You are on page 1of 8

GELFAND-GRAEVS RECONSTRUCTION FORMULA IN THE 3D REAL SPACE

Yangbo Ye
1,2
, Hengyong Yu
3,4
, Ge Wang
4,5


1
Department of Mathematics, University of Iowa, Iowa City, Iowa 52242, USA
2
School of Mathematics, Shandong University, Jinan, Shandong 250100, China
3
Department of Radiology, Division of Radiologic Sciences, Wake Forest University Health
Sciences, Winston-Salem, NC, 27157, USA
4
Biomedical Imaging Division, VT-WFU School of Biomedical Engineering and Sciences, Wake
Forest University Health Sciences, Winston-Salem, NC, 27157, USA
5
Biomedical Imaging Division, VT-WFU School of Biomedical Engineering and Sciences,
Virginia Tech., Blacksburg, VA 24061, USA

yey@math.uiowa.edu, hengyong-yu@ieee.org, ge-wang@ieee.org
ABSTRACT

Purpose: Gelfand and Graev performed classic work on the inversion of integral transforms in different spaces (I. M.
Gel'fand, and M. I. Graev, Crofton function and inversion formulas in real integral geometry. Functional Analysis and its
Applications, 1991. 25(1): p. 1-5). This paper discusses their key results for further research and development.
Methods: The Gelfand-Graev inversion formula reveals a fundamental relationship between projection data and the Hilbert
transform of an image to be reconstructed. This differential backprojection (DBP)/backprojection filtration (BPF) approach
was re-discovered in the CT field, and applied in important applications such as reconstruction from truncated projections,
interior tomography, and limited-angle tomography. Here we present the Gelfand-Graev inversion formula in a 3D setting
assuming the 1D x-ray transform.
Results: The pseudo-differential operator is a powerful theoretical tool. There is a fundamental mathematical link between
the Gelfand-Graev formula and the DBP (or BPF) approach in the case of the 1D x-ray transform in a 3D real space.
Conclusion: This paper shows the power of mathematics for tomographic imaging, and the value of a pure theoretical
finding, which may appear quite irrelevant to daily healthcare at the first glance.
Index Terms X-ray transform, computed tomography (CT), image reconstruction, pseudo-differential operator,
differential backprojection (DBP), backprojection filtration (BPF).

1. INTRODUCTION

In a series of papers [1-3] and a classical book [4] I. M. Gelfand, M. I. Graev and others proved inversion formulas for
integral transforms of various dimensions in different spaces. In particular, Gelfand-Graevs inversion formula [1] reveals a
fundamental relationship which expresses the Hilbert transform along a chord in the field of view in terms of projection data.
This relationship was re-discovered in recent years; see [5-10]. It has wide applications in the CT field, including
reconstruction from truncated projections [11, 12], differential backprojection (DBP)/backprojection filtration (BPF) [13],
interior tomography [14-18], and limited-angle tomography [19]. For a survey of these results, please see [20-22]. In this
paper, we will focus on [1] and present their inversion formula and proof for the 1D x-ray transform in a 3D real space in an
engineer-friendly language.

2. NOTATION

We follow closely the notations in [1] and consider the 1D x-ray transform in a 3D real space R
3
. Set o = (o
1
, o
2
, o
3
) e
R
3
-{0] and [ = ([
1
, [
2
, [
3
) e R
3
. Denote by E
1,3
the set of all straight lines in R
3
. Then, any (o, [) e (R
3
-{0]) R
3

defines a line b e E
1,3
by ot +[, t e R (Figure 1). Two (o, [), (o', [') e (R
3
-{0]) R
3
define the same b if and only if
o
i
= oz, [
i
= [ +ot
0
for some z e R-{0], t
0
e R. (1)
Let be a compactly supported function in R
3
. The 1D x-ray projection is an integral transform given by
J(o, [) = _ (ot +[)
R
Jt. (2)
Denote (o, [) = J(o, [) and define a differential on (R
3
-{0]) R
3
by
=
o(o, [)
o[
]
3
]=1
Jo
]
. (S)

