Proceedings of the 2005 IEEE
Engineering in Medicine and Biology 27th Annual Conference
Shanghai, China, September 1-4, 2005
Data Correction for Gantry-tilted Local CT
Hongzhu Liang, Cishen Zhang, Ming Yan, Jiayin Zhou
Abstract— Gantry-tilted helical multi-slice Computed Tomog-
raphy (CT) refers to the helical scanning CT system equipped
with multi-row detector operating at some gantry tilting angle.
Its purpose is to avoid the area which is vulnerable to the
X-ray radiation. The local tomography is to reduce the total
radiation dose by only scanning the region of interest for
image reconstruction. In this paper we consider the scanning
scheme, and incorporate the local tomography technique with
the gantry-tilted helical multi-slice CT. The image degradation
problem caused by gantry tilting is studied, and a new error
correction method is proposed to deal with this problem in the
local CT. Computer simulation shows that the proposed method
can enhance the local imaging performance in terms of image
sharpness and artifacts reduction.
I. INTRODUCTION
In the past decade, there are two major technical achieve-
ments on Computed Tomography (CT): the helical scanning
CT (HCT) and multi-slice CT (MCT) [1]-[4]. The HCT
enables a continuous and fast scanning than the step-and-
shoot operation, while the MCT offers a big volume coverage
leading to enhancement of the scanning speed and image
resolution in the patient feeding direction.
In medical diagnosis, it sometimes needs to tilt the CT
gantry with respect to the patient and couch for some special
scanning. After the gantry tilting, however, the geometry of
helical MCT is made more complex, and the image quality
can be unfortunately degraded. This introduced errors in the
projection data set and this problem has been studied and
some methods have been proposed for the error correction
to improve the processing and image quality of the gantry-
tilted CT [5][6].
The gantry-tilted CT scanning is applied to the situation
where higher scanning safety is needed and scanning of the
vulnerable to radiation area is to be avoided. In this sense, it
is quite meaningful to apply the local tomography technique
in the gantry-tilted scanning. The local tomography refers
to reconstructing the image of local region of interest (ROI)
using only local projection data and aims at reducing the
X-ray exposure to the patient.
In this paper we incorporate the wavelet based local
tomography ([7]-[10]) into the gantry-tilted MCT reconstruc-
tion, and propose a data-correction method for reducing the
image degradation caused by the gantry tilting. We will show
that imaging performance can be enhanced by the proposed
method.
All authors are with School of Electrical & Electronic
Engineering, Nanyang Technological University, 639798, Singapore.
jyzhou@ntu.edu.sg
Cishen Zhang (Dr) is also with School of Chemical & Biomedical
Engineering, Nanyang Technological University.
0-7803-8740-6/05/$20.00 ©2005 IEEE. 1798
x
A.
Gantry rotation Tilting angle
plane
s Projection angle
Source zt
O z
Couch
y Translation
t
B.
Reconstruction
Gantry isocenter isocenter
Source
Patient
Translation Detector
Fig. 1. The gantry-tilted helical multi-slice scanning.
The outline of this paper is as follows. In Section II the
geometry of gantry-tilted helical multi-slice CT scanning is
described and the image degradation due to the isocenter
derivation in the gantry tilted scanning is analyzed. The
proposed error correction method for dealing with the image
degradation is presented in Section III, followed by the
local reconstruction procedure in Section IV. Computered
simulation results of the proposed and existing methods’
comparison are given in Section V.
II. TILTED HELICAL MULTI-SLICE CT SCANNING
The geometry of gantry-tilted helical multi-slice scanning
is shown in Fig.1. The couch and patient is translating
through the gantry along axis z, while the X-ray source
rotating in the gantry circle which has axis zt . Between axis
z and zt is the gantry tilting angle, named as θ. On the
gantry tilted plane, we define two coordinates x − y − zt and
t − s − zt . The axis x − y are fixed on the gantry plane while
the t − s rotate with the X-ray source. The rotation angle
between axis s and y is named as β.
In helical multi-slice system, since the reconstruction slice
moves at couch translation direction while scanning, an
accurate projection data set could hardly be achieved. This
requires approximation of the data in the z-direction using
appropriate projection data. A straightforward way is to find
the data closest to the expecting slice (named as z0 ) at each
 A.
projection angle. This is a “one-data” approximation which Detector
yields considerable errors. In [2], two data being closest to Row n+1
slice z0 from both sides are employed for linear two-point Slice Z1

