echo imaging sequences The spin-echo imaging sequence can be considered as an imaging version of the spin-echo sequence used to measure tissue 2 values. Compared to gradient-echo based sequences, the + spin-echo method has the advantage of refocusing 2 effects, so that the magnitude of the signal detected is governed by the 2 , rather than 2 value of the tissue. Two RF pulses are used: the 90 pulse creates components of precessing transverse + magnetization and the 180 pulse refocuses the effects of 2 relaxation. The 90 pulse is applied simultaneously with to select the desired slice: the 180 pulse is also applied simultaneously with to allow multislice imaging as we will cover later. Phase-encoding is carried exactly as described previously, with the number of increments defining the spatial resolution in this dimension. The time between successive phase encoding increments is termed TR.

Regarding the frequency-encoding axis, instead of applying a negative dephasing gradient as in the gradient-echo sequence, the dephasing gradient in a spin-echo sequence is usually applied between 90 and 180 pulses with a positive polarity. The blank periods are introduced to give certain values to TR and TE in order to achieve corresponding 1 and 2 -contrast weighting into the image as discussed in the next section. 59

4.5.2. and -weighted imaging sequences The intensity of an axial image acquired using a spin-echo sequence is given by:

The term 1 /1 determines the 1 weighting of the sequence. If the TR value (which can be set by the operator) is set to a value much greater than 1 , then the image has no 1 weighting as the term 1 /1 is very similar for all tissues. If the TR value is set closer to the tissue 1 then the image becomes more 1 -weighted. As the TR values becomes smaller, the SNR decreases, although the total imaging time ( ) also decreases (the goal is to decrease imaging time while maintaining sufficient SNR for accurate diagnosis). The same considerations apply to the term /2 , which determines the degree of 2 weighting: if the value of TE is set much shorter than the tissue 2 , then no 2 contrast is present; if the value of TE is too long, then the contrast is high, but the SNR is low (the goal is to have the highest CNR).

4.5.2.1. Which physical properties influence and ? The mechanism of 1 relaxation involves the protons losing their excess energy via interactions with oscillating magnetic fields, produced by nuclei in surrounding molecules as they execute random Brownian motion. When these randomly fluctuating magnetic fields have a component at the Larmor frequency, they can stimulate transitions between the antiparallel and parallel states, and thereby cause the component to return to 0 . Thus, slowly moving molecules (low frequency) in a viscous media have higher contributions to 1 at low magnetic field strengths, while rapidly moving molecules (high frequency) in mobile liquids produce fluctuating magnetic fields over a wide frequency range. The spectral density () is a measure of the number of nuclei capable of producing magnetic fields at the Larmor frequency, thus the greater (), the shorter is 1 : 60

The above picture also shows that higher contrast is achieved for low field strengths, however SNR is also lower, which leads to a tradeoff. The same mechanisms affect 2 , however there is an extra contribution arising from magnetic field fluctuations at near-zero frequencies, which determines that 2 can never be as long as 1 . 4.5.3. Multislice Imaging Unlike most other imaging modalities, MRI can acquire multiple slices in essentially the same time as for a single slice run. This is done by making more acquisitions during the TR-TE time and using RF pulses with different center frequencies. In practice, the odd-numbered slices are acquired first to avoid partial excitation of adjacent slices due to the nonideal frequency profile of the RF pulses.

61

4.5.4. Rapid gradient-echo sequences and 3D Imaging Some applications (ex: abdominal imaging) require very fast imaging times to avoid motion artifacts, which is achieved by reducing TR. For a given TR, the value of that maximizes SNR is referred as the Ernst angle:

Rapid gradient-echo imaging also allows 3D imaging within time scales commensurate with clinical practice. They use a conventional frequency-encoding gradient, but two incremental phase-encoding gradients:

4.5.5. Echo-planar imaging In applications like diffusion and perfusion weighted imaging of the brain, even faster imaging times are required. The fastest type of imaging sequence uses a single RF pulse (singleshot) to excite the protons in the chosen slice, followed by full k-space sampling in a single echo train. The most common sequence, called echo-planar imaging (EPI), which is able to obtain scans in less than 100 ms, is depicted below:

62

This sequence can introduce an unacceptable level of image blurring due to the broad PSF, which arises from 2 relaxation during the sequence. To overcome this, a segmented mode can be used, where, for instance, only every fourth data point is acquired and the process is repeated 4 times.

