central slice theorem · npre 435, principles of imaging with ionizing radiation, fall 2018 signals...

y

xr x’

IncidentX-rays

f(x,y)

Detected p(, x’)

R

Central Slice Theorem

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

The thick line is described byx cos + y sin = R

Properties of Fourier Transform

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Fourier Transform

Spatial Domain

Spatial Frequency Domain

Linear shifting in spatial domain simply adds some linear phase to

the pulse Magnitude is unchanged

ajexfaxf 2)]([)]([ FF

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Signals and Systems

Line Impulse Signal (1)

otherwiselyx

xwhere

lyxyxL

,0sincos,0

)(

)sincos(),(

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Signals and Systems

Line Impulse Signal (2)

×

x’

2-D integral

The value of the projection function p(x’)at this point is the integral of the functionof f(x,y) along the straight line:x’=xcos+ysin

dxdyxyxyxfxp )sincos(),(),(

The integral of a line impulse function and a given 2-D signal gives theprojection data from a given view …

)xsinycosx(

Central Section Or Projection Slice Theorem

F{p ( , x’)} = F(r,)

So in words, the Fourier transform of a projection at angle gives us a line in the polar Fourier space at the same angle .

Central slice theorem is the key to understand reconstructions from projection data

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

∑ ∑ ∑ ∑ ∑

CT projection measure line integralsx

y

f(x,y)

p(x) = CT detector array output

p(x)

y

y

dyyxfxp ),()(

X-ray Projection Revisited

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of 2-D Analytical Reconstruction MethodsProjection Data

Projection data p(, x’)

y

xr x’

IncidentX-rays

f(x,y)

Detected p(, x’)

R

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

dxdyxyxyxf

xp

)sincos(),(

),(

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Signals and Systems

Radon Transform and Sinogram

http://tech.snmjournals.org/cgi/content-nw/full/29/1/4/F3

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Signals and Systems

Radon Transform and Sinogram

http://tech.snmjournals.org/cgi/content-nw/full/29/1/4/F3

Simple Backprojection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of 2-D Analytical Reconstruction MethodsBack Projection Operation

y

xr x’

IncidentX-rays

f(x,y)

R

Detected p(, x’)

Back projection from a single view angle:b (x,y) = ∫ p (x’) (x cos + y sin - x’) dx’

x’

Simple Backprojection

Adding up all the back projections from all the angles gives, fback-projected (x,y) = ∫ b (x,y) d

π ∞fb (x,y) = ∫=0 d ∫x’ = -∞ p (x’) (x cos + y sin - x’) dx’

Simple Backprojection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Crude Idea 1: Take each projection and smear it back along the lines of integration it was calculated over.

Result from a back projection from a single view angle:

b (x,y) = ∫ p (x’) (x cos + y sin - x’) dx’x’

Adding up all the back projections from all the angles gives,

fback-projected (x,y) = ∫ b (x,y) d

π ∞fb (x,y) = ∫ d ∫ p (x’) (x cos + y sin - x’) dx’

0 -∞

Simple Backprojection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

From Medical Physics and Biomedical Engineering, Brown, IoP Publishing

Simple Back-projection and the 1/r Blurring

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

hb(r) = 1/r

fb (x,y) = f(x,y) * 1/r

Fb (,) = F (, ) / since F{1/r}= 1/

Impulse Response Function of Simple Backprojection Operator

Back projected image is blurred by convolution with 1/r

The nature of the 1/r blurring:Radon transform produced equally spaced radial sampling in Fourier domain.

Simple Back-projection and the 1/r Blurring

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Crude Idea 1: Take each projection and smear it back along the lines of integration it was calculated over.

