PET Instrumentation

In many ways PET instrumentation is similar to gamma cameras. These similarities include:
1. Photon interaction with a solid crystal which causes scintillation
2. PMTs convert the scintillation to an electron pulse
3. Amplification of the pulse and discrimination circuitry
4. Storage of data, application of filters, and the diagnosing of disease
5. While similarities are identified there is myriad of differences. So let us begin our exploration.
6. It's going to take a semester…
Before we discuss anything further, exactly what does PET detectors detect? Gamma-rays? The answer is NO! They detect photons. Why are they photons and not gamma rays? Hm ... consider the source in the production of the photon

Starting with the Basics of Instrumentation

Collimation in PET
1. In general PET does not employ collimators, however, this concept is utilized in a different manner and will be discussed later
2. First we have to ask, why are collimators not needed? Photon event (not gamma) radiation comes from an annihilation reaction that creates two distinct 511 keV annihilation photons traveling in opposite direction (~180^o) from one another. This is known as coincidence detection because there are two photons striking the crystal, in opposite directions, at the same time. Therefore "focusing" of the events to a specific area in the gantry of the FOV. Hence septa are technically not needed since we have a good idea on where the event occurred via a photon pathway
3. Sometimes literature defines this as electronic collimation, yet I have a problem with this definition
4. When coincidences occur, a straight line can be drawn that connects both ends of the annihilation event as it strikes the opposing detectors. This straight lines are referred to as Line of Response (LOR). To demonstrate this concept 4 LORs are drawn through a hot spot within a tumor demonstrated as a transverse slice
5. LORs are stored in the computer as sinograms (that's the next lecture)
  1. The hot spot is actually comprised of many LORs, all of which occur from individual annihilation events
  2. In relationship to regular nuclear medicine an LOR would be equivalent to count (or gamma event) being detected on a gamma camera (1 LOR = 1 count)
  3. As these LORs overlap each other, counts are recorded and the count density increases eventually defining a hot lesion. The more LORs the greater the resolution. In the above example, detailed information can be visualized, not just of the hot spot, but also of the surrounding tissue. While there is less concentration of the radiopharmaceutical around the lesion (this could just be background or another structure), LORs are still produced and recorded to the point where there is enough data to visualize the entire axial projection. You could also consider the concept of target to background. Question - Are there LOR in areas that we define as background?
  4. Concept - When an LOR is generated these means that the location of the annihilation event occurred somewhere along the path of coincidence. Therefore, the exact location is not known. In a future lecture we will discuss "Time of Flight" (TOF) which greatly enhances our ability to determine the approximate location of the annihilation event
6. The PET/SPECT comparison
  1. In conventional SPECT imaging location of the gamma event can only be defined by lead septa plays the role of collimating the incoming photon
  2. Collimation in SPECT via lead septa reduces count rate which effects reduces resolution
  3. This is one reason why PET can resolve around 5 - 10 mm, while SPECT has a the range of 15 - 20 mm
  4. In addition, this improve resolution, is due in part, to increased counts via the removal of lead septa causing an increase in the amount of photon flux being detected by the imaging system

The Crystal - Is yet another influencer on PET resolution. Related to this topic is to remember there is no such thing as an ideal crystal

Crystal density has to increase since we are dealing with a more energetic photon - Increased density should translate to greater "stopping power" and improved efficiency in capturing the incoming photon
Increasing Linear Energy Transfer (LET) Increases crystal density. Do you recall where we usually apply LET?
The amount of light being produced by the scintillation event translates to a greater amount of light means better detection
Variations of light translates to the amount of keV of the photon gives better sensitive
Acquiring a short decay constant, or how quickly can the electron move back into its normal energy state is yet another critical factor - if you shorten deadtime you increase the amount photon detection
The question that needs to be considered is, why is NaI(Tl) not a viable option in the use of PET? The answer, read on, I think you will figure it out
Why is the use of NaI(Tl) not a good idea?
1. Its density is too low to capture 511 keV photons efficiently - to make an efficient NaI crystal it would need to ~ 5 inches thick (thickness of the old rectilinear scanners had this technology)
2. Low (LET) and low density means poor "stopping power"
3. Stopping power is defined as the mean distance a photon travels until it stops completely, depositing the energy in the crystal which causes scintillation