By Eq. (1), we have
J(o
i
, [
i
) = _ (o
i
t +[
i
)
R
Jt = _ (ozt +[ +ot
0
)
R
Jt
= _ (o(zt +t
0
) +[)
R
Jt =
1
z
_ (ot +[)
R
Jt =
1
z
J(o, [) (4)
and

o(o', [')
o[
]
i
3
]=1
Jo
]
i
=
1
z

o(o, [)
o[
]
i
3
]=1
Jo
]
i
=
o(o, [)
o[
]
3
]=1
Jo
]
, (S)
i.e., the differential depends on b only, being independent of the choice of (o, [).
Let E
0
be the set of oriented 1D lines in R
3
through the origin. Since an oriented line has a direction, E
0
can be viewed as
the unit sphere S
2
in R
3
, and o e S
2
defines the oriented line ot e E
0
, t e R. Let us fix a 1D oriented curve y. As shown in
Figure 2, y can be represented as an oriented curve C
y
c S
2
. For any vector x e R
3
, denote by x +y the set of oriented lines
in y shifted to x. Then, x +y can be expressed as ot +x, o e C
y
, t e R. Define an operator
y
on = J by
(
y
)(x) =
1
2ni
_
x+y
=
1
2ni
_
o(o, [)
o[
]
_
[=x
3
]=1
Jo
]
C
y
. (6)
Take e R
3
-{0]. Denote by 0
{
the set of all lines b e E
0
on the plane x = u. Then, 0
{
can be represented as a
large circle C
{
, which is the intersection between the unit sphere S
2
and the plane x = u, oriented by the right-hand rule
with pointing to the direction of the thumb.
Let us define the Crofton function Cif
y
() as the index for the intersections between C
y
and C
{
. An intersection between
C
y
and C
{
contributes 1 to Cif
y
() if u
y
> u, where u
y
is the tangent vector of C
y
at the intersection (see Figure 3). An
intersection contributes -1 if u
y
< u, and 0 if u
y
= u. Note that and Cif
y
(-) = -Cif
y
(), and for a continuous or
otherwise properly constructed y Cif
y
() = u, _1.
Based on the notations defined above, in the next section we will introduce the pseudo-differential operator defined by
Gelfand and Graev, and show that the composition operator

J is a pseudo-differential operator whose symbol is the


Crofton function. In Section 4, we will present the Gelfand and Graev inverse formula
y
J
y
for the integral transform
= J, which was derived from the properties of the pseudo-differential operator. In Section 5, we will show that the
composition operator
y
J in the 3D real space reveals a Hilbert transform relationship between the differentiated
backprojection image and the original image, which was first revealed by Gelfand and Graev. In the last section, we will
discuss some related issues.

3. PSEUDO-DIFFERENTIAL OPERATORS

Let P() = o
n n

n
be a linear differential operator with constant coefficients o
n
and on smooth functions on R
3
, and
P() = o
n n

n
be a polynomial of three variables, where
n = (n
1
, n
2
, n
3
) , = (
1
,
2
,
3
),
n
=
1
n
1

2
n
2

3
n
3
, = (-io
1
, -io
2
, -io
3
),
n
= (-io)
1
n
1
(-io)
2
n
2
(-io)
3
n
3
.
Define the Fourier transform and its inversion as

`
() =
1
(2n)
32
_ c
-x{
(x)Jx
R
3
, (x) =
1
(2n)
32
_ c
x{

`
()J
R
3
. (7)
Then, by the Fourier inversion formula
P()u(x) = P() _
1
(2n)
3 2
_ c
x{
u()J
R
3
_
=
1
(2n)
3 2
_ P()c
x{
u()J
R
3


=
1
(2n)
3 2
_ _ o
n
(-i)
n
1
+n
2
+n
3
o
n
1
+n
2
+n
3
c
x{
ox
1
n
1
ox
2
n
2
ox
3
n
3
n
_u()J
R
3

