Exam 3

The amount of attenuation from 10 cm depth for 99mTc is approximately A. 1% B. 25% C. 75% D. 99%

(Knowing that 99mTc 1st HVL is 4.5 cm), ○ 1st HVL: 10 / 2 = 5 cm ○ 2nd HVL: 5 / 2 = 2.5 cm. ○ Since this number is somewhere in the middle, we add the two to get 5 + 2.5 = 7.5 Amount attenuation = 7.5 / 10 = 0.75 x 100% = 75% attenuated from a 10 cm depth for 99mTc

-----

-----

-------

---------

4 SPECT characteristics that are better than planar (1 is effectively the same)

1. Improvement in image contrast (differentiating shades of gray is better in SPECT) 2. Improved anatomical localization (planar: only A/P) 3. Effectively the same spatial resolution (effectively the same. planar is actually a bit better because in SPECT we can't get as close to the patient due to the orbit of the detectors) 4. Depending on the filter used, variable improvement in noise

Iterative reconstruction allows modeling in the algorithm to compensate for factors that degrade the reconstruction. Give some factors that degrade resolution but that can be fixed with iterative reconstruction.

1. Resolution loss with depth due to use of collimator 2. Attenuation of photons 3. Scatter 4. Patient Motion

Dot he math for determining 128 million counts for a uniformity flood on a 128x128 matrix

128x128 = 16384 ~ 16000. FOV is rectangular, which results in 3/4 of the matrix being used, thus 16,000 x 3/4 = 12000 To obtain a 1% STD, you need 10,000 counts/pixel 10,000 x 12,000 = 120 million counts

What is the total # of counts in a SPECT study?

2-8 million counts

Sinogram Describe how many sinograms would be in a matrix

2D image which displays all projections of a single slice. If there is a 64x64 matrix, then there are 64 lines across which means 64 projection profiles and thus 64 sinogram's (1 sinogram per slice)

Uniformity flood requirements What does this mean for 64x64 and 128x128?

30-100 million counts for flood imaging Need at least 10,000 counts/pixel 64x64 - 30 million counts 128x128 - 120 million counts

A COR offset greater than 0.5 can create 2 defects:

360 degrees - donut "ring" artifact 180 degrees - pitchfork artifact - (cardiac 90)

CT Attenuation Correction

512x512 matrix representing a map of average linear attenuation coefficients. To correct for attenuation in a SPECT study, these attenuation coefficients are used for each projection and utilized in the iterative recon method

Do the math for determining 30 million counts

64x64 = 4000 pixels. FOV is rectangular, which results in 3/4 of the matrix being used, thus 4000 x 3/4 = 3000 To obtain a 1% STF, you need 10,000 counts/pixel 10,000 x 3,000 = 30 million counts 16384 = 12288

Circular "ring" artifacts (often called "bullseye artifacts") in tomographic reconstructions are the result of: A. Uncorrected flood field non-uniformities B. Inadequate statistics C. Insufficient angular sampling D. Using an inappropriate filtering algorithm E. Insufficient linear sampling

A C and E are when you don't have enough stops

A filter will never affect the _________, it just affects the _________

A filter will never affect the frequency, it just affects the amplitude

Hanning Filter

Adjusts what the *cuttoff* is based on your frequency. If the cutoff is a high frequency, then there are sharper edges but more noise. If the cutoff is a low frequency, the image is more smooth but more blurry.

Ramp Filter (High Pass Filter)

Applied before every low frequency filter. This filter amplifies high frequency data and minimizes (or suppresses) low frequency data. This results in removal of the blurriness from the unfiltered back projection and restores edges to the 3D image. A pitfall is that it's amplifying noise. This is why we typically don't use the ramp filter by itself

Transmission Scan RP Advantage Pitfall

As the transmission source (Gd-153) scans the object, it will give off radiation that is detected by detectors on the opposing side of the transmission arms. Advantage is that Gd-152 has different energy windows than 99mTc which allows the computer to depict Tc information versus Gd information Pitfall: decays after time. Thus, at approx one year, there won't be as many activity (counts given off) as there were when we originally received it.