Consider the table below and review the different crystal characteristics: Density, scintillation decay, photon yield, light output, energy resolution, and LET

Properties*	NaI	BGO	GSO	LSO	LYSO	LaBr₃	BaF₂
Density - gm/cm³	3.7	7.1	6.7	7.4	7.3	5.3	4.9
Scintillation decay time in ns	230	300	60	40	42	15-25	0.6-0.8
Photon yield in keV (%)	38	6	10	29	~75	63	2
%Light Output	100	15	27.5	25-30	40	1-12	160
% Energy Resolution at 511 keV	6.6	10.2	8.5	10	14	N/A	10
Linear attenuation Coefficient (μ)	0.34	0.95	0.7	0.83	0.87	0.48	0.45

BSO is Bismuth Germinate, LSO is Lutetium oxyorthosilicate with Ce, BaF₂ Barium Fluoride, LaBr₃ Lanthanum Bromide, LYSO Lutetium Yttrium Oxyorthoscilicate
*One thing that I have come across regarding the above values is that there is some variation in the literature. As an example I refer you to http://www.sinocera.net/en/crystal_lyso.asp

Density, and μ all relate to the crystal's stopping power. In general, as the density increases so will the stopping power
How thick does does NaI have to be in order stop a 511 keV gamma?
1. Keep in mind that a gamma camera usually has a crystal thickness of 3/8 inch or about 9.5 mm
2. If NaI has 9.5 mm thickness the probability of detecting a single photon event is ~12%. To record the coincidence photon event it drops to 1.4%
3. If you double the thickness of the crystal you increase the detection rate to 26% for the first event and coincidence detection is about 6.8%
4. If 30 mm of BGO is used the initial gamma detection is 70% with a coincidence detection of 49%
5. Why is BGO stopping significantly better than NaI? (look at its μ and its density)
Consider thickness of the crystal and its density. Hypothetically the thicker a crystal is the poorer its energy resolution, however, the greater its stopping power. Therefore, increased density and reduced thickness should improve spatial resolution
Based on μ values of the different crystals, NaI would have to be more than twice as thick to be able to have the same stopping power as BGO or LSO
Opacity of the crystal is another crucial role because improved crystal clarity allows for better transfer of light scintillation
Photon yield in keV relates to the system's sensitive and how many photons are released from scintillation when the 511 keV photon is absorbed. The greater the yield the better sensitivity the crystal has to keV variation
Light output effect image resolution, therefore the greater the light output the greater the system's sensitivity
Energy resolution - Greater the energy resolution the better the spatial resolution
Scintillation decay time relates to dead time which is the crystal's ability to capture two distinct photon events. Should the crystal be in the process of scintillating and another gamma ray comes in it cannot be recorded. Therefore, the shorter the decay the better, since it can more quickly response to the next incoming event

Further review of the crystals
1. For many years BGO has been the crystal of choice and the reason for this is its stopping power. However, compare BGO to the other crystals and evaluate: μ, photon yield, energy resolution, and light output. Is there a better candidate?
2. HOMEWORK - Break down the above characteristics of a crystal and rate them in descending order: decay time, photon yield, energy resolution and linear attenuation. See
3. Important when considering PET/MR - A typical PMT will not function in a magnetic field, therefore semiconductor technology must be applied