=
1
(2n)
3 2
_ _ o
n
(-i)
n
1
+n
2
+n
3
(i
1
)
n
1
(i
2
)
n
2
(i
3
)
n
3
c
x{
n
] u()J
R
3

=
1
(2n)
3 2
_ _ o
n

1
n
1

2
n
2

3
n
3
n
] c
x{
u()J
R
3

=
1
(2n)
3 2
_ P()c
x{
u()J
R
3
.
This suggests the following definition. A pseudo-differential operator C() on R
3
is defined by
C()u(x) =
1
(2n)
3
_ _ c
(x-){
C()u(y)
R
3
Jy
R
3
J
=
1
(2n)
32
_ c
x{
C()u()J
R
3
(8)

for a certain function C() which is called the symbol of the pseudo-differential operator C().

Theorem 1. (Gelfand and Graev [1]) The composition operator
y
J defined by Eqs. (2) and (6) is a pseudo-differential
operator on R
3
with the symbol Cif
y
().
Proof. By Eqs. (2) and (6), we have
(
y
J)(x) =
1
2ni
_
o(o, [)
o[
]
_
[=x
3
]=1
Jo
]
C
y

=
1
2ni
_
o
o[
]
__ (ot +[)Jt
R
__
[=x
3
]=1
Jo
]
C
y

=
1
2ni
_
o
o[
]
__
1
(2n)
32
_ c
(ut+[){

`
()JJt
R
3
R
__
[=x
3
]=1
Jo
]
C
y

=
1
(2n)
52
_ __ _ c
(ut+x){

`
()
]
JJt
R
3
R
_
3
]=1
Jo
]
C
y

=
1
(2n)
52
_ __ c
x{

`
()
]
R
3
_ c
tu{
JtJ
R
_
3
]=1
Jo
]
C
y

=
1
(2n)
32
_ __ c
x{

`
()
]
o(o )J
R
3
_
3
]=1
Jo
]
C
y

=
1
(2n)
32
_ c
x{

`
()
R
3
__ o(o )
]
3
]=1
Jo
]
C
y
_J, (9)
where we have used the Fourier inversion formula and the fact that
1
2n
_ c
tu{
Jt
R
= o(o ),
with o() being the Dirac delta function. By Eq. (9),
y
J is a pseudo-differential operator with the symbol
C() = _ o(o )
]
3
]=1
Jo
]
C
y
= __ o(o )Jo
]
C
y
_
3
]=1

]
.
We want to prove that C() = Cif
y
() for e R
3
-{0]. To that effect, we only need to prove C() = Cif
y
() for a small
local piece of y, as illustrated in Figure 3. Let s, s
1
s s
2
, be a variable. For each oriented line b(s) e y depending on s
smoothly, we have its representation o(s)t, o(s) e C
y
, t e R. Denote u(s) = o(s) , we have
C() = _ o(u(s))
Ju
Js
Js
s
2
s
1
= _ o(u)Ju
b
u
,
where o = u(s
1
) = o(s
1
) and b = u(s
2
) = o(s
2
) . If C
y
does not intersect C
{
, u(s) is never zero, and ] o(u)Ju
b
u
= u.
If C
y
intersects C
{
once, then C() = 1 if o < u < b, and C() = -1 if o > u > b, which is exactly Cif
y
(). If C
y
is tangent
to C
{
at one point, o and b are of the same sign, and C() = u. Alternatively, we may compute
C() =
1
2n
_ _ c
tu
Jt
R
Ju
b
u
=
1
2n
_ _ c
tu
Ju
b
u
Jt
R
=
1
2ni
_
c
bt
-c
ut
t
Jt
R
, (1u)

which is a Froullani integral. Eq. (10) equals 0 if o and b are of the same sign, must be 1 if o < u < b, and be -1 if o > u >
b. Consequently, C() = Cif
y
(), and Theorem 1 follows.