interpolation to approximate the real data. For a higher z- Row n

Slice Z0
resolution, in [3] a “z-filtering” method was proposed in
which at each projection angle two or more than two data
are adopted for resampling and filtering in z-direction before Translation
forming a new data. From these methods, we can see that at axis
each projection angle the data needed for reconstruction is Source
formed by two or more data at different z-location. y
Unlike the normal scanning, in the gantry-tilted scanning B. S
s S’
system the objective moving direction (z-axis, in Fig.1)
deviates from the gantry rotation axis (zt -axis). It is that
axis deviation (or isocenter deviation) that ruins the sampling
continuity of the original data set along the z direction. So
there will be error if we directly treat these data without t
correction as we do in the gantry-nontilted scanning system. C0 C1
In the next section, this error will be investigated and a new O
data correction approach will be proposed. x
III. ERROR CORRECTION IN LOCAL
TOMOGRAPHY p

A. Error Analysis D1
.
It is already known that the isocenter deviation of tilted O
D’
or
HCT scanning will cause the sampling deviation at the tect
De
patient translating direction producing reconstruction errors. D0
As shown in Fig.2.B, the ROI’s center C0 at expected slice
z0 and C1 at slice z1 are both projected onto detector plane. Fig. 2. The deviation between two slices z0 and z1 .
Referring to Fig.2.A, C0 C1 is parallel to x axis, |C0 C1 | =
(z0 − z1 )tanθ.
Assuming the x − y coordinates of C0 and C1 are (x0 , y0 ) The above value ∆d0 is the deviation of the local area’s
and (x1 , y1 ), and the t − s coordinates are (t0 , s0 ) and center. For an arbitrary point (x, y) at the projection path
(t1 , s1 ). Then we have: SD0 , the deviation value ∆d can be calculated similarly.




 In this paper the “two-point linear interpolation” method is
t0 x0 t1 x
=Λ· , =Λ· 1 , (1) considered, which is commonly adopted in the 2-dimensional
s0 y0 s1 y1 cone-beam reconstruction [1], [2], as follows:


cosβ sinβ z2 − z0 z0 − z1
where Λ = . Pn (p, β, z0 ) = P (p, β, z1 ) + P (p, β, z2 ),
−sinβ cosβ z2 − z1 z2 − z1
As Fig.2 shows, the X-ray projection of C0 and C1 onto (4)
detector are named as D0 and D1 . Now we can calculate the where Pn () is the new data set to be formed, P () is the
deviation, as: original data. p denotes the data position coordinates on row,
β is the rotation angle, and z is the data’s z-location: z0
Dsd t0 Dsd t1
D0 D1 = − denotes the expected slice, z1 and z2 are the two closest
Dso − s0 Dso − s1 slices at opposite sides. Because there is isocenter deviation
 x0 cosβ + y0 sinβ
= Dsd − in the gantry-tilted system, the continuous samples along the
Dso − (−x0 sinβ + y0 cosβ) z-axis no longer locate at the same p position like they
x1 cosβ + y1 sinβ  do in gantry-nontilted scanning. It can be seen from (3)
. (2)
Dso − (−x1 sinβ + y1 cosβ) that two corresponding points on neighbour slices will have
where Dsd denotes the distance from source to detector, their projection on detector with a deviation D0 D1 , which
and Dso denotes the source to detector. Since y0 = y1 , the varies with tilting angle θ, projection angle β and the point’s
difference between x0 sinβ and x1 sinβ is small compared position.
with Dso . So we use these facts to have: B. Proposed Data Correction
Dsd (x0 − x1 )cosβ Although the corresponding pairs of points on neighbour
∆d0 = D0 D1 ≈
Dso + x0 sinβ − y0 cosβ slices have parallel and equal shift to each other, the deviation
Dsd (z0 − z1 )tanθ cosβ of their projection on the detector are different since the fan
= . (3)
Dso + x0 sinβ − y0 cosβ magnification varies with the different coordinates. Referring
1799
to Fig.2, for one projection path SD0 on z0 , there is only one 360LI:
“ideal” but “virtual” corresponding path S
D
on z1 , which z2 − z0  
Pn (p, β, z0 ) = P p + ∆d1 , β, z1 +
is parallel to SD0 with a distance |C0 C1 |cosβ. Only this z2 − z1
“ideal” path could bring to SD0 the continuous sample along z0 − z1  
P p + ∆d2 , β, z2 , (8)
the z-axis. To choose a projection path (named as SD1 ) z2 − z1
being closest to the ideal one S
D
, the distance between 180LI:
the real path and ideal path is worked out by calculating the z1 − z0  
average distance of n pairs of points locating on two paths Pn (p, β, z0 ) = P p + ∆d1 , β, z1 +
z2 − z1
respectively and at same y level: z0 − z1  
P p + ∆d2 , βφ , z2 , (9)
1 2 1/2
n
z2 − z1
Dist = (∆xi ) , n ≥ 2, (5)
n i=1 where ∆d1 and ∆d2 are respectively the deviation of slice
“z0 − z1 ” and slice “z0 − z2 ” which is calculated in (6) and
where ∆xi is the distance of two corresponding points at
(7), and βφ is the projection angle of another beam along
real and ideal path. The projection path SD1 is chosen to
the same way but in the opposite direction.
let the Dist be least. This choice not only depends on the
Besides the linear interpolation, a “z-filtering” approach
rotation angle β and ROI’s position, but also considers the
based on the proposed data correction is proposed as follows:
ROI’s shape to a great extent: more points should be involved M
in calculating Dist at wide area than at thin area. Assume Pm (p, β, zm ) fz (zm − z0 )
Pn (p, β, z0 ) = m=1 M , (10)
the expected SD1 joins the ideal path S
D
at (xc , yc ), we
m=1 fz (zm − z0 )
can have the deviation value ∆d on detector as:
where
Dsd (z0 − z1 )tanθ cosβ
∆d ≈ z2 − z1
Dso + (xc + C0 C1 )sinβ − yc cosβ zm = z1 + m, M ∈ Z, M ≥ 2, m = 1, 2, ...M ;
M
Dsd (z0 − z1 )tanθ cosβ (11)
= (6)
Dso + [xc + (z0 − z1 )tanθ]sinβ − yc cosβ z2 − zm 
Pm (p, β, zm ) = P p1 + λ1 , β, z1 ) +
In terms of the detector cell size, the compensation value z2 − z1
zm − z1 
can be shown as: P p2 + λ2 , β, z2 ), (12)
∆dn = ∆d,
N
(7) z2 − z1
L m m
λ1 = (∆d1 + ∆d2 ) , λ2 = (∆d1 + ∆d2 )( − 1). (13)
where n is slice number, L is the length of detector slice, N M M
is the whole number of detector cell per slice. For practice, The fz (z) is a window filter chosen by user. The data
the deviation values can be calculated and stored in stacks compensation in (6, 7) and the interpolation methods in
in the order of p value and angle β before scanning. The (8)–(10) are the proposed work which will be simulated in
original projection data set is compensated by these values Section V.
to form a new data set.
The above results are calculated in the equally-spaced fan- IV. RECONSTRUCTION
beam scanning. For the equally-angle scanning the results In this paper we incorporate the wavelet-based local
can be calculated similarly. tomography into the gantry-tilted MCT reconstruction. To
To neutralize these deviations, it was proposed in [6] to reconstruct the local image, the 2-dimensional(2D) separable
implement linear or nonlinear or frequency domain inter- wavelet-based approach is adopted [9]. The wavelet-based
polation among these slices to compensate the deviation. local reconstruction can be concluded as three steps: 1,
That linear interpolation method compensates the projection Filtering: The ramp filter H is replaced by four 2D wavelet-
data with deviation value calculated only at reconstruction modified filters. The completed data set P is respectively
center. The nonlinear or frequency domain methods re- filtered into four data sets. 2, Reconstruction: Four sub-
quire many more calculations. This paper proposes a data- images could be reconstructed, each of which represents the
corrected linear interpolation approach. A position-and-shape objective image’s approximate or detail component. Beware
dependent compensation value is worked out, which leads to that the reconstruction slice is moving at both z and x di-
a more accurate correction to the isocenter deviation in local rection during rotation. 3, Inverse Wavelet Transform (IWT):
reconstruction. It includes the “360LI” and “180LI” linear Combine the four sub-images with IWT to achieve the final
interpolation [2]: expected image.