4.5.6. Spiral imaging Instead of a rectangular k-space trajectory, spiral imaging uses a spiral trajectory, which presents two basic advantages: Low spatial frequencies are sampled more densely, because the spiral is tightest close to the origin of the k-space axis. This results in higher SNR and compensation for patient motion; The gradient slew rate (rate at which strength is changed) is slower.

Prior to the 2D inverse Fourier transform, an interpolation onto a rectangular grid must be performed. A segmented mode is also commonly used. 4.6. Image characteristics The tradeoffs and factors that affect SNR, spatial resolution and CNR are discussed below. 4.6.1. Signal-to-noise ratio The main factors affecting SNR are: Magnetic field 0 strength: improves nuclear polarization and voltage measured by Faraday induction () RF coil sensitivity: defined as the 1 field produced per unit current () Noise contributions from the RF coil and the sample: random voltage fluctuations () Number of phase encoding steps () Number of frequency encoding steps () Slice thickness () 63

4.6.2. Spatial resolution The image PSF is different in each spatial dimension. In the slice-select direction, the PSF is simply related to the RF pulse frequency profile (slice thickness). In the frequency and phase encoding directions, three factors affect the PSF: Digital resolution: defined as FOV/Number of data points acquired.

Data truncation: because only a finite number of frequency and phase encoding data points are taken, the data are effectively truncated and the corresponding PSF is a sinc function. The width of the sinc function in each dimension is inversely proportional to the number of data points acquired; relaxation time: the exponential decay corresponds to a Lorentzian PSF, which FWHM decreases when 2 increases:

4.6.3. Contrast-to-noise ratio CNR is essentially dependent on the difference between the relaxation times and proton density and the values of TR and TE. There is also a dependence on the strength of 0 since 1 weighted contrast is higher for lower 0 . Since CNR between two tissues depends on the respective SNRs, CNR is also affected by the factors listed for SNR.

4.7. MRI contrast agents In certain situations such as the detection of very small lesions, where the signal from the lesion is effectively averaged with that of healthy tissue, the CNR can be very low. In this case, one of two basic types of MRI contrast agents can be used: Paramagnetic agents: shorten 1 ; tissues on which it accumulate appear bright; Superparamagnetic (or ferromagnetic): shorten 2 and 2 , accumulate primarily in healthy tissues.

4.7.1. Paramagnetic agents Paramagnetic agents are based on metal ions such as gadolinium that have a large number of unpaired electrons (with high magnetic moments). The interactions between the unpaired electrons and the water molecules cause the proton 1 to be shortened, with the effect being quantified by the relaxivity of the agent:

64

Where 1, is the 1 of water containing a concentration C of the contrast agent, 1,0 is the corresponding value without contrast agent and 1 is the relaxivity. There are two mechanisms which result in enhanced relaxation efficiency, depicted below:

4.7.2. Superparamagnetic agents Superparamagnetic agents consist of small magnetic particles containing iron, which possess extremely high magnetic moments. They cause inhomogeneities in the local magnetic field, causing a reduction in signal intensity. Therefore, pathological tissue (which do not absorb these agents) appear unaffected as a relatively bright area, while the healthy tissue appears with reduced intensity.

