IMAGE, RADON, AND FOURIER SPACE
F1D R f ( , t )(t )  F2 D  f ( x, y)(x  t cos , y  t sin  )
FILTERED BACK-PROJECTION FOR PLANAR PARALLEL PROJECTIONS
f ( x, y)  
2
0
R f ( , t )  g (t )( , t  x cos  y sin  ) d
P
convolution
back-projection
ramp filter
1 
j t t
g (t )  F t  

e
dt
t



2π
P
1
1D
FILTERED BACK-PROJECTION FOR FAN BEAM
PLANAR DETECTOR CASE
y
a
Fan beam projection on linear detector: pF(,a)
        arct an
P(x,y)
a
,
R
P’
a

t  R sin   R sin  arct an 
R


S

x
R
U
1) Weighing by cos and ramp filtering
 F
R

p (, a )  p (, a )

R2  a2

F

  g P (a )  p F (, a ) cos   g P (a )




2) Back-projection
R2
F
f FBPF (x, y)  
p
(, a (x, y, ))d
2
U(x, y, )
2
0
 x sin   y cos
a ( x, y, )  R
, U( x, y, )  R  x cos  y sin 
R  x cos  y sin 
CONE BEAM PROJECTIONS ON FLAT PANEL DETECTORS
Flat
Panel
Detector
RX
CONE
z, b
RX
SOURCE
y
a  R t an 
a
b  R 2  a 2 t an 
  arct an
a
R
  arct an


b
R2  a2

VIRTUAL
DETECTO
R
x
FELDKAMP, DAVIS, KRESS (FDK) ALGORITHM (1984)
Approximated Filtered Back-Projection for cone-beam and circular trajectory
Satisfactory approximation even with quite high copolar angles (e.g., 20°)
It reconstructs the volume crossed by rays at any source position on the circles; hence
a cilinder plus two cones.
z, b
detector
inclined
fan
z
S
Sorgente
Fan
a
FDK ALGORITHM
z, b
1) Weighing by coscos :
R
R 2  a 2  b2
p F (, a, b)  cos  cos p F (, a, b)
y
P(x,y)
a
P’
2) Row by row filtering with the ramp filter




p F (, a, b)  cos cosp F (, a, b)  g P (a)
S

virtual
planar
detector
3) Back-projection
f FDK ( x, y, z ) 

2
0
R2
U ( x, y,  )
2
p F (  , a( x, y,  ), b( x, y, z,  ))d
 x sin   y cos
R
, b( x, y, z, )  z
R  x cos  y sin 
R  x cos  y sin 
U( x, y, )  R  x cos  y sin 
a ( x, y, )  R
x
FDK PROPERTIES
FDK algorithm is an approximate extension to the 3D cone beam case on planar
detector of the 2D
1) Exact on the central plane, z=0, where it coincides with the Fan Beam solution
2) Exact for objects homogeneous along z, f(x,y,z) = f(x,y).
3) Integrals along z,  f(x,y,z)dz, is preserved
4) Integrals on moderately tilted lines preserved as well
Main artifact: blurring along z at high copolar angles  FDK artifact
Satisfactory reconstructions were demonstrated even at fairly high copolar angles
(40°-50°). Usually much lower copolar angles are exlored (10°) in the field of
view, with higher precision.
Theory of 3D reconstruction from projections
RADON TRANSFORM IN 3D
The full Radon transform implies integration of volume over planes which are
projected on a point located at the intercept of the normal line through the origin
integration plane
Eˆ ,t

: r  ˆ  t ,
z

x  ( x, y , z ) T
integration plane
0
versor normal to the integration plane
ˆ( , )  (sin cos , sin  sin , cos )T
Radon value
t ˆ
  
Rf (, , t)     f (x )(x  ˆ  t) dx

R f ( , , t )  p


2 D ( , )
(t )
x

x
y
CENTRAL SECTION THEOREM IN 3D – FULL RADON TRANSFORM
Full Radon transform, 2D projection of parallel planes on the orthogonal axis
R f ( , , t )  p
2 D ( , )
(t )
F3 D  f ( x, y, z )(0,0, t ) 
  

f ( x, y, z )e ( r r  s s tt ) dr ds dt |r 0 
  
s 0

 2tt
     f ( x, y, z )drds e
dt  F1D p2 D (t )(t )
    