Angular Sampling

As you decrease the number of projections acquired, you increase the amount of noise because you aren't getting enough information per projection

Alignment QC in SPECT/CT Importance of the source used When is this performed Important for CT for 2 reasons:

Assessment of whether the fusion of the CT and SPECT is within acceptable limits Gd-153 is important because it can act as a radio dense (CT) and radioactive (SPECT) source to see if the fusion between the 2 is correct Performed monthly Important for CT for 2 reasons: 1. Attenuation correction 2. Localization information

Why is attenuation correction important? For what study is it more important and why?

At depth, there is a greater % of emitted photons attenuated which will appear to have reduced counts without attenuation correction. Most important in heart due to presence of soft tissue, lung, and bone, each of which attenuates to a different degree.

Patient motion may be most easily detected in: A. Individual frames prior to reconstruction B. Cine display C. Reconstructed images

B

Which of the following components has the greatest influence on both the spatial resolution and detection efficiency of a SPECT system? A. Scintillation crystal B. Collimator C. Image display system D. Reconstruction filter

B. .05% of photons actually interact with the detector due to the collimator, attenuation in the patient, etc

Fournier Transformation

Because back projections lead to blurry images, we need a mathematical algorithm (called the Fournier Transformation) to remove this blurriness. The Fournier transformation converts 2D data (sinogram) to spatial frequency domain data Recap: Fournier Transformation converts 2D data to spatial frequency domain data

What would be 2 instances where you would use a 128x128 matrix?

Bone scans - they have anatomy with high contrast Using software that can account for resolution recovery (which is typically what is always done now. Thus, if we acquire in 64x64, iterative reconstructions process to a 128x128 matrix.)

The center of rotation look up table A. Should be performed daily B. Interprets the patient's axis of rotation C. Compensates for slight misalignments D. Is the same as the COR test

C

To evaluate uniformity of the SPECT system you would: A. Acquire a 30 million count flood B. Acquire floods at every 10 degrees C. Acquire and reconstruct a cylindrical source of activity D. Acquire and reconstruct a point source of activity

C A would be for planar imaging

3 QC Tests that should be performed with SPECT Imaging?

COR - make sure the camera is still holding up and correcting properly. Another way to check for COR is using a level and checking the head tilt. SPECT Jaszack Phantom - looks at uniformity, low contrast resolution (spheres), and high contrast resolution (rods) throughout the phantom

COR calibration Name the other term (s) Define Performed when? Source (s)

COR Calibration / COR Correction / COR Study Correction matrix (look up table) acquired that shifts data. Done by manufacturer, engineer, or physicist. Performed annually. Uses 3 sources to do this test (triangular data)

COR Evaluation Name the other term (s) Define Define x-offset Performed when? Source (s)

COR Evaluation / COR Test Determines if the camera is still holding up and correcting properly. Data is compared to the COR calibration data to see if anything has changed. Does not create a new map, just compares the data. x-offset: if this is measured greater than 0.5 pixels, the stored values should be cleared and a new COR calibration should be determined to create a new look up table Performed monthly and need to be done according to degrees (180 versus 90) and collimators (if SPECT is performed on those collimators) Uses 1 source

Matrix of CT versus SPECT

CT: 512x512 SPECT: 64x64 or 128x128 The CT matrix is averaged to match our SPECT reconstruction matrix

Chang Attenuation Correction Can this be used with iterative reconstruction?

Chang assumes your slice is a uniform distribution (typically attend. coe. of water) and applies the same linear attenuation coefficient (0.15 cm^-1, sometimes 0.14 cm^-1) r to every single pixel, despite the varying densities in the image. Chang is only used with FBP and can't be used with iterative reconstruction.

Ring Artifact Can be due to 3 things Single head versus dual head

Consequence due to the rotational nature of SPECT imaging Can be due to flood non-uniformities, COR errors, and collimator defects Single head 360 degrees: ring Dual head 360 degrees: arc

For many low pass filters, what is the one variable that can be modified? Define this

Cutoff frequency. This is defined as the frequency in which the filter function reaches zero. The cutoff determines the maximum frequency in the filtered image. For butterworth, the cutoff is the maximum frequency (i.e. typically the value of the Nyqust. Can not exceed). For hanning, the cutoff is based on the frequency Important to keep in mind that its when the filter reaches an AMPLITUDE of zero. This point is the cutoff frequency. (amplitude: y axis. frequency: x axis)

Define COR error

Defined as an offset during backprojection

Nyquist Frequency What is the rule of thumb? Equation?