Result from a back projection from a single view angle:

b (x,y) = ∫ p (x’) (x cos + y sin - x’) dx’x’

Adding up all the back projections from all the angles gives,

fback-projected (x,y) = ∫ b (x,y) d

π ∞fb (x,y) = ∫ d ∫ p (x’) (x cos + y sin - x’) dx’

0 -∞

Simple Backprojection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

The Nature of the 1/r Blurring

The nature of the 1/r blurring:

Radon transform produced equally spaced radial sampling in Fourierdomain.The low frequency components are over sampled, which causes

hb(r) = 1/r and fb (x,y) = f(x,y) * 1/r in spatial domain

and

Fb (,) = F (, ) / in spatial frequency domain

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Inverse Radon Transform

Suppose the sample projection data preserves all informationcontained in the original function f(x,y), can we recover the exactfunction f(x,y) with an Inverse Radon Transform?

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Inverse Radon TransformThe estimate of the original image f(x,y) can be obtained as

where

andThe Central Slice Theorem

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Simple and Filtered Back-projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Mathematically, we can define a filtered back-projection FBP operation to remove the 1/r blurring.

Filtered Back-projection

where

Can this be realized ??

Due to the diverging nature of the |w| function, the corresponding filter kernel does not exist in spatial domain!

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Inverse Radon TransformThe estimate of the original image f(x,y) can be obtained as

where

The inverse Radon transform can be represented as a filtering processfollowed by a back-projection operation.

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Simple and Filtered Back-projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Simple and Filtered Back-projection

From Computed Tomography, Kalender, 2000.

Filtered Backprojection (FBP), What and why?

Central Slice Theorem

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

http://engineering.dartmouth.edu/courses/engs167/12%20Image%20reconstruction.pdf

One would need to use filter to compensate for this effect.

Ideally, one would use a perfect filter as in the Inverse Radon Transform

The Ram-Lak filter

Filtered Back-projection

Ram-Lak filter

0

Ideal filter

Consider the bandwidth-limited nature of most projection data, we have …

Filtered Back-projection

The Ram-Lak filter in spatial domain

Simple and Filtered Back-projection

From Computed Tomography, Kalender, 2000.

Simple and Filtered Back-projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Signals and Systems

Radon Transform and Sinogram

http://tech.snmjournals.org/cgi/content-nw/full/29/1/4/F3

Filtered Back-projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

https://www.youtube.com/watch?v=ddZeLNh9aac

But is there something missing in this discussion?, such as

Possible artifacts?

Noise in data?

Filtered Back-projection –The Optimum Filter

Have we forgot something? What about the noise in the projection data?

In reality, the true projection is

Where is the noise coming from?

Review of Fourier Transform and Filtering Spectral Filtering

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of Fourier Transform and Filtering Spectral Filtering

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection

The “ringing” artifacts in reconstructed images.

In this context it is usually manifest itself as "streak artifacts". You may, for example, seethis as lines radiating from the center and outwards. The term comes from electronics andi s m e a n t i n t h e s e n s e o f a b e l l - r i n g i n g .

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection

The Ram-Lak filter in spatial domain

Filtered Back-projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-Projection

Sampled version

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection

The sharp boundary of the Ram-Lak filter often make the spatial domainfilter kernel oscillatory.It sometime introduces the “ringing” artifacts in reconstructed images.This can be effectively resolved by the Shepp and Logan filter

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

- Low spatial frequency data is overweighted. Filter to compensate for this. Weighted by 1/.

- Solution - filter each projection by || to account for the uneven sampling density

Steps:1) Projection operation2) Transform projection3) Weight with || 4) Inverse transform5) Back project6) Add all angles

Filtered Back-projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Any chance we can define an OPTIMUM filter function for FBP?

Filtered Back-projection

For FBP with noisy projection data, one can derive an optimum filterfunction. The FBP reconstruction has the minimum mean square error

In other words, there exists a filter function that can be used in the FBPreconstruction that produces the most faithful reproduction of theoriginal image f(x,y)

2)),(ˆ),((.. yxfyxfEESM

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projectionFor a situation in which we know the power spectra for both the imagefunction f(x,y) and the noise, the optimum filter function is

Where

Hw(,) is the Wiener filter function.