4. INVERSION FORMULA

First, we have a corollary.
Corollary. (Gelfand and Graev [1]) For two oriented curves y
1
and y
2
, the symbol of the pseudo-differential operator
(
y
1
J)(
y
2
J) is
C() = Cif
y
1
()Cif
y
2
().
Proof. This is the composition law of pseudo-differential operators. In fact, by Eq. (8) we have
C
1
()C
2
()u(x) =
1
(2n)
32
_ c
x{
C
1
()(C
2
()u)

()J
R
3
=
1
(2n)
32
_ c
x{
C
1
()C
2
()u()J
R
3
.
A curve y c E
0
is called a quasicycle if |Cif
y
()| is a constant function for almost every . In particular, y is a quasicycle
if |Cif
y
()| = 1 for almost all e R
3
-{0]. This is the case when C
y
is a smooth curve on S
2
whose end points are
diametrically opposite C
y
(s
2
) = -C
y
(s
1
). Figure 4 shows several quasicycles. For a quasicycle y, we denote the constant
|Cif
y
()| by c(y). Note that in this case Cif
y
() itself is not a constant in general.
By the Corollary, if y is a quasicycle, then the symbol of (
y
J)
2
is
[Cif
y
()
2
= |Cif
y
()|
2
= (c(y))
2
,
which is a constant. A pseudo-differential operator with a constant symbol (c(y))
2
is indeed (c(y)
2
)E, where E is the
identity operator, by the Fourier inversion formula. Therefore, we have proved an inversion formula for the x-ray transform
J.
Theorem 2. (Gelfand and Graev[1]) If y is a quasicycle in E
0
, then (
y
J)
2
= (c(y)
2
)E. Thus, for the integral
transform = Jf, one has the inversion formula

y
J
y
= c(y)
2
. (11)
In the next section, we will show that a major implication of Eq. (11) for CT reconstruction for any curve y that is a
quasicycle.


5. HILBERT TRANSFORM

From Eqs. (9) and (10), we have
(
y
J)(x) =
1
(2n)
3
2
_ c
x{

`
()
R
3
1
2ni
_ (c
t(u(s
2
){)
-c
t(u(s
1
){)
)
Jt
t
J
R

=
1
2ni
_
1
(2n)
32
R
_ (c
t(u(s
2
){)
-c
t(u(s
1
){)
)c
x{

`
()J
Jt
t

R
3

=
1
2ni
PI _ ((o(s
2
)t +x) -(o(s
1
)t +x))
Jt
t
R
, (12)
where the Fourier inversion formula has been used, and the PV integral takes the Cauchy principal value. Eq. (12) shows
that (
y
J)(x) is the difference between the Hilbert transforms at the point x along the oriented lines o(s
2
)t +x and
o(s
1
)t +x, t e R. The Hilbert transform relationship Eq. (12) is exactly the only example given in Section 8 in [1]. When y
is a closed curve, i.e., when the corresponding C
y
is a closed curve on S
2
with o(s
2
) = o(s
1
), from Eq. (12) we can see
(
y
J)(x) u. When y is a quasicycle, i.e., when the corresponding C
y
is a smooth curve on S
2
whose end points are
diametrically opposite, Eq. (12) can be simplified as
(
y
J)(x) =
1
2ni
PI _ ((o(s
2
)t +x) -(-o(s
2
)t +x))
Jt
t

