Lecture 5 Flashcards
What is tomographic imaging?
Data collected as a series of 2D projections at different angles around an object or directly as sinograms in PET. Method for obtaining 3D activity distribution within an object.
How is tomographic imaging carried out in SPECT?
Camera rotation- continuous or step and shoot
180 or 360 degrees rotation- Steps usually 3-6 degrees
Requirement for collimator- no intrinsic line-of-response like PET or CT
What are the two main times of reconstruction for SPECT?
Filtered back projection
Iterative
Describe filtered back projection
Assume that each line through the collimator is a cylinder and the point on the detector only sees things from a cylinder in front of it. The projection profile is a series of ray sums which are integrals through the object. 1D projection profiles are generated at all planar projection angles
Describe a sinogram
profiles from the same rows from each projection angle are stacked to form an image. There are as many sinograms as there are rows in the projection data
What are star artefacts and how can they be reduced
Start artefacts are inherent because counts are ‘smeared’ along each projection. This effect can be reduced by increasing the number of projections. around 120 projections are used in practice.
What is a fourier transform used for in FBP?
The effect of blurring is reduced by filtering projection images in frequency space- it is easier than filtering in spatial domain. The Fourier transform converts the image from object space to frequency space.
Why is a ramp filter applied and how is it applied?
Standard back projection causes 1/r blurring and a loss of high frequency of fine detail. The ramp filter is only applied in the transaxial plane as this is the only plane to experience 1/r blurring.
Blurring is minimised by projecting negative counts around each object.
What is iterative reconstruction?
A statistical trial and error method- commonly used in PET. It produces an image via successive approximations
Describe the four steps of iterative reconstruction
- Obtain an initial guess image
- The image is forward projected
- The estimated projections are compared with the real projection data and then back projected and adjusted accordingly
- The process is repeated over a number of iterations
What is maximum likelihood expectation maximisation (MLEM)
The most popular type iterative algorithm. Includes statistical considerations to calculate the maximum likelihood source distribution i.e. doesn’t attempt to find a single correct solution but MOST LIKELY solution.
Describe expectation maximisation
Estimate the distribution of the tracer that is most likely to give the acquired data. Ordered subset EM (OSEM) is a clever computational approach to EM. It ahs been shown to give better signal to noise ratio with low count data. Attenuation and scatter are more naturally accommodated into the reconstruction.
Why is iterative reconstruction slow?
- Need several iterations to converge- each iteration requires both a forward and back projection
- Include models of imaging system i.e. collimator scatter, resolution recovery- includes consideration of most pixels
- Difficult to optimise number of subsets and iterations
- Different regions will converge at different speeds
- Needs to be optimised for each clinical situation
Give two advantages and two disadvantages of FBP
Pros:
1. Computationally quick
2. Simple to perform and understand
Cons:
1. Artefacts are produced as a result of back scatter
2. difficult to incorporate scatter, attenuation and detector blur
Give two advantages and disadvantages of OSEM/MLEM ( iterative reconstruction)
Pros:
1. physical properties e.g. attenuation are modelled at projection level
2. reconstruction noise is lower than FBP
Cons:
- Not simple to define when image quality is final
- Takes longer than FBP