M17-41
2012 IEEE Nuclear Science Symposium and Medical Imaging Conference Record (NSS/MIC)
Development and Assessment of Statistical Iterative Image Reconstruction for CT on a Small Animal SPECT/CT Dual-Modality System Tiantian Dai, Xiao Deng, Ciprian N. Ionita, Qingyang Wei, Tianyu Ma, Member, IEEE, Yaqiang Liu, Daniel R. Bednarek, Stephen Rudin, Life Member, IEEE, Rutao Yao, Senior Member, IEEE Abstract-We developed a statistical iterative CT image reconstruction software for a newly constructed high-resolution small animal SPECT/CT dual-modality system, and assessed its performance
at
different
radiation
exposure
levels.
The
objective of this work was to preserve or improve reconstructed image quality at either the same or reduced animal x-ray radiation
exposure.
The
SPECT/CT
system
used
a
single
detector for both the CT and SPECT modalities that consists of a micro-columnar CsI(TI) phosphor, a light image intensifier (LII) and a CCD sensor. The CT reconstruction software was based on the ordered-subset-convex (OSC) algorithm, and the system matrix was calculated through a ray-driven approach. A self-calibration method was implemented to calculate the offset of the axis of rotation (AOR), an important geometry parameter
is based on a statistical noise model and can accommodate a sophisticated system response model, is a widely accepted solution. The objective of this work was to preserve or improve reconstructed image quality at either the same or reduced x ray radiation exposure using statistical reconstruction for a newly constructed high-resolution small animal SPECT/CT dual-modality system. The reconstruction performance at different radiation exposure levels was assessed. This paper reports only our work on the CT aspects of the system. The overall system development and performance are reported in a separate paper [1].
of the system. An endovascular stent was imaged to evaluate the
II.
high resolution performance of the statistical reconstructed image. A sacrificed mouse was scanned at different exposure levels to assess the effect of statistical noise on the image. The mouse studies were reconstructed with both the statistical reconstruction software and a filtered back-projection (FBP) program. The images were assessed and compared by contrast to noise ratio (CNR) in the region of interest. The images yielded by the statistical reconstruction software were artifact free and show
superior
reconstruction
noise at
performance
different
radiation
to
those
exposure
from
FBP
levels.
The
statistical reconstructed images with reduced exposure showed obviously
higher
image
quality
than
those
from
FBP
reconstruction at full exposure.
I.
A. Small Animal
INTRODUCTION
978-1-4673-2030-6/12/$3l.00 ©2012 IEEE
CT System
Setup
The small animal CT system of the dual function SPECT/CT prototype [1] we constructed consisted of an x ray tube, an animal holder supported by a rotational stage, and a micro-angiographic fluoroscopy (MAF) detector [2], as shown in Fig. l(a). Tomography imaging was implemented by rotating the imaged object while the x-ray tube and detector remained stationary. The diagram of the CT system configuration is shown in Fig. l(b).
MAF Detector
With heightened awareness of health-related risks from radiation exposure, radiation dose reduction has become an important subject in CT technology development. Reducing radiation dose, however, increases the statistical noise in the acquired data. To achieve high quality images with low radiation exposure, statistical iterative reconstruction, which Manuscript received November 16, 2012. This work was supported in part by China Scholarship Council and the Pilot Studies program of the University at Buffalo Clinical and Translational Research Center and the Buffalo Translational Consortium, NIH Grants ROI EB002873 and EB008425. Tiantian Dai, Qingyang wei, Tianyu Ma, Yaqiang Liu are with the Department of Engineering Physics, Tsinghua University, Key Laboratory of Particle & Radiation Imaging (Tsinghua University), Ministry of Education, Beijing 100084 China. (telephone: 86-10-62789644, e-mail:
[email protected]) Xiao Deng, Rutao Yao are with the Department of Nuclear Medicine, State University of New York at Buffalo, Buffalo, NY, 14226, USA. (telephone: 716-838-5889, e-mail:
[email protected]) Ciprian N. Ionita, Daniel R. Bednarek, Stephen Rudin are with the Toshiba Stroke and Vascular Research Center, State University of New York at Buffalo, Buffalo, NY, USA.