R

=
1
ni
PI _ (o(s
2
)t +x)
Jt
t
R
, (1S)
which shows a Hilbert transform along the line parallel to o(s
2
) and through the point x. In other words, the composition
operator
y
J is a Hilbert transform, which is the DBP approach. It is well known that the Hilbert transform can be inverted
using itself. Then, performing a Hilbert filtering operation on Eq. (13) along the line parallel to o(s
2
) and through x, we can
exactly reconstruct the image (x) on that line, which is a good example of the general inversion scheme Eq. (11).
In the CT field, projection data is usually measured in a parallel-, fan- or cone-beam geometry. When a parallel-beam
geometry is assumed, the scanning trajectory can be viewed as C
y
and represented by o(s) on the unit sphere, where s with
s
1
s s
2
is a variable. For any fixed s, J(o(s), [) will be measured for all [ e R
3
. Particularly, (x) will be measured
by J(o(s), x) for s
1
s s
2
. When (s
1
) = -o(s
2
) , we have Eq. (13). Because o(s) is independent of x, Eq. (13) holds
for any x. Therefore, Eq. (11) becomes the backprojection filtration (BPF) reconstruction formula in a parallel-beam
geometry.
When a fan- or cone-beam geometry is considered [10], the scanning trajectory can be represented as a curve (s),
where s with s
1
s s
2
is a variable. For any fixed s, J(o, (s)) will be measured for all o e S
1
or S
2
. Particularly, (x)
will be measured by J(o(s, x), (s)) = J(o(s, x), x) with o(s, x) = (x -(s)) [x -(s)[ for s
1
s s
2
(see
Figure 5(a)). In this case, the curve C
y
depends on x and should be denoted it as C
y
x
as pointed out in Section 9 in [1].
Substituting o(s, x) into Eq. (6), we obtain
(
y
x
J)(x) =
1
2ni
_
o(o(s, x), [)
o[
]
_
[=x
3
]=1
J(o(s, x))
]
C
y
x

=
1
2ni
_
o(x -(s), [)
o[
]
_
[=x
3
]=1
J(x -(s))
]
, (14)
(x-A(s))

where we have used the fact proved in Eq. (5) that the differential depends on the line only and is independent of the choice
of (o, [). Thus, by Eq. (4)
(
y
x
J)(x) =
1
2ni
_ ('(s))
j
o(x -(s), [)
o[
]
_
[=x
3
]=1
Js
s
2
s
1

=
1
2ni
_
('(s))
j
[x -(s)[
o(o(s, x), [)
o[
]
_
[=x
3
]=1
Js. (1S)
s
2
s
1

Consequently, we have
(
y
x
J)(x) =
1
2ni
_
('(s))
j
[x -(s)[
3
]=1
_
i
(2n)
32
_ _ c
(u(s,x)t+x){

`
()
]
JJt
R
3
R
_Js
s
2
s
1

=
1
2ni
_
1
[x -(s)[
_
i
(2n)
32
_ _ c
(u(s,x)t+x){

`
()('(s) )JJt
R
3
R
_Js
s
2
s
1

=
1
2ni
_
1
[x -(s)[

o(o(s, x), (q))
oq
_
q=s
Js
s
2
s
1
, (16)
where in the last step we have used the fact

o(o(s, x), (q))
oq
_
q=s
=
o
oq
_
1
(2n)
32
_ _ c
(u(s,x)t+A(q)){

`
()JJt
R
3
R
__
q=s

=
i
(2n)
32
_ _ c
(u(s,x)t+A(q)){

`
()('(q) )JJt
R
3
R
_
q=s

=
i
(2n)
32
_ _ c
(u(s,x)t+A(s)){

`
()('(s) )JJt
R
3
R

=
i
(2n)
32
_ _ c
(u(s,x)t+x){

`
()('(s) )JJt. (17)
R
3
R

Combining Eqs. (12) and (14)-(16), we arrive at
(
y
x
J)(x) =
1
2ni
_
o(o(s, x), [)
o[
]
_
[=x
3
]=1
J(o(s, x))
]
C
y
x

=
1
2ni
_
1
[x -(s)[

o(o(s, x), (q))
oq
_
q=s
Js
s
2
s
1

=
1
2ni
PI _ ((o(s
2
, x)t +x) -(o(s
1
, x)t +x))
Jt
t
R
. (18)
When x belongs to the chord connecting the two end points of the scanning trajectory (s), we have o(s
1
, x) = -o(s
2
, x)
and Eq. (18) is a DBP formula, which clearly suggests the BPF approach for image reconstruction. Combined with the finite
Hilbert transform inversion technique [23, 24], the BPF approach immediately lead to the BPF algorithms for standard and
non-standard helical cone-beam scanning in the CT field [6-10, 13, 25-29].