Photomultiplier Tube and the crystal
1. A look at the PET crystals from Sinocera. Notice the clarity
2. When comparing PET to a gamma camera technology, the role of the PMT is the same
3. When the crystal and PMT are combined together with PET this concept evolves (This is an example of design 1)
4. Blocks Crystals and PMT Design
  1. Detector Blocks
    1. A PET gantry is composed of multiple rings
    2. In each ring are a series of blocks
    3. Within these blocks are is the scintillation crystals and attached to those crystals are the PMTs
    4. Crystals and PMTs are arranged in some type of 'block' arrangement that shares the location of the scintillation
    5. The question here is, what is the most efficient setup for the crystal and PMTs?
  2. Block designs
    1. In general, when a photon scintillates with the crystal, X, Y, and Z coordinates are identified by the specific element within the crystal block. In turn, the detector identification table reads its location and sends that data to a flood histogram.
    2. What is a flood histogram? A QC procedure referred to as normalization that acquire the actual locations of coincidences throughout each ring to generate uniformity corrections within each block. The histograms generate a detector identification map (See normalization scan). Hence the flood histogram is a location circuit and a correction matrix from the acquired scintillation
    3. Process of detection - scintillation --> X Y location -->PMT and system amplification -->PHA -->Coincidence Timing Circuit -->LOR--> Sinogram
    4. In some systems blocks are placed into cassette or buckets and in order for coincidence to occur it must be recorded in opposite cassette/buckets
    5. In design 1 all 4 PMTs are all within the the crystal block
      1. Prior to this design larger PMTs where placed on a sets of panels and placed into the detection ring
      2. By decreasing the size of the PMTs, adding cutting "elements" into crystal, reduced the amount of PMTs needed: camera resolution improved and manufacturing costs went down
      3. Each PMT is responsible for its scintillation and location is determined within each element
      4. Dead-time is less with this design when compared to design 2
      5. Elements within the crystal helps reduce the amount of PMTs needed, therefore reducing the system's cost
    6. In design 2 the PMTs overlap the crystal
  3. Pixelated (Quadrant sharing)
    1. Set of PMTs are arranged in a panel
    2. Crystal is composed of GSO
    3. Because of the PMT arrangement each scintillation event is seen by 7 PMTs and a more exact location is identified
    4. Advantages indicate¹
      1. Better positron location
      2. High sensitivity
      3. Improved count-rate
5. Blocks and Rings
  1. Blocks encompass 360 degrees of detection around an axial projection which allows it to capture the coincident event
  2. There are different designs with these PET rings, and three different examples are noted. If you would like to know more about them place your cursor over the image and click it to go to the appropriate link
  3. The more popular design if a circular one
  4. With the PET gantry there are between 18 to 32 rings that acquire in the axial projection. Consider the advantages of more rings
  5. The average size of an axial acquisition is 70 cm
  6. In a situation where the a whole body image has to be acquired (or where the area of interest is greater than a single axial acquisition) the table moves the patient into appropriate position, acquires the data, and then moves the patient down to the next position for further data acquisition. This process continues until the entire imaging sequence is done
  7. While there is no such thing as collimation, there is lead between each ring which helps reduce the recording of cross-talk between the rings. This improves resolution and reduces count sensitivity
6. PET Gantry Overview
  1. While all PET detectors have the same components to detect the incoming photons there is a significant difference on how much there is of each component. Look at the above image for the comparison
  2. In general, one should be able make some assumptions
    1. The greater the axial dimension the more the rings
    2. More rings means more crystals and PMTs
    3. Consider the different factors above and how it may effect and/or improve image quality
  3. Yet another type of PET gantry can image a whole body scan in one bed
  4. Another view of the gantry with the components labeled
7. The Coincidence factor
  1. The ability to detect the two 511 keV gammas and define the specific locations of the event is unique to a PET scanner
  2. Each element within a block is connected to a coincident circuit which is connected to all other elements in the opposite location within the ring, located in the same block. However, there is more ..
  3. The image above shows how angles of coincidence detection can occur in the x, y, and z planes
    1. The above box is emitting photons in the x and y axis (left) which demonstrates how one specific element within a block can share its event with a multitude of elements/blocks in the coincident direction
    2. In the Z axis one element/block within a block (colored red) can share the opposing coincident event with any element/block on the opposing side (all blocks are red)
8. Timing Window
PHA - Window

Essentially the same as a gamma camera, if a photon fall above or below the window it is rejected and within the window it is accepted
What differs from a gamma camera perspective are the LLD and ULD settings. With PET the keV settings is usually set between 350 and 650 keV. This translates to a 60% window
Here is some food for thought!
1. Reducing the window improves energy resolution, by eliminating unwanted scatter
2. However, limiting the window also reduces counting efficiency
3. There are two rejection circuits in PET - PHA and timing window

Return to the beginning of the document
Next Lecture - LORs and Imaging
Return to the Table of Content

1/24