SPECT Reconstruction (FBP, OSEM,

filters)

Dr Arun Prashanth K
 Previous Exam Questions
• Describe different types of Filters in SPECT
reconstruction and elaborate Butterworth and
Shepp-Logan filter applications in different
clinical settings.
• Discuss the different reconstruction
techniques in nuclear medicine imaging.
• PSF
• Nyquist frequency
 SYNOPSIS
• Basic concepts
• Back projection
• Filters
• Iterative methods
Basic concepts
Representation of nuclear medicine imaging in form of frequency
 PSF
• Describes the blurring of a point source of
radioactivity by the nuclear medicine imaging
system.
• Ideal vs. Real Imaging: An ideal system would
image a point as a point. Real systems produce
a blurred distribution of counts
• simply response of an imaging system to
a point source or point object
Narrower PSF = better resolution.
 Factors affecting PSF
• Collimator: Hole design, length, and source distance.
• Detector Intrinsic Resolution: Detector element size,
crystal properties.
• Scattered Radiation: Compton scattering within the
patient and detector.
• Photon Range & Acollinearity (PET): Physical limits in
PET imaging.
• Reconstruction Algorithms: Processing can influence
the apparent PSF.
• Patient Motion: Blurs the image, effectively widening
the PSF.
Factors affecting PSF
 Why PSF is important?
Resolution Recovery
• Quantifying Resolution: Allows objective
measurement (e.g., FWHM).
• Image Processing: Enables deconvolution to
improve sharpness.
• Accurate Interpretation: Helps understand
limitations when assessing small structures.
• Optimizing Parameters: Guides selection of
collimators and settings
 Clinical Benefits of PSF Correction
In PET Imaging:

• Improved spatial resolution (up to 20-30% sharper images).

• Better small lesion detection (e.g., small tumors, brain PET).

• Reduced Partial Volume Errors (more accurate SUV

measurements).

• Enhanced contrast (especially in oncologic PET/CT).

 Clinical Benefits of PSF Correction
In SPECT Imaging:

• Sharper images despite collimator limitations.

• Better defect delineation (e.g., myocardial perfusion

SPECT).

• Quantitative SPECT becomes more reliable (e.g., in ¹⁷⁷Lu-

DOTATATE therapy dosimetry).
 Challenges & Limitations
• Noise Amplification – PSF correction can
increase noise; thus, regularization is needed.
• Computational Cost – More complex than
standard reconstruction.
• System-Specific PSF – Requires calibration for
each scanner.
• Not a Magic Fix – Cannot recover information
lost due to physical limits (e.g., positron range
in PET).
SPECT RECONSTRUCTION
Size of a pixel
=1/dimensions
of a matrix

given field of
view.
• Noise usually present in all frequency ranges
• Higher frequency = better resolution/edge
detection
• So if resolution then noise
• Also increase in resolution requires more
sensitivity which in turn increases time of
examination.
• High frequency noises produce loss of
resolution than low frequency.
 Back Projection
• Fastest and the simplest method of
reconstruction-simple back projection
• Radon Transformation
Various positions of scanner and image
representation in respective positions
How back projection is done?
Making of sinogram
Various number of Back projections
Removal of star artefact by decreasing
counts on either sides
1/r effect and its removal
Simple back projection Filtered back projection
FILTERS
Common characteristics of low and
high pass filters
 Cut off frequency
• Power of the filter
• Maximum frequency which it allows to pass=?
• If more than nyquist frequency then abrupt
termination results.
Butterworth filter
Butterworth filter
Shepp Logan

•Linear Increase at Low Frequencies

•Roll-off at High Frequencies
•Often Includes a Cutoff Frequency

•Least Smoothing
•Highest resolution
Parameters Determining the Choice of
the SPECT Filter
• energy of isotope, number of counts, activity
administered
• statistical noise and the background noise
level.
• type of organ being imaged.
• kind of information we want to obtain from
the images.
• collimator used.
 Filters Type