6. DISCUSSIONS AND CONCLUSION

For practical CT systems, the projection is a half-ray transform and can be defined as
J
+
(o, ) = _ (ot +)
+
0
Jt. (19)
Because an object (x) to be reconstructed in biomedical applications is compactly supported and the trajectory (s) is
always outside the convex hull of the object support in biomedical applications, we have J
+
(-o, ) u and J
+
(o, )
J(o, ). As a result, the object is compactly supported on the chord, as shown in Figure 5 (b). Thus, the image on the chord
can be exactly reconstructed by performing the inverse Hilbert transform or inverting the truncated Hilbert transform [23,
24]. As a side note, when the BPF algorithms were developed in the CT field, two data extension schemes were used in the
proofs. They are the even data extension (J
+
(o, ) +J
+
(-o, )) and the odd data extension (J
+
(o, ) -
J
+
(-o, )). Clearly, Eq. (18) is based on the even extension. Normally, the two extensions are equivalent. However, in a
rare case (the scanning trajectory intersects the convex hull of the object support), J
+
(-o(s, x), (s)) u, and only the
odd extension allows theoretically exact reconstruction on the chord. Please see the relevant references for more details [6, 8-
10, 25]. We would like to point out that Gelfand and Graev [1] provided no suggestion on how the non quasi-cycle formula
could be beneficial for cone-beam reconstruction, whereas the results in [9] described a powerful methodology using the
concepts of R-lines and M-lines.
After [1], the Hilbert transform relationship was extended by Rullgrd in the case of the attenuated Radon transform for
single photon emission computed tomography [5]. Also, we would like to make a few comments on a Fourier-transform-
based general CT scheme presented by Wei et al. [10]. First, Theorem 1 (Galfand and Graev) in the parallel-beam scenario
was re-generated in R
3
when As=1 in Case II (Eq. 27 in [10]) of that general CT scheme. The formula Eq. (13) is closely
related to Eq. (28) in [10] and the existing FBP formulas with the even extension in the fan- and cone-beam geometry, which
were re-derived in the same way by Wei et al. [10]. Second, while the theory in [1] provides a unified reconstruction formula
in a higher dimensional space using the pseudo differential operator, the scheme in [10] works in 2D and 3D through a
general weighting function. This weighting function can be specialized to give not only the BPF formulas but also other
important CT algorithms [30, 31].
A few more comments are in order on the relationship between the Gelfand and Graev formula and some recent work in
the CT field. First, the DBP approach (the same as the BPF approach) requires the inverse Hilbert transform after Gelfand
and Graevs formulation in Section 8 in [1]. Evidently, Gelfand and Graev revealed a fundamental link between the
differential backprojection and the Hilbert transform. The overall purpose of their work is the recovery of an underlying
function from its integral transform. Hence, the 2D or 3D image reconstruction from parallel-, fan- or cone-beam projections
is a special case of their general scheme. Therefore, once they presented the example in Section 8 in [1], it goes without
saying that they had suggested the BPF in the parallel-beam geometry. Furthermore, their Hilbert transform relationship
assumes any path on a unit sphere, including but not restricted to a quasicycle. That is, they are the first who proposed the
BPF approach. On the other hand, we want to point out that the application of the inverse truncated Hilbert transform and the
aforementioned odd data extension for DBP-based CT reconstruction of a compactly supported object is an original
contribution by CT researchers[6, 8, 11-14, 16, 27-29].
Second, it is mathematically trivial to extend the parallel-beam BPF formula to divergent-beam BPF formulas. Indeed,
Gelfand and Graev pointed out this connection in a more general setting in Section 9 in [1]. Note that they used the word
easily, and indeed we exactly followed their instruction in the case of the x-ray transform in a 3D real space, and re-derived
the same fundamental link between the differential backprojection and the Hilbert transform, which is Eq. (18) in our
manuscript.
In other words, both the DBP (or BPF) approach/formulas suggested by Gelfand and Graev and the DBP (or BPF)
algorithms developed by CT researchers are valuable in different ways. Mathematically, the Gelfand and Graev theory is
very abstract and yet powerful, which gives a recipe dealing with image reconstruction in a general space from its integral
transform of various forms, which are expressed in mathematical jargons. Technically, it is not trivial to have working
algorithms designed and coded directly from their theory. When an object to be reconstructed is compactly supported, the
truncated Hilbert transform technique is crucial, which was not mentioned by Gelfand and Graev. Thus, the mathematical
conception of DBP/BPF was suggested by Gelfand and Graev, and at the same time its application for reconstruction of a
compactly supported object and technical implementation should be credited to CT researchers.