MATERIALS AND METHODS
CT
(a)
(b)
Fig. I. (a) Photo, and (b) Diagram illustration of the small animal CT system setup.
3409
The MAF detector can operate in both energy integration (EI) and single photon counting (SPC) mode [3]. The MAF detector, as shown in Fig. 2, consisted of a 300 f!m thick CsI(TI) phosphor whose light is amplified by a light image intensifier (LII) whose output in turn is sensed by a CCD. The field of view (FOV) of the MAF was 3.6 cm x 3.6 cm. The image from the MAF is read out by a lO24 x 1024 array with a pixel size of 35 f!m. The LII was a dual micro-channel plate with 4 cm diameter. The size of the x-ray focal spot used was 0.3 rum. The x-ray source-to-imager distance was 652 rum, and the source-to-object distance was 570 rum. Csl(TI)
A 2-dimensional filtered back projection (FBP) reconstruction was used as a reference method for evaluating the image quality from the OSC approach.
D. Phantom and Mouse Experiments A standard endovascular stent with wire diameter of 0.14 mm, as shown in Fig. 3(a), was imaged by a 180 steps x 20 rotation/step scan protocol. One projection frame was acquired after each rotation step. The x-ray tube parameters used were 60 kVp, 100 rnA, and 9 ms exposure time. 3mm
���ii��
Light Image Intensifier (LlI) ....,;:::======::1...
Fig. 3 Photo of the stent
Fig. 2 Structure scheme ofMAF detector. B.
Self-calibration ofthe AOR Offset.
For accurate system response modeling, geometrical calibration is one important step needed to be performed. Generally the calibration techniques use a specific calibration object and need to scan the object in certain orbits. For this experimental system, we used a self-calibration method [4] to calculate the offset of the axis of rotation (AOR) from the detector center-axis by only using the projections acquired during an object scan. The method uses the assumption that the projections from opposite views are identical except that the sequence is reversed. C.
Reconstruction Algorithm and System Matrix Derivation
Our statistical reconstruction software was developed based on the ordered-subset-convex (OSC) algorithm [5]:
f.1;+1 (j)
=
I lij(Y;Cu;) Y,) (j) (j) f.1; + f.1; �/ S_ n � n k ' L.. ,JY, (f.1s )L.. IIkf.1s ( )
;
-
(1)
where the f.J; (j) is the attenuation coefficient of the j-th voxel in the object calculated at the n-th iteration and s-th subset, Ii) is the intersection length of projection line i though j-th voxel, Y, represents the measured counts in projection line i, and the expected projection counts Y;(f.J;) is given by (2) where d; represents the blank scan counts in projection i. In this work, the system matrix, {Ii)}, was pre-calculated using the conventional ray-driven method [6].
A �20g normal mouse was sacrificed and scanned at 3 radiation exposure levels. The x-ray tube voltage and current were fixed 60 kVp and 100 rnA. The x-ray exposure time was set at 15 ms, 9 ms and 6 ms for the 3 scans. The respective measured radiation exposure to the mouse was 153 f!Rlf, 88 f!Rlf and 60 f!Rlf, with 1 frame-per-second (fps) frame acquisition rate; and they were recorded as Level l, Level 2, and Level 3, respectively. At each x-ray setting, the whole body mouse was scanned in 3 axial volumes with 20.6 mm between two consecutive volumes which assured the X-ray beam could cover the whole body of the mouse with 10.9 rum overlap in sagittal plane between the adjacent volumes. At each axial step position, the mouse was scanned with a rotation protocol of 180 steps x 20 /step. A 2.1 cm thick aluminum plate was attached in front of the x-ray tube to filter the low energy x-ray photons which could be absorbed in animal body for mouse imaging. With the acquired projections, we reconstructed the 3 parts of mouse body respectively, cropped the redundant overlapping parts of each volume, and combined the rest together to a whole mouse image. For each CT scan, dark field and flat field data were acquired under the same system settings and applied for corresponding corrections before reconstruction. III.
