Medical Imaging System (2024 fall) - Exercise 6 Instructions/Questions

Due: 11:59 pm, Dec 14 (Sat), 2024 (By E3)

In the first part of this assignment, you will answer questions related to filter backprojection
reconstruction of X-ray data.

Figure 1:
Image of the original object to be used for the exercise –– the program will supply the result.

Figure 2:
Sinogram (i.e., projection) of the original image computed at 202 angles equally distributed over
180 degrees and 128 lateral positions at each angle –– the program will supply the result.

Fig 1. Image of the original object Fig 2. Sinogram of disk phantom

QUESTION 1: Using figures 1 and 2, justify why the sinogram looks as it does given the
original object image in figure 1 and the sinog
ram in figure 2.

The sinogram in Figure 2 is created by collecting projections of the object in Figure 1 (Disk
Phantom) at various angles (from 0 to 180°). Each circle contributes a sinusoidal wave with
unique frequency, amplitude, and brightness.
Outer Circle (Largest):
The center of the outer circle is aligned with the center of the image, meaning its projection
position does not shift significantly with changing angles. As a result, its sinogram spans
approximately positions 20 to 110. Its sinusoidal wave has a near-zero frequency due to the
uniformity of its projection across all angles, producing a very smooth and flat pattern.Gray

1
Circle (Second Largest):
The center of the gray circle is slightly offset from the image center, located at approximately
[−25,12.5].As the projection axis rotates, this offset causes a sinusoidal wave to appear in the
sinogram, with a slightly higher frequency than that of the outer circle.The brightness of this
wave is higher than the outer circle's due to the gray circle's density being twice that of the outer
circle.
White Circle (Smallest):
The white circle is located at [25,−25] from the image center.Its offset position results in the
highest-frequency sinusoidal wave in the sinogram.Its brightness is the highest because the white
circle's density is four times that of the outer circle and twice that of the gray circle.
Between projection angles 30 and 100, the sinogram shows the brightest regions. This is because
all three circles overlap in projection within this range, leading to maximum intensity in the
sinogram. In conclusion, the sinogram is a superposition of sinusoidal patterns from the three
circles, with the brightness and frequency determined by their geometry, position, and density.

Figure 3:
Plot the 0th moment of the projection at each angle – the program will supply the result once
you provide the correct expression for the 0th moment.
QUESTION 2: The 0th moment is defined as the area under the curve of a function. Compute
the 0th moment of the projection as a function of angle using the sinogram function created in the
program. Plot this result in Figure 3.
Provide an analytic proof regarding the shape of the zeroth moment of the projection as a
function of angle you see in Figure 3. You do not have to justify the magnitude of this function -
it's related to the details of the object, but the shape you can show analytically.

Fig. 3 0th moment projection

2
Physical Meaning of the Zeroth Moment:
The zeroth moment represents the total density of the object
integrated over all projections at different angles. Mathematically,
this is equivalent to the integral of the object density. If the object’s
total mass (or density) is uniform and finite, M0(θ) will remain
constant across all angles θ. In Figure 3, the zeroth moment
appears as a horizontal line. This matches the conclusion of the
analytical proof: the symmetry and conservation of total mass
lead to a constant M0(θ).

Figure 4:
Display the spectrum of the filter function used for filtered backprojection – you must provide
the code to produce this spectrum from spatial frequencies ranged from – to  over nr
pixels - the program will supply the figure.

Figure 5:
Compute the filtered sinogram. The program will supply the final plot once code has been
added to compute the proper filter function.
QUESTION 3: Add code to provide the ramp function specifying the Fourier Transform of the
filter function required for filtered backprojection. This function will be plotted in Figure 4. Once
you have the correct filter function, the program will compute the filtered sinogram in Figure 5.
Contrast the unfiltered sinogram in Figure 2 with the filtered sinogram in Figure 5. What are the
major differences and similarities?

Fig 4. Ramp up filter function Fig 5. Sinogram after filtered backprojection

Similarities:
1. Structure Retention: The overall structure of the sinogram remains the same. Both
sinograms represent the projections of the object from different angles.