1800
 A
V. PERFORMANCE COMPARISON
Computer simulation has been carried out to compare the
proposed data-corrected linear interpolation method with the
existing linear approach in [6].
The low-contrast Shepp-Logan cubical phantom is used
for simulation [1]. The grey value of the main area ranges in
[1.0, 1.04] (whole grey range is [1, 2]). It is highly sensitive
to imaging noise. A central area in phantom is chosen as
ROI for the local tomography simulation. The ROI area B
is scanned while the gantry tilts 10o , 20o and 30o . The
proposed data-corrected “360LI” linear z-interpolation (8)
is simulated to compare with the existing approach in [6].
For the wavelet-based local tomography, the “Haar wavelet”
and its corresponding scaling function are adopted for the
multi-resolution reconstruction.
As shown in the reconstruction results in Fig.3 that:
when the tilting angle is relatively small (e.g., 10o ), the
reconstruction performance of existing method is nearly as C
good as that of the proposed approach. While the tilting angle
increases, the difference becomes more and more visible. The
images reconstructed with the existing linear data correction
have severe artifacts. In the right column, however, it can
be seen that the proposed data compensation method can
provide better results: the ellipses in the phantom image still
preserve clear and sharp edges at the big tilting angles.
VI. CONCLUSIONS
The gantry-tilted CT scanning allows scanning the ROI Fig. 3. Local imaging comparison: images of left column are results of
while avoiding the vulnerable-to-radiation area. It is quite existing data correction approach; images of right column are results of
our proposed approach. From upper row to lower row are respectively the
meaningful and useful to employ the local tomography results at tilting angle θ = 10o , θ = 20o , θ = 30o .
technique in the gantry-tilted CT for safer scanning. To
solve the image degradation in the gantry-tilted CT, in this
paper we analyze the data error caused by the isocenter [4] S. Schaller, et al. “Spiral interpolation algorithms for multi-slice spiral
deviation, and propose a linear interpolation method based CT I: Theory,” IEEE Transaction on Medical Imaging, vol.19, No.9,
pp. 822-834, Sep. 2000.
on new data correction scheme to form a completed data [5] I. Hein and K. Taguchi, M.D. Silver, M. Kazama, and I. Mori, “Tilted
set for local reconstruction. Comparing with the existing helical Feldkamp cone-beam reconstruction algorithm for multi-slice
methods, the new proposed approach can provide a better CT,” Proceding of SPIE, vol.5032, pp. 1901-1910, 2003.
[6] J. Hsieh, “Tomographic reconstruction for tilted helical multi-slice
reconstruction performance, as shown in the computer sim- CT,” IEEE Transaction on Medical Imaging, vol.19, No.9, pp. 864-
ulation results. The advantages of proposed method are that 872, Sep. 2000.
the data compensation can vary with the ROI’s position and [7] J. DeStefano, T. Olson, “Wavelet localization of the Radon transform,”
IEEE Trans. Signal Processing, vol. 42, pp. 2055-2067, Aug.1994.
shape, and its computational load is considerably light since [8] A.H. Delaney and Y. Bresler, “Multiresolution tomographic recon-
the compensation matrix can be pre-determined. struction using wavelets,” in Proc. IEEE Int. Conf. Image Processing,
vol.ICIP-94, pp. 830-834, Nov. 1994.
R EFERENCES [9] F. Rashid-Farrokhi, K.J. Ray Liu, C.A. Berenstein, and D. Walnut,
“Wavelet-based multiresolution local tomography,” IEEE Trans. Image
[1] Ge Wang, T. Lin, P. Cheng, and D.M. Shinozaki, “A general cone- Processing, vol. 6, pp. 1412-1430, Oct. 1997.
beam reconstruction algorithm,” IEEE Transaction on Medical Imag- [10] F. Rashid-Farrokhi, K.J. Ray Liu, “Local tomography in fan-beam
ing, vol.12, No.3, pp. 486-496, Sep. 1993. geometry using wavelets,” IEEE Press, pp.709-712, 1996.
[2] Hui Hu, “Multi-slice helical CT: Scan and reconstruction,” Medical
Physics, vol.26, No.1, pp.5-18, Jan. 1999.
[3] K. Taguchi and H. Aradate, “Algorithm for image reconstruction in
multi-slice helical CT,” Medical Physics, vol.25, No.4, pp.550-561,
Apr. 1998

1801