Remember that

Image degradation function

Signal power spectrum of theprojection data at view angle

Power spectrum of the noisein the projection data at viewangle

222)()()(),( xpimagxprealxpWp FFF

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Fourier Transform

),(),(),( 22 vuFvuFvuF IR

),(),(tan),( 1

vuFvuFvuF

R

I

),(),(),( vuFjevuFvuF

• In general, Fourier transform is a complex valued signal, even iff(x,y) is real valued.

• It is sometimes useful to consider the magnitude and phase ofthe Fourier transform separately.

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018 Fourier Transform

• The square of the magnitude |F(u,v)|2 is referred to as thepower spectrum of the original function.

),(),(),( vuFjvuFvuF IR Fourier coefficients are complex:

Magnitude:

Phase:

An alternative representation:

Filtered Back-projection with Optimum FilterThe optimum filter function can also be written as

And then

where is the signal-to-noise ratio (SNR) ofthe projection data at a given view angle .

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection

Therefore the optimum estimator (the optimum reconstruction) of theoriginal image function f(x,y) is

where the filtered projection data is given by

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Most of current X-ray CT work in fan-beam mode …

Filtered Back-projection in Fan-beam Mode

Why fan beam mode?Most modern X-ray CT system use fan (or cone) beam data acquisitionscheme.Image reconstruction with fan beam mode often provide better spatialresolution with the same dimension of sampled data as the parallel case,due to the improved sampling at the central region. This is found to beimportant for PET, where the intrinsic limitation on spatial resolution ison the finite detector size.

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection in Fan-beam Mode

Comparing parallel beam and fan beam geometries

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection in Fan-beam Modewhere ’ is the angle between central line and the line passing through thereconstructed point at (x,y). And v is the distance between the apex of fan andthe reconstruction point p.

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection in Fan-beam Mode

The basic idea:

The mathematical treatment for fan beam mode is almost identical tothat for parallel beam case with changing parameters!

Starting from the one-to-one relationship between the parallel beamprojection space (, x’) and fan beam projection space (,),

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection in Fan-beam Mode

Where the Jacobian |J| is

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection in Fan-beam Mode

Where the Jacobian |J| is

and the FBP in fan beam mode becomes

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filter Function in Fan-beam Mode

Similarly the filter function used in fan beam mode is a directtransformation from the parallel beam counterpart

Filtered Back-projection in Fan-beam ModeFBP in parallel beam case

can be converted to FBP in fan beam mode by changing integrationvariables

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Filtered Back-projection in Fan-beam Mode

Where the Jacibian |J| is

and the FBP in fan beam mode becomes

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Back-projection and Filtering Method for Reconstruction

Simple Back-projection and the 1/r Blurring

From Medical Physics and Biomedical Engineering, Brown, IoP Publishing

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

From Medical Physics and Biomedical Engineering, Brown, IoP Publishing

Simple Back-projection and the 1/r BlurringRevisited

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

hb(r) = 1/r

fb (x,y) = f(x,y) * 1/r

Fb (,) = F (, ) / since F{1/r}= 1/

Impulse Response Function of Simple Backprojection Operator Revisited

Back projected image is blurred by convolution with 1/r

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Reconstruction with Back-projection and Filtering

If we know that the consequence of simple back-proejction (undercertain assumptions) is to apply a 1/r blurring on the input image …

Can we envisage an alternative reconstruction method that

Back projection first

and thenDe-convolve the 1/r blurring?

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Reconstruction with Back-projection and Filtering

The scheme leads to the back-projection and filtering reconstructionmethod

where ρ is the radial spatial frequency and

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Crude Idea 1: Take each projection and smear it back along the lines of integration it was calculated over.

Result from a back projection from a single view angle:

b (x,y) = ∫ p(, x’) (x cos + y sin - x’) dx’

Adding up all the back projections from all the angles gives,

fback-projected (x,y) = ∫ b (x,y) d

π ∞B (x,y) = fb (x,y) = ∫ d ∫ p(, x’) (x cos + y sin - x’) dx’

0 -∞

Simple Backprojection --BB

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Reconstruction with Back-projection and Filtering

BPF reconstruction provides relatively poor images compared with FBPresults because of the following two issues:

The back projection step results in an image with infinite extend. Cuttingit to NxN points for filtering leads to loss of information.