3
 2. Projection Geometry: The locations of the main intensity peaks in the sinograms are
consistent, as they correspond to the same object.
Differences:
1. Noise Reduction: The filtered sinogram (Figure 5) has reduced noise compared to the
unfiltered sinogram (Figure 2).
2. Sharpness: Boundaries and edges in the filtered sinogram are much sharper due to the
ramp filter amplifying high-frequency details.
3. Low-Frequency Suppression: The filtered sinogram suppresses low-frequency
components, removing gradual intensity variations that obscure details.

Figure 6:
Compute the reconstructed image by backprojecting the filtered sonogram of Figure 5. The
program will supply result.
QUESTION 4: Compare and contrast the reconstructed image in figure 6 with the original
object in Figure 1.

Fig 6. Reconstructed image

Similarities:
1. Shape and Structure:
o The reconstructed image (Figure 6) retains the overall shape and geometry of the
original object (Figure 1), including key features like circular or elliptical regions.
o Major boundaries and regions of contrast in the original object are well-
reproduced.
2. Gray-Scale Intensity:
o Intensity variations in the reconstructed image match those of the original object.
Bright regions in Figure 1 correspond to bright regions in Figure 6.

4
Differences:
1. Blurring and Artifacts:
o Figure 6 may have slight blurring at the edges due to the interpolation in the
backprojection process.
o Artifacts, such as streaks or rings, may appear in Figure 6, especially if the
filtering or the sinogram resolution is suboptimal.
2. Noise and Smoothing:
o The reconstructed image might appear slightly smoother or less detailed in
regions of low contrast due to the suppression of noise and low-frequency
components by the ramp filter.
3. Reconstruction Resolution:
o Depending on the number of angles and the resolution of the sinogram, Figure 6
may not perfectly match the resolution of Figure 1. Increasing the number of
projection angles improves the resolution.

In the second part of this assignment you will analyze a simple 1-D Anger Camera used to
spatially localize incident gamma-rays in nuclear medicine. The geometry for the Gamma
Camera is shown below.

All relevant spatial information is included in the program, where “depth” is the distance from
the center of the scintillator (assume the scintillator is a plane – i.e., gamma-ray photons are
stopped at one depth to produce the optical photons detected by the photomultiplier tubes),
“left_camera_position” is x1 and “right_camera_position” is x2, the total camera extent is
“x_position”, and there is no gap between the two tubes at the center of the camera.

5
Figure 7:
Compute estimated position of the captured gamma-ray versus the actual position of the entering
gamma-ray. You will have to add code to compute this function but then the program will
compute the graph.
QUESTION 5: Compute the Gamera Camera position estimate for a 1-D camera consisting of
two photomultiplier tubes. Assume for this simple 1-D case that the output from the camera at
any position is simply proportional to 1 over the distance between the photon capture site and the
position on the camera. The output from each photomultiplier tube is simply the sum of the 1
over distance measure to each point on the surface of that tube. There are two outputs from the
camera representing the intensity from each tube. You must add code to compute the estimated
position using the standard Anger Camera formula and compare it to the actual position of the
captured gamma-ray in Figure 7. Does the estimated position equal to the actual position? If not,
why not? Also, given the function in Figure 7, can the output position estimated by the Anger
Camera be used to determine the true position? Explain your answer.

Fig. 7 Anger camera position estimate

The plot demonstrates a clear linear relationship between the photon capture lateral position and
the Anger Camera position estimate, as expected. The estimated position is increasing with the
lateral position of the photon, which matches the expected behavior of the Anger Camera.

Does the estimated position equal the actual position?

1. No, they do not match perfectly.
o The estimated position deviates from the actual position due to the inverse-
distance dependence of the photomultiplier outputs. This formula assumes an
ideal response, but the interaction geometry introduces systematic non-linearities.
2. Non-Linear Distortion:

6
 o Near the center (between the two tubes), the estimated position is fairly accurate
because the distances to both tubes are similar.
o As the gamma-ray moves closer to one tube, the estimate tends to
overcompensate, leading to non-linear distortion.

Can the estimated position determine the true position?

1. Yes, but with corrections:
o The relationship between the estimated and actual positions is systematic and
predictable. A calibration curve can be constructed from Figure 7 to map
estimated positions back to actual positions.
2. Limitations:
o Near the edges (far from the tubes), the error increases because the signal from
one tube dominates, reducing the resolution and accuracy.
o This behavior limits the range of accurate position estimation without corrections.

7
8