RESULTS
A. Stent Study Fig. 4 shows the CT cross-section view of the reconstructed stent image with the statistical iterative software we developed (2 iterations, 18 subsets). The reconstructed image size is 256 x 256 x 256 with a voxel size of 63 x 63 x 63 f!m3. All wires in the cross-section view can be identified clearly.
3410
were artifact free and provided superior noise performance compared with the filtered back-projection (FBP) reconstructed images, and the OSC software reconstruction resulted in high image quality with reduced radiation exposure. OSC reconstruction Fig. 4 The reconstructed cross-section view of the stent.
B.
Mouse study and contrast assessment
The volume rendering display of the reconstructed mouse whole-body image is shown in Fig. 5. The voxel size in image space is 126 x 126 x 126 11m3. The skeletal structure of the mouse can be identified clearly, which shows the good resolution performance.
Fig. 5 The volume rendering display of the reconstructed mouse whole-body image. (OSC reconstruction, 2 iterations, 18 subsets). The x-ray parameters for this acquisition are 60 kVp, 100 rnA, and 15 ms exposure time.
Fig. 6 compares the sagittal view images of the mouse reconstructed with the developed iterative program and the FBP algorithm at the 3 exposure levels. The voxel size in all images is 0.13 x 0.13 x0.13 mm3• For comparing the noise performance, in the sagittal slices showed in Fig. 6 we selected a circular region of interest (ROI) (""'3.1 mm2) over the selected bone area (A region) and another circular region (B region) with the same area outside the skeleton. The contrast to noise ratio (CNR) for the selected structure was calculated by: CNR
JSA -S8 1 a
8
,
(3)
where SA represents the average signal in A, SB represents the average signal in B, and aB represents the standard deviation in the B region. The CNR in selected region under different exposures are shown in Table. I. The results show that the statistical iterative reconstruction provides superior noise performance compared with FBP analytical reconstruction method, and even the image quality with the reduced exposure is much better than the FBP reconstructed results at the same exposure. IV.
Fig. 6 Sagittal views of reconstructed mouse images with 3 different exposure levels. Top: OSC reconstruction, bottom: FBP reconstruction. Left to Right: the exposure times were 15 ms, 9ms, and 6ms, and measured exposures were 153 IlRlf, 88 IlRif and 60 IlRlf, respectively, and the frame rate was Ifps.
REFERENCES
CONCLUSION
We developed a statistical iterative CT image reconstruction software program and assessed its performance at different radiation exposure levels on a newly built high-resolution small animal SPECT/CT system. The stent study showed it achieved high resolution in the reconstructed image. The mouse experiments showed that the images yielded by the statistical reconstruction software
3411
[I]
X Deng, T Dai, C N Ionita, et al. " Development of a Small Animal SPECT and CT Dual Function Imager with a Microcolumnar CsT(TI) and CCD based Detector ", 2012 IEEE Nuclear Science Symposium and Medical Imaging Conference, MI0-58, Anaheim, CA.
[2]
A. Jain, D. R. Bednarek, S. Rudin, "A theoretical and experimental evaluation of the micro-angiographic fluoroscope (MAF): a high resolution region-of-interest x-ray imager", Medical Physics, 38 (7), pp. 4112-4126, July 2011.
[3]
A. Jain, A. Panse, D. R. Bednarek, R. Yao, S. Rudin, "Evaluation of the micro-angiographic fluoroscope as a high resolution, single-photon counting and energy integrating imager for transmission and emission imaging." IEEE Trans. Nucl. Sci., 59 (5), pp. 1927-1933, 2012.
[4]
V Patel, R N Chityala, K R Hoffmann, et al."Self-calibration of a cone beam micro-CT system", Med. Phys., pp. 2686-2689, 2009.
[5]
F J Beekman, C Kamphuis, "Ordered subset reconstruction for x-ray CT", Phys. Med. BioI., vol. 46, pp. 1835-1844, 2001.
[6]
Siddon RL, "Fast caldulation of the exact radiological path for a three dimensional CT array", Med. Phys., 12 (2): 252-255, 1985