Defines the maximum spatial frequency that may be acquired (2 other ways of looking at this: maximum difference between 2 pixels in an image, highest frequency on the graph). The rule of thumb is that you need at least 2 pixels to display a cycle First, you need to determine your pixel size. Pixel size = FOV/matrix size. Once you have your effective pixel size, then nyquist frequency = 1 / (2)(pixel size in mm) ^^ thus, the Nyqvist frequency is determined by the pixel size and number of pixels

Filtered Backprojection

Filtered backprojection occurs because we know that there is a loss of resolution and contrast as we backproject. The image is made up of low frequencies and to account for this loss we apply a ramp filter. The ramp filter accentuates high frequencies and suppresses low frequencies which helps compensate for the original low frequencies. The pitfall to this filter is noise is added, thus we apply a low pass filter to smooth the data and get rid of some of these high frequencies.

Assume you have a FOV of 20 cm (200 mm) and the FWHM resolution at 10 cm depth is 12 mm. Determine matrix size

First, determine effective pixel size i. 12 mm x 1/3 = 4 mm Then, you can determine matrix size i. 200 mm / 4 mm = 50, so the appropriate matrix to use would be 64x64.

Matrix size equation

First, may need to determine pixel size Pixel size = FWHM x 1/3 Then, matrix size Matrix size = FOV / pixel size

What is one instance in which we would just use the ramp filter?

For a QC phantom, where we can acquire as many counts as we please (50 million counts. Could acquire for hours. We can't do this with patients because they can't sit there that long)

Approximate how the time per stop would be determined for a given study.

For the given number of stops, you want the total acquisition time to not exceed 20, at most 30, minutes so this determines how much time, accounting for head movement between stops, can be allocated to each projection. Note that dual head allows more counts, faster acquisition, or a tradeoff between the two.

Ray Artifact

From FBP reconstruction

GE What its called What software is correlated with it

GE: EVOLUTION Reconstructs with OSEM, resolution recovery (accounts for how far the camera is from the patient), attenuation correction (CT data), scatter correction (having a window around another photo peak and getting rid of anything within that energy range)

Amplitude Why is amplitude important?

How much of a particular signal (i.e. the height) Amplitude is important because it is the factor that is always changing in filters. Amplitude is the y axis and frequency is the x axis.

If you don't have enough stops/projection in a SPECT acquisition, what are 2 things that can occur

Insufficient linear sampling Insufficient angular sampling (You made up this question)

Compare Iterative with FBP

Iterative does not manipulate the data obtained, it uses it for comparison. Iterative is better able to handle low data (low statistic) information. Iterative can model corrections, to include resolution loss with depth and attenuation correction and also scatter correction.

Low resolution SPECT. Matrix size and how many stops High resolution SPECT. Matrix size and how many stops How does dual headed systems change this?

Low resolution SPECT: 64x64 matrix, 60 images every 6 degrees (360/6 = 60) Dual head: 30 images every 6 degrees per detector High resolution SPECT: 128x128. 120 images every 3 degree stops (360/3=120). Dual head: 60 images every 3 degrees per detector

Examples of: Low spatial frequencies Middle spatial frequencies High spatial frequencies

Low spatial: large organs (liver) and background Middle: desired object information (edges, anatomy of interest) High: sharp edges (bone), noise

Equation for matrix size

Matrix size = FOV / pixel size

What does MLEM stand for

Max Likelihood Expectation Maximization (Type of iteration)

Spatial Frequency Domain How do we convert useful information into spatial frequencies?

Measure of the change in count densities from pixel to pixel across a matrix The Fournier transformation converts 2D projections into the spatial frequency domain. From there we can determine the nyquist frequency

More iterations =

More iterations = more noise in the image. Need to find a balance between how many iterations is good/too many

Results in blurring due to activity being placed in 2 different locations in the 3D image matrix

Motion artifact

What artifact will occur in SPECT on uniformity analysis if we have the wrong uniformity map?

Multiple ring artifacts

What spatial frequencies (low, medium, high) have noise?