    
Result:
The 1D Fourier transform of the projection axis t gives the 3D Fourier values on
the corresponding axis t
THE FULL RADON TRANSFORM IS A SUFFICIENT DATA SET FOR 3D
IMAGE RECONSTRUCTION
Rf(,,t)
z
f(x,y,z)
z
F3Df(x, y, z)
z
F1D
t
t


y
y
y

x

x
x
.
F3D
A complete set of data is defined if all integration planes through the object are
present. The subset of parallel planes defined by the normal direction (,) fill the
radial axis with the same direction in F3Df(x,y,z). So we need 2 directions times 
parallel shifts; i.e., 3 planar integrations. These can be provided by 2 planar
projections.
CENTRAL SECTION THEOREM IN 3D – PARTIAL RADON TRANSFORM
Partial Radon transform, p1D, r (s,t) = projection of lines parallel to a ray axis r on the
orthogonal projection plane at coordinates (s,t)
The 2D Fourier transform of projected values on the projection plane (s,t) gives the
3D Fourier values on the corresponding plane frequency domain plane (s , t)
F3 D  f ( x, y, z )(0, s , t ) 
  

f ( x, y, z )e  2 ( rr  ss tt ) dr ds dt |r 0 
   

  2 ( ss tt )
f
(
x
,
y
,
z
)
dr
ds dt  F2 D p1D ,r ( s, t )(s , t )
 e


    

Deriving Full Radon (integrals on planes) Transform from Partial Radon
Transform (line integrals)
projection plane
integration plane
E rˆ,0 (s, t )
z
z
E ˆt ,t (r, s)
0
s
s
r
F3D
r
t
t

t0
y
x
p
1D , r

y
x
( s, t  t0 )ds  p2 D ,( , ) (t  t0 )
t t 0
Points on a projection plane represent 1D line integrals parallel to r. A further integration on a
line t=t0 correspond to a 2D integration on a plane orthogonal to t and a distance t0 from the
origin. Hence the Full Radon Transform sample in position
(r=0, s=0, t=t0) .
Note that 1D projections are much more informative than 2D projections, due to lesser
integration: a 1D projection on a plane fills an entire plane in the 3D Radon and Fourier spaces
REDUNDANCY OF PROJECTIONS ON ALL PLANES
1. Projections planes are 2 (azimuth and polar angle).
In each plane 2 integration lines can be defined.
Hence 4 values are found; i.e., each Radon Transform point is found in  ways.
Indeed, the same can be computed on the family of  projection planes containing
axis t and fill the corresponding axis in the Full Radon Transform and in the 3D
Fourier space.
2. Equivalently, the Central Section Th. (in the Partial Radom Transform version),
says that by transforming a planar projection an entire plane (2 points) is filled in
the 3D Fourier space.
Given the 2 planar projections we obtain a redundancy of 3 over 4 , again.
In conclusion, a set of 1 directions filling the Radon Space and the Fourier space is
sufficient for the reconstruction.
Passing from parallel projections to cone beam projections an appropriate
trajectory of the focal spot passing through 1 points can cover the entire 3D
transform spaces if it satisfies proper conditions (see Tuy Smith sufficient
condition in the next slide).
SUFFICIENT SET OF DATA IN 3D FOR CONE BEAM
Tuy-Smith sufficient condition (1985): A cone-beam projection permits to derive the
integral of each plane passing through the source S. Hence, if the source in its trajectory
encounters each plane through the object a sufficient set is obtained
Explanation: Any plane (,,t) through S is filled by X-rays and we know the integral (1D
projection) along them; the 2D projection of the plane (i.e., integral over the plane) can be
derived by summing up the 1D projection values giving a 3D Radon sample Rf(,,t).
Note that 2 planes pass through a point S, which multiplied by  positions of S on the
trajectory can fill all the 3D Radon space, if a proper trajectory is chosen.
trajectory of S
z
t
S
plane (,,t) touched by S

y

x
corresponding point of
Radon Transform Rf(,,t)
ALMOST SUFFICIENT SET OF DATA IN 3D FOR CONE BEAM
A cone-beam projection permits to derive the integral of each plane passing through
the source. Hence, if the source in its trajectory encounters each plane through the
object a sufficient set is obtained. This is the Tuy-Smith sufficient condition (1985).
A circular trajectory, most often used, satisfies this condition only partially: planes
parallel to the trajectory are never encountered. Hence, a torus is filled in Radon space
with a hole, called shadow zone, close to the rotation axis z.
z
z
shadow zone
x
y
x
x
TRAJECTORIES SATISFYING TUY-SMITH CONDITION
1. helics (used in multislice “spiral” CT)
2. two non parallel circles (possibly used in C arm cone beam)
3. circle and line (just theoretical)