Ram-Lak star artefact removal; noise sensitivity

Butterworth noise reduction

Metz noise reduction; contrast enhancement

Wiener noise reduction; contrast enhancement

Scramp noise reduction; contrast enhancement

Inverse MTF noise reduction; contrast suppression

Hamming noise reduction

Parzen noise reduction

Shepp-Logan noise reduction

Hann noise reduction

 Min and Max
• Ramp - best resolution & most noise
• Butterworth – smooth image least resolution
• Shepp Logan – Best resolution least smooth
• Quantifying best contrast and SNR – Gaussian
• Best defect size - Parzen
• Optimal balance - Butterworth
 Organ Filter

Myocardium Butterworth (at 0.45nq and 11 order)

Brain SPECT Butterworth (0.5 nq and 10 order)

Bone Metz, Butterworth

Renal Hamming and Metz

Lung Hann and OSEM

Thyroid Butterworth
Sequence of applying filters

Pre filtration by low pass filters

then
Ramp/high pass filter
 ITERATIVE RECONSTRUCTION
• Guess estimates
• projection views
• Difference between ratio of projection views
• New estimates
• Repeated/iterated again
 Iteration and linear regression
• 1 variable requires 1 equation
• 2 variable requires 2 equations and so on to
solve it.
• What if only 1 equation available for “n”
number of variables.
Iterative method
 MLEM
• The expectation-maximization (EM) algorithm
incorporates statistical considerations to
compute the “most likely,” or maximum-
likelihood (ML), source distribution that would
have created the observed projection data.
• greater weight to high-count elements of a
profile and less weight to low-count regions.
 OSEM-MLEM
• Ordered subsets of images are used instead of
separate images.
• Likelihood always increases with iteration
• non-negativity of solution guaranteed
• Use
– Drastic time reduction
– Better computational efficiency
 Difference between regular
iteration, OSEM-MLEM at 8 and 16
subsets
 Advantages of OSEM/MLEM
• Attenuation correction
• Compensation for distance-dependent
resolution
• Scatter correction
• Motion correction
• Incorporation of anatomical information
 Disadvantages of OSEM/MLEM
• slow: requires multiple iterations
• noise control: unsure when to stop
• At higher iterations image is more noisy.
 FBP vs. Iterative methods
• Assumes very simple • more exact projection
projection model model
• Non-uniform • includes variable
attenuation not attenuation
included(attenuation • flexible detector
correction to be done geometry
separately) • reduces streak artefacts
• Streak artefacts • handles missing data
• Noise amplification • improved noise
characteristics
 Feature Filtered Back-Projection (FBP) Ordered Subsets Expectation Maximization (OSEM)

Significantly slower per iteration, but can converge faster

Computational Speed Generally fast
overall in some cases.

Sensitive to noise in the projection data. Noise can be

Noise Handling More robust to noise, especially at lower count statistics.
amplified.

Can produce images with sharp edges but may have star Tends to produce images with better contrast, reduced
Image Quality
artifacts and increased noise. noise, and fewer artifacts, especially with limited data.

Can incorporate attenuation and scatter correction

Requires explicit, often separate, correction before
Handling of Attenuation & Scatter within the iterative process (more complex
reconstruction.
implementations).

Difficult to directly incorporate PSF modeling during Can incorporate models of the PSF to improve resolution
Handling of System Response (PSF)
reconstruction. and reduce the partial volume effect.
 Feature Filtered Back-Projection (FBP) Ordered Subsets Expectation Maximization (OSEM)

Iterative process that converges towards a solution.

Convergence Single pass algorithm; no iteration needed. Convergence speed depends on the number of subsets
and iterations.

Can be higher, especially for 3D reconstructions and

Memory Requirements Generally lower.
storing intermediate images.

Generally performs better, yielding more diagnostic

Suitability for Low Count Data Can produce noisy images.
images.

Generally has better potential for accurate quantification,

Can be less accurate, especially with uncorrected
Quantification Accuracy especially with appropriate corrections and sufficient
attenuation and scatter.
iterations.

Can exhibit streak artifacts if too few iterations are used;

Common Artifacts Star artifacts, ringing artifacts, noise amplification.
may have a "blocky" appearance with too many subsets.

More flexible in incorporating complex physical effects

Flexibility Less flexible in incorporating complex physical effects.
like attenuation, scatter, and detector response.

More complex to implement due to the iterative nature

Implementation Complexity Relatively simpler to implement.
and potential for incorporating various corrections.
QUESTIONS