Noise is throughout low, middle, and high spatial frequencies but it only becomes an issue at high spatial frequencies (think back to the graph. Noise should increase at the end - where the high frequencies are)

Number of stops rule of thumb

Number of views should approx. equal the number of pixels (or matrix size) used to acquire the image (i.e. 64x64, at least 60 views)

Truncation Artifacts

Occurs in transmission based attenuation (CT, transmission scan) when the patient's entire body is not seen in all projections (NM field of view is bigger, so many be seen in the NM data but there is no CT data for these outside areas. Thus, the CT can't perform any attenuation correction in that area)

What does OSEM stand for Describe it

Ordered Subset Expectation Maximization (Type of iteration) Same idea as MLEM but OSEM breaks different sinograms up into subsets or sections and reconstructs them at the same time. Benefit is this is faster than MLEM.

Phillips Whats it called What software is correlated with it

Phillips: ASTONISH Reconstructs with OSEM, Resolution recovery (accounts for how far the camera is from the patient), attenuation correction (CT data), scatter correction (having a window around another photo peak and getting rid of anything within that energy range

Equation for pixel size

Pixel size = FWHM x 1/3

What filter is typically used with FBP?

Ramp, then Butterworth filter (cutoff, power)

SPECT Jaszczak Phantom Where are the rods and spheres What do the 3 areas measure How much activity is within the phantom When is this performed

Rods - bottom Spheres - middle Spheres - measures low contrast information (low contrast resolution) Rods - measures high contrast information (high contrast resolution) Water portion (top) - measures uniformity 25-30 mCi. 6 hours before you measure the phantom, you put 60 mCi of Tc in and roll the phantom throughout the day to let it disperse. Performed quarterly

99mTc Rule of Thumb for HVL

Rule of thumb for 1st HVL of 99mTc: Every 4.5cm will be the 1st HVL for 99mTc

Unfiltered Backprojection (Simple Backprojection) How do we correct for this?

Smearing "paint-rolling" the data across the matrix in every projection. The pitfall is the image becomes blurry with edges hard to depict and counts at incorrect locations. These images use low frequency data and to correct for this we apply a ramp and low pass filter. The ramp filter compensates for the overuse of low frequencies in the UBP by accentuating high frequencies and suppressing low frequencies But, ramp filters add noise. To reduce noise, the computer uses a low pass filter to smooth data and keep noise out of the reconstructed 3D image matrix.

Gd-153 T1/2 Energies

T1/2: 241 days Energies: 70, 97, 103 keV

Butterworth Filter

The *cuttoff* in Butterworth is the highest frequency. This is often given as a decimal of the Nyquist frequency. The *power* is the rate at which the weighting "rolls off" (how the curve changes) and thus determines how much each of the frequencies is weighted in the final reconstruction

What eliminates scatter?

The PHA (NOT the collimator. Scatter can still get through the collimator). But, the PHA throws the scatter information out

What makes the 128x128 matrix possible in Astonish?

The ability to use resolution recovery

If there is misalignment in a dual detector SPECT image, when reconstruction occurs, what images will show the biggest offset?

The cine images.

Iterative Reconstruction

The computer starts out with a uniform distribution (every pixel has the same activity). The computer will guess that this is what the projection is going to look like (estimated projection). Based off of this estimated projection the computer made, its going to compare that estimated projection with the measured projection (measured projection is a sinogram) which is what the technologist imaged. During the "compare" process, there are mathematical equations that the computer is going through to create a new guess projection or otherwise known as an error projection (a new sinogram). It will backproject to make an error image, and the error image will update and now be the new guess.

Matrix Size rule of thumb

The dimension of a pixel should be approx. 1/3 the FWHM at depth (Example: if you have a collimator resolution of 10.53, then you take x 1/3 of it to get a little over 3. That becomes the effective pixel size which you can use to determine the matrix size if you know the FOV

Why do we acquire cardiac's in 180 degree projections?

There is no additional information given in the posterior right side of the patient. It just degrades the image

If you exceed the resolution capabilities of the system, what benefit does it give by going up a matrix size?

There is no benefit. You will just have less counts per pixel

In broad terms, what does the Butterworth and Hanning filter do? What are the parameters within these filters?

They reduce the noise that was accentuated with the ramp filter. This results in smoothing of the image Parameters: Butterworth: Cutoff and Power Hanning: Cutoff

TBAC

Transmission based attenuation correction This can be represented for the transmission scan as well as CT.

T or F? Incorrect COR in a clinical study often produces very subtle artifacts

True

UltraSPECT Whats it called What software is correlated with it

UltraSPECT: Wide Beam Reconstruction Used for cardiac recons (Brightview). Iterative with noise reduction and resolution recovery. Supposed to allow imaging at half the activity or half the time