4.8. Magnetic resonance angiography Unlike X-ray angiographic techniques, magnetic resonance angiography (MRA) does not require the use of a contrast agent, although they can be used to increase signal difference. There are two major techniques:

4.8.1. Time-of-flight methods Time-of-flight methods (TOF) are based on the shortening of the effective 1 , 1() , of blood as it flows into and through the image slice during data acquisition. The reason for the effective shortening of 1 is that, at each incremental value of the phase-encoding gradient, protons in the blood that have not experienced the previous RF pulse enter the slice with full magnetization = 0 . For a given thickness and blood velocity v, the value of 1() is given by

65

The simplest implementation of the TOF principle uses a rapid gradient-echo sequence with , which produces a high 1 - weighted image.

4.8.2. Phase-contrast methods Phase-contrast (PC) angiographic techniques add a bipolar gradient pulse, which introduces a velocity dependent phase into the signal. For stationary protons, the phase displacement is zero, however, for protons in blood flowing at constant velocity , the phase is given by:

The phase of a single image cannot be used directly to measure the velocity. A second image must be acquired in which the polarity of the bipolar gradient is reversed. While the phase of a static proton remains the same in both images, the phase of flowing protons is reversed in sign.

4.9. Diffusion-weighted imaging Measurement of the rate of water diffusion is often indicative of the tissues health. For example, in conditions such as stroke, cells swell and cell membranes can rupture, thus higher diffusion rates can mean fewer physical barriers. The simplest pulse sequence used is based on a spin-echo sequence, with symmetric diffusion-encoding gradients applied at either side of the 180 refocusing pulse. The first diffusion gradient encodes each proton with a certain phase. If the proton does not diffuse, then the second diffusion gradient will impose an opposite phase on the proton, thus there is no net dephasing of the magnetization. However, if the proton diffuses, the proton is only partially rephased, which leads to a loss of signal intensity. The higher the diffusion, the greater the loss in signal. To quantify the diffusion coefficient the imaging sequence is repeated a number of times using different values of diffusion gradients and fitted to the basic equation describing diffusive signal loss. 66

4.10. Functional MRI fMRI makes use of the sensitivity of the MRI signal intensity to the level of oxygen in the blood in the brain (oxygen-level-dependent BOLD effect) to determine which areas of the brain are involved in specific cognitive tasks. In gray matter in the brain, a modest increase in the cerebral metabolic rate of oxygen (MCRO2) and of glucose occurs in areas involved in neural activation, which leads to the release of vasodilatory compounds and increase of blood flow in the capillary bed. The rate of delivery of oxygen increases, but to a much lesser extent than the increase in blood flow, which leads to an oxygenation of blood. This, in turn, increases the level of oxyhemoglobin and decreases the level of deoxyhemoglobin. The decrease in deoxyhemoglobin reduces the local magnetic field gradients between the blood in the capillary bed and tissue, which leads to an increase of 2 and 2 in areas of the brain associated with neural activation. The most common sequence used is the multislice EPI because it is fast enough to obtain whole brain coverage in a few seconds. The changes in image intensity in activated areas are very small, thus the experiments are repeated a number of times and the data is subjected to statistical analysis and filtering. The changing in MRI signal intensity at sites associated with neuronal activation has three components: Small signal decrease immediately after the onset of the stimulation, which corresponds to the total deoxyhemoglobin increasing for the first 3 s due to an initial increase in oxygen extraction before the increase in blood flow; Increase in signal intensity, which reaches a maximum at about 7 s after the onset of the stimulus (hemodynamic response time); Signal undershoot which can last up to 1 min and corresponds to the blood volume in the venules.

67

Apart from the intrinsic increase in image SNR, operating at higher field strengths increases the contribution of the BOLD component and eliminates the contribution of the venules.

4.11.

Clinical applications of MRI

The majority of clinical diagnosis using MRI rely on the intrinsic contrast between pathological and healthy tissue. It is often coupled with other imaging techniques and associated with contrast agents such as Gd-DTPA, using the appropriate protocol (1 -weighted, 2 weighted or proton density-weighted).

68