7. ACKNOWLEDGEMENT

This work was partially supported by the NSF/MRI program (CMMI-0923297) and NIH/NIBIB grants (EB009275 and
EB011785). The authors thank Dr. Yuchuan Wei for his analysis on the relationship between the Gel'fand formula in [1] and
the general CT scheme in [10], and very insightful and constructive comments from the editor, reviewers, Dr. Yu Zou, and
Dr. Xiaochuan Pan.


8. REFERENCES

[1] I.M. Gel'fand, and M.I. Graev, Crofton function and inversion formulas in real integral geometry. Functional Analysis and its
Applications, 1991. 25(1): p. 1-5.
[2] I.M. Gel'fand, and S.G. Gindikin, Nonlocal inversion formulas in real integral geometry. Functional Analysis and its Applications,
1977. 11: p. 173-179.
[3] I.M. Gel'fand, M.I. Graev, and Z.Y. Shapiro, Integral geometry on manifolds of k-dimensional planes. Soviet Mathematics - Doklady,
1966. 7: p. 801-804.
[4] I.M. Gelfand, M.I. Graev, and N.Y. Vilenkin, Generalized functions. Vol. 5: Integral geometry and representation theory. 1966, New
York-London: Academic Press.
[5] H. Rullgard, An explicit inversion formula for the exponential radon transform using data from 180 ~. Ark. Mat.,, 2004. 42: p. 353-
362.
[6] Y.B. Ye, S.Y. Zhao, H.Y. Yu, and G. Wang, Exact reconstruction for cone-beam scanning along nonstandard spirals and other
curves. Proceedings of SPIE,Vol. 5535, p:293-300, 2004, Denver, CO, United States.
[7] S.Y. Zhao, H.Y. Yu, and G. Wang. A family of analytic algorithms for cone-beam CT. Proceedings of SPIE, Vol. 5535, p:318-328,
2004, Denver, CO, United States.
[8] Y. Zou, and X.C. Pan, An extended data function and its generalized backprojection for image reconstruction in helical cone-beam
CT. Phys. Bio. Med., 2004. 49(22): p. N383-N387.
[9] J.D. Pack, F. Noo, and R. Clackdoyle, Cone-beam reconstruction using the backprojection of locally filtered projections. IEEE
Transactions on Medical Imaging, 2005. 24(1): p. 70-85.
[10] Y.C. Wei, et al, Scheme of computed tomography. Journal of X-ray Science and Technology, 2007. 15(4): p. 235-270.
[11] F. Noo, R. Clackdoyle, and J.D. Pack, A two-step Hilbert transform method for 2D image reconstruction. Phys. Bio. Med., 2004.
49(17): p. 3903-3923.
[12] M. Defrise, F. Noo, R. Clackdoyle, and H. Kudo, Truncated Hilbert transform and image reconstruction from limited tomographic
data. Inverse Problems, 2006. 22(3): p. 1037-1053.
[13] Y. Zou, X.C. Pan, and E.Y. Sidky, Image reconstruction in regions-of-interest from truncated projections in a reduced fan-beam scan.
Phys. Bio. Med., 2005. 50(1): p. 13-27.
[14] Y.B. Ye, H.Y. Yu, Y.C. Wei, and G. Wang, A general local reconstruction approach based on a truncated Hilbert transform.
International Journal of Biomedical Imaging, 2007. Article ID: 63634, 8 pages.
[15] Y.B. Ye, H.Y. Yu, and G. Wang, Exact interior reconstruction with cone-beamCT. International Journal of Biomedical Imaging,
2007. Article ID: 10693, 5 pages.
[16] H.Y. Yu, Y.B. Ye, and G. Wang, Local reconstruction using the truncated Hilbert transform via singular value decomposition.