The 2-D filter function discussed above has a slope discontinuity andtherefore leads to ringing artifact.

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Reconstruction with Back-projection and Filtering

If high quality reconstructions were to be achieved the BPF method, thefollowing aspects have to be considered

Although the final reconstruction is on NxN points, the back projectionstep should use an matrix that is at least 2Nx2N.

We would need to apply appropriate windowing to the filter functiondiscussed above, just like in the parallel beam case …

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Reconstruction with Back-projection and Filtering

The procedures of BPF algorithm is VERY SIMPLE!

From Page 88, Foundation of Medical Imaging, Z. H. Cho, John Wiley & Sons, 1993.

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of 2-D Analytical Reconstruction MethodsKey concepts:Mathematical framework for modeling the projection process•Radon transform•Central slice theorem

Characteristics of typical projection data•Non-uniform sampling of the 2-D Fourier transform space•Actual measurements are associated with statistical noise and detector imperfections.

Analytical reconstruction methods•Inverse Radon transform•Filtered back-projection (FBP)•FBP in fan-beam geometry•Back-projection and filtering (BPF)

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of 2-D Analytical Reconstruction MethodsProjection Data

Projection data p(, x’)

y

xr x’

IncidentX-rays

f(x,y)

Detected p(, x’)

R

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Projection data is a 2-D function representing the original function f(x,y)transformed into the projection data space.

Given (a) a sufficiently sampling rate and (b) a band-limited image and (c) aperfect measurement, the projection data contains all information requiredto recover the original image.

Projection operation either through actual measurements or Radontransform, map 2-D function into the sinogram space.

Projection data are obtained as line-integrals from view different angles andradial distances from the center.

Review of 2-D Analytical Reconstruction MethodsProjection Data

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of 2-D Analytical Reconstruction MethodsBack Projection Operation

y

xr x’

IncidentX-rays

f(x,y)

R

Detected p(, x’)

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Back projection is NOT an exact inverse operation of the projection due tothe lack of the information regarding the actual distribution of the originalfunction along the LOR.

Back projection does NOT provide an exact reproduction of the originalfunction, f(x,y). Instead, it provide an image that is the convolution of theoriginal image with a 1/r blurring.

Back projection assign a constant value to each element along theprojection lines (or lines of response, LOR). The value assigned isproportional to the projection function evaluated at the correspondinglocation on the x’ axis.

Review of 2-D Analytical Reconstruction MethodsProjection Data

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

This 1/r blurring reveals an important aspect of the projection operation:The original function was sampled non-uniformly sampling rate acrossboth (x,y) and (u,v) planes.

Review of 2-D Analytical Reconstruction MethodsBack Projection Operation

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

The basic idea of analytical reconstruction is to recover the original image(a) using back projection operation and (b) applying filters in either spatialfrequency domain or spatial domain to compensate for the non-uniformsampling.

Review of 2-D Analytical Reconstruction MethodsFiltered Back Projection (FBP)

The approach of correcting for the non-uniform sampling in theprojection data space and then performing back projection leads to theFiltered Back Projection (FBP) Method

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Review of 2-D Analytical Reconstruction MethodsBack Projection and Filtering (BPF)

The approach performing back projection first and followed by thecorrection for the non-uniform sampling in the (x,y) space is called BackProjection and Filtered (BPF) Method

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

As a matter of fact, FBP provides almost optimum image quality with goodquality projection data, but performs poorly when the projection datacontains significant noise.

Why?

Are we missing something in the design of the FBP or BPF methods?

Review of 2-D Analytical Reconstruction MethodsFiltered Back Projection

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018

Basic Problem of Image Reconstruction

NPRE 435, Principles of Imaging with Ionizing Radiation, Fall 2018