Scribd
Upload
44 views19 pages

PET Image Reconstruction

This document discusses image reconstruction techniques for small PET systems. It outlines algorithms like statistical reconstruction using a system response matrix (SRM) and incorporating time-of-flight information. Images are shown for different detector spacings and a real 4x4 detector. Fast iterative shrinkage-thresholding (FISTA) is also introduced. Phantom and source descriptions are provided for image reconstruction of a large dataset.

Uploaded by

Anonymous uw72xEZT
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
44 views19 pages

PET Image Reconstruction

This document discusses image reconstruction techniques for small PET systems. It outlines algorithms like statistical reconstruction using a system response matrix (SRM) and incorporating time-of-flight information. Images are shown for different detector spacings and a real 4x4 detector. Fast iterative shrinkage-thresholding (FISTA) is also introduced. Phantom and source descriptions are provided for image reconstruction of a large dataset.

Uploaded by

Anonymous uw72xEZT
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 19

Image Reconstruction for

Small PET System

1
Outlines
• Image Reconstruction Algorithm and Detector Evaluation

• Image reconstruction using SRM (Using geometric method)

• Image reconstruction with Time of Flight Information

• Image variation with detector spacing

• Image of real 4 x4 Detector

• Fast Iterative Shrinkage-Thresholding Algorithm (Ke-Chia)

• Summary

2
Image Reconstruction Algorithm and Detector geometry

Maximum Likelihood Estimation Method 3


Detection Efficiency and sensitivity
Detection Efficiency alone the Y dimension

Sensitivity of PET: Total number of coincidences to total detected events.

4
Image Reconstruction for a large Data Set
Phantom:
A cylinder placed along the X-Axis.
Radius=10mm Height=40mm location center of the mother volume.
Sources:
S1: Activity 986640 decays per secondsx1200, R=1.2mm, height=16 mm
Coordinates (0,0,0)
S2: Activity 586640 decays per secondsx1200, R=1.1mm, height=4 mm
Coordinates (0.0,-5.0,0)
S3: Activity=586640, decays per secondsx1200, R=1.1mm, height=5 mm,
Coordinates (0.0,5.0,0)
5
Iteration # 10

Voxel size=0.8x0.8x0.5 𝑚𝑚 3 Voxel size=0.8x0.8x0.5 𝑚𝑚3

6
Source and Phantom Description
Phantom
Volume : Cylinder
Material: Water
Diameter 20mm
Height 40mm

Source
Activity 90000*100
Diameter 14 mm
Height 20 mm
Particle Type Positron
Isotope Oxygen-15
Volume: Cylinder

7
Images by Change in detector spacing Voxel size=0.8x0.8x0.5 𝑚𝑚3
Detector Spacing =6cm Detector Spacing =4 cm

3 cm
Voxel size=0.8x0.8x0.5 𝑚𝑚3 Voxel size=0.8x0.8x0.5 𝑚𝑚3

8
Detector Spacing =4cm Detector Spacing =3cm

Voxel size=0.8x0.8x0.5 𝑚𝑚3


Voxel size=0.8x0.8x0.5 𝑚𝑚3

9
Accounting the time of Flight Information
Voxel size=0.8x0.8x0.5 𝑚𝑚3 TOF Image
Voxel size=0.8x0.8x0.5 𝑚𝑚3
time resolution=10 ps

10
Detector Schematic
Source Properties

• Source Na22

• Diameter 25 cm Double of the detector size

• Height 0.5 m

• We are not very sure about the position of the phantom and the
separation been the detector planes precisely.

• We are not sure about the actual size of the isotope strip inside the
Na22 source

Following Data Files are used

• One from Calibration run 25 mm

• One from the simulated data 11


Reconstructed Image from the simulated Data

12
Image for 4 by 4 Detector with real data
Detector Separation 6 cm Geometric SRM Detector Separation 6 cm Geometric SRM

Voxel size=0.8x0.8x0.5 𝑚𝑚3 Voxel size=0.8x0.8x0.5 𝑚𝑚3

13
FISTA
• This algorithm is based on gradient descent , but the advantage is it chose
the step size and descent direction more smarter than only use gradient
descent
• FISTA achieves a second-order convergence rate
• FISTA can be employed to minimize a cost function by convex problems.
• The objective function
• F sought-after object
• H is the system matrix
• b is the project data
• FISTA algorithm
• t is the step size
• x is intermediate image
Phantom and source setting

Phantom:

• Sphere, Radius = 15mm X_plane = 0, for ite =5

Sources:
• Sphere, Radius = 6 mm
• Activity = 8000 BQ
• Start time = 300 s

Reconstructed Image using coincides data


Source Location of the all (PET Sensitivity 5.2 %)
annihilations (singles)
Summary and Future study
• The Image reconstruction using geometrically calculated system response matrix (SRM) is checked.

• The image is reconstruction with the maximum likelihood estimation (MLE).

• Depth of Interaction information (DOI) will be in cooperated in the algorithm.

About FISTA
• Reconstruct of whole phantom .

• Make two same activity sources but different size and one activity 1.3e^+07 source, compare them to see the
difference.

• Find the best iteration number to reconstructed image.

16
• To include the depth of interaction information (DOI) in algorithm.
Backup

17
Distribution of Crystal Ids

18
Detector Evaluation
• Detector Efficiency
• Detector Sensitivity
Sensitivity of PET: Total number of coincidences to total events.
(Prompt coincidences) include all kind of coincidences and Random
True coincidences)
True Coincidences: EventID of first single is equal to EventID of second
single and no Compton scattering inside the phantom.
Random Coincidences: Comes in coincidence window but EventID of
one single is not equal to the EventID of second single.

19

You might also like