Journal of X-Ray Science and Technology, 2008. 16(4): p. 243-251.
[17] H. Kudo, et al., Tiny a priori knowledge solves the interior problem in computed tomography. Phys. Bio. Med., 2008. 53(9): p. 2207-
2231.
[18] M. Courdurier, et al., Solving the interior problem of computed tomography using a priori knowledge. Inverse Problems, 2008. 24,
Article ID 065001 , 27 pages.
[19] Y.B. Ye, H.Y. Yu, and G. Wang, Exact interior reconstruction from truncated limited-angle projection data. International Journal of
Biomedical Imaging, 2008. Article ID: 427989, 6 Pages.
[20] G. Wang, Y.B. Ye, and H.Y. Yu, Approximate and exact cone-beam reconstruction with standard and non-standard spiral scanning,
Physics in Medicine and Biology, 2007, 52(6):p. R1-R13..
[21] Y.B. Ye, H.Y. Yu, and G. Wang, Interior tomography: Mathematical analysis, in Biomedical mathematics: Promising directions in
imaging, therapy planning, and inverse problems, Y. Censor, M. Jiang, and G. Wang, Editors. 2010, Medical Physics Publishing.
[22] G. Wang, H.Y. Yu, and Y.B. Ye, Interior tomography: Practical applications, in Biomedical mathematics: Promising directions in
imaging, therapy planning, and inverse problems, Y. Censor, M. Jiang, and G. Wang, Editors. 2010, Medical Physics Publishing.
[23] H. Soehngen, Die losungen der integralgleichung under deren anwendung in der tragflugeltheorie. Mathematische Zeitschrift, 1939.
45: p. 245-264.
[24] F.G. Tricomi, On the finite Hilbert Transform, Quarterly Journal of Mathematics 1951. 2(1): p. 199-211.
[25] Y.B. Ye, S.Y. Zhao, H.Y. Yu, and G. Wang, A general exact reconstruction for cone-beam CT via backprojection-filtration. IEEE
Transactions on Medical Imaging, 2005. 24(9): p. 1190-1198.
[26] S.Y. Zhao, H.Y. Yu, and G. Wang, A unified framework for exact cone-beam reconstruction formulas. Medical Physics, 2005. 32(6):
p. 1712-1721.
[27] T.L. Zhuang, et al., Fan-beam and cone-beam image reconstruction via filtering the backprojection image of differentiated projection
data. Phys. Bio. Med., 2004. 49(24): p. 5489-5503.
[28] Y. Zou, and X.C. Pan, Exact image reconstruction on PI-lines from minimum data in helical cone-beam ct. Phys. Bio. Med., 2004.
49(6): p. 941-959.
[29] Y. Zou, X.C. Pan, and E.Y. Sidky, Theory and algorithms for image reconstruction on chords and within regions of interest. J. Opt.
Soc. Am. A, 2005. 22(11): p. 2372-2384.
[30] A. Katsevich, A general scheme for constructing inversion algorithms for cone-beam CT. Int. J. of Math. and Math. Sci., 2003. 21: p.
1305-1321.
[31] A. Katsevich, Improved cone beam local tomography. Inverse Problems, 2006. 22(2): p. 627-643.




FIGURE CAPTIONS

Figure 1. Representation of a line with two vectors.
Figure 2. Representation of oriented lines with a curve which can be shifted to x.
Figure 3. Crofton function in different cases.
Figure 4. Various quasicycles.
Figure 5. Fan- or cone-beam geometry for (a) the Gelfand and Graev formula and (b) the chord-based reconstruction.

You might also like