CT Reconstruction
What is CT Image
Reconstruction?
• Mathematical process that generates images for x-
ray projection data acquired @ many different
angles around the patient
• Two major categories of methods exist:
• i) Analytical reconstruction: filtered back projection
methods
• ii) Iterative reconstuction
• Clinical CT scanners have very limited control over
the inner workings of the reconstruction method &
are confined principally to adjusting various
parameters specific to different clinical applications
• The reconstruction kernel, or ‘filter’ or ‘algorithm’ is
one of the most important parameters that affects
image quality
• Smooth kernel generates images with less noise but
less spatial resolution. Sharp kernel generates
images with increased spatial resolution but
increased noise
• Selection of kernel based on clinical applications. EG:
Smooth kernels used in brain exams or liver tumor
assessment to decrease noise and enhance low
contrast detectability —> radiation dose increases
• Sharper kernels used in exams to assess bony
structures because of increase spatial resolution —
>lower radiation dose due to increased contrast
• Slice thickness is another important parameter that
controls image quality in longitudinal direction. This
influences trade offs among resolution, noise and
radiation dose
• CT users have the added responsibility of choosing the
appropriate reconstruction kernel and slice thickness for
each application to decrease radiation dose and ensure
image quality isn't compromised
• Iterative reconstruction has advantages that includes
important physical factors including focal spot & detector
geometry, photon stats, x-ray beam spectrum and
scattering that can be more accurately incorporated into
iterative reconstruction—>yielding lower image noise &
higher spatial resolution when compared to FBP
• Iterative reconstruction can also reduce image artifacts
such as beam hardening, windmill and metal artifacts
Filtered Backprojection
• In simple backprojection, we sum projections from a large number of angles around 360
degrees
• This results in the 1/r blur function
! Blurring caused by the
geometry of the back
! projection is corrected by
deconvolving the measured
! projection data prior to back
projection —>FBP!
!
!
!
• A mathematical operation of convolution corrects for the impact of the blurring function
• Deconvolution is used when we want to ‘undo’ an
effect caused by convolution
• Convolution backprojection is considered a specific
implementation of filtered back projection
Fourier Based
Reconstruction
! Measured projection passes
through deconvolution kernel
! to give filtered projection data
which is used for back
! projection resulting in an
image that reflects the
! properties of the original
object
!
!
• In filtered backprojection, we start off using the ramp filter!
• The ramp filter is used to compensate for the sparser sampling at higher densities. 1/r
blurring effect corresponds to a 1/f effect in the frequency domain. If an image has a 1/f
dependency, the correction process would involve multiplying the image by a function that
has a f dependency. Therefore 1/f * f eliminates any frequency dependency
CT Dosimetry
• Radiation dose distribution is more homogenous in
CT or radiography/fluoro? Why?
• Dose distribution in CT is more homogenous due to rotational
irradiation geometry used
• CT dose gradients are very slight and distribution depends on the
diameter and shape of the point and on the beam quality (kV)
• CT dose index (CTDI) is an index for patient dose assessment
• CTDI100 involves the use of a 100mm long cylindrical chamber,
approx. 9mm in diameter, inserted into either the centre or
peripheral hole of a PMMA phantom (Acrylic material)
• 100mm chamber length is useful for x-ray beams for thin slices
(5mm) to thicker beam collimations such as 40mm
• Equation describes measurement of dose
distribution, D(z) along the z-axis, from a single
circular rotation of the scanner with a nominal x-ray
beam width of nT
• CTDI100 measurements are made for both the
centre and periphery. They are both then combined
using a 1/3 to 2/3 weighting scheme
• CTDIvol = CTDIw/pitch ; where CTDIw is the
combined CTDI100 from centre and periphery
• Dose length product, DLP = CTDIvol * L ; where L is length of CT scan
along z-axis of patient
• Limitations of CTDIvol:
• CTDIvol is a dose index and not a measurement of dose!
• CTDIvol results from air kerma measurements at two locations to a very
large cylinder of PMMA plastic. When related to human dimensions, the
phantom corresponds to a person with a 47” waistline —> large patient!
Therefore for small pts, doses are larger than CTDIvol for same technique
factors
• Researchers have included a patient size conversion factor to deal with
this
• CTDIvol is calculated as the dose in air at the centre of a 100 mm long
phantom. What is neglected when compared to a real life situation?
• Scatter dose distribution! This is an important component of
radiation dose to the patient in CT
• Dose from scatter most intense along z-axis close to the primary
beam and decreases as the distance along z from primary CT
beam increases
! Higher tube voltages
generate scatter tails
with greater range
!
!
Image Quality in CT
• Spatial Resolution:
• Depends on fundamental resolution properties of image acquisition and
resolution characteristics of reconstruction filter used
• The ultimate resolution is determined by: focal spot size & distribution,
detector dimensions, magnification factor, whether or not gantry motion is
compensated for, patient motion etc.
• X-Ray tube focal spot distribution:
• i) focal spot reduces SR in CT
• ii)Object is highly magnified relative to projection radiography
• iii) CT runs @ very high mA & this can increase in size of x-ray focus
• Gantry motion:
• i) x-ray source and detector moving relative to stationary pt both in
angular dimension and along z-dimension for helical acquisition
• ii)This reduces SR and methods such as focal spot rastering can be
used to compensate for this
• Detector size and sampling:
• i) Smaller detector dimensions and oversampling methods can increase
SR
• Reconstruction filter:
• i) SR is, more often than not, intentionally reduced by selection of
reconstruction filter with significant roll-off at high spatial frequencies.
This reduces the appearance of image noise
• We can have reconstruction taking place on
multiple occasions to increase SR and decrease
noise, with no added dose to patient
Factors that affect contrast
resolution (noise in CT)
• Technique factors:
• i)kV, mA, time & pitch affects dose levels
• ii)mAs have linear rel. with noise, kV dose not,
pitch?
• Slice thickness:
• i) Thicker slices —> more or less noise? why?
• Less noise with thicker slice due to more detected x-rays
• Reconstruction filter (FBP):
• i) Choice of filter results in trade-off between SR and image
noise
• Reconstruction methods:
• i) Iterative reconstruction reduces image noise vs. FBP
• ii) Lower dose image using IR = higher dose studies with
FBP
