High-Resolution Cardiac Rat SPECT: Effect ofDose and Acquisition Time

2008 IEEE Nuclear Science Symposium Conference Record M10-80

High-Resolution Cardiac Rat SPECT:

Effect of Dose and Acquisition Time

Chao Wu, Brendan Vastenhouw, Frans van der Have, Rudi A. Dierckx, and Freek J. Beekman

Abstract-SPECTimaging with low dose or short acquisition time can lead to very noisy reconstructed images. With the U-SPECT-II system we study the effect of low-count scans, and 3D Gaussian blurring to suppress statistical errors. Three different groups of short-time cardiac rat scans were derived from a long-time scan. Images were reconstructed and then smoothed with 3D Gaussian filters with different kernel sizes. We used the standard deviations of voxel values in the reconstructions from each short-time group to indicate the quality of the blurred images. The results show that a good cardiac rat SPECT image can be obtained from a scan in the U-SPECT-II system within very short time (2 minutes) or with very low dose (0.30 mCi activity injected) if some additional blurring is accomplished to the reconstruction.

I. INTRODUCTION

T

HE pre-clinical medical studies benefit much from small animal models. Dedicated radionuclide imaging technology, such as micro-SPECT (single photon emission computed tomography) imaging for rodent models, has provided many advantages for the researches of molecular process.

In order to reduce the radiation dose used in animal studies, or to accomplish dynamic (4D) imaging with short time-frames, a high-sensitivity imaging system together with some image processing methods for suppressing statistical noises in the reconstructions is needed. A common method for reducing the noises is the Gaussian blurring, which can be implemented easily and fast to N-dimensional images due to the linearly separable property of the Gaussian kernel.

In this paper, we study the effect of low-count cardiac rat SPECT scans which are accomplished by using the U-SPECT-II system, and analyze their reconstructions with different extent of Gaussian blurring.

II. MATERIALS&METHODS

A. U-SPECT-II System

Figure IA shows an overview of the U-SPECT-ll device, a small-animal SPECT system introduced in [1]. An interchangeable cylinderic collimator containing 75 pinholes is mounted in the center of a triangular gantry where three ultra-large NaI gamma-detectors are installed. An object such as a mouse or a rat can be moved inside the collimator (Fig. 1B) during scanning with anXYZstage, and the sequence of those movements is created and executed automatically by acquisition software according to region of interest (ROI). A novel graphical user interface (GUI) makes users select the ROI easily with the aid of three optical pictures taken by three webcams (Fig. 1C).

Fig.I. (A) U-SPECT-II system. (B) Mouse in a tungsten collimator with 75 gold pinholes. (C) GUI of acquisition software for ROI selection.

1.2

Fig. 2. Capillary phantom (Tc-99m) reconstructions. (A) 0.35-mm pinhole mouse collimator. (B) 1.0-mm pinhole rat collimator.

B

o~

:-1.5mm

0.8~

0111

This focusing multi-pinhole system can reach an ultra-high resolution, which is smaller than 0.35 mm for mouse-sized objects, or less than 0.8 mm for rat-sized objects (Fig. 2). Its peak geometric sensitivity is from 0.07% to 0.18% for different collimators [1]. A O~25 .O.6mm

-:.

...:

..

0.3,..-:

-Si!

0.5

o~

1iJi:4

Manuscript received November 14, 2008.

C. Wu is with the University Medical Center Groningen, Groningen, The Netherlands, and the University Medical Center Utrecht, Utrecht, The Netherlands (e-mail: c.wu@umcutrecht.nl).

B. Vastenhouw is with the University Medical Center Utrecht, Utrecht, The Netherlands, and the Molecular Imaging Laboratories, Utrecht, The Netherlands (e-mail: b.vastenhouw-2@umcutrecht.nl).

F. van der Have is with the University Medical Center Utrecht, Utrecht, The Netherlands, and the Molecular Imaging Laboratories, Utrecht, The Netherlands (e-mail: f.vanderhave@umcutrecht.nl).

R. A. Dierckx is with the University Medical Center Groningen, Groningen, The Netherlands (e-mail: r.a.dierckx@ngmb.umcg.nl).

F.1. Beekman is with the Delft University of Technology, Delft, The Netherlands, the University Medical Center Utrecht, Utrecht, The Netherlands, and the Molecular Imaging Laboratories, Utrecht, The Netherlands (e-mail: f.beekman@umcutrecht.nl).

Fig. 5. SA slices from reconstructed cardiac rat images acquired by using V-SPEeT-II. Each row from left to right: acquisition for 30 min, 450 s, 225 s, and 112.5 s; each column from top to bottom: FWHM of Gaussian blurring ranging from 0.22 to 2.2 mm. 2.5 112.5 2 225 0.5 I I I I - - - -1- - - --t - - - t- - - 1 -I 1 1 1 1 1 I I 2.21 1.32 1.55 1.77 1.10 1.99 0.88 0.66 0.44

Filter".. Scan time (s)

FWHM ".. 1800 450 (mm) 0.22 0.16r---r----.---~---;:::=:c:::====:::=::::;] 0.14 0.12 o en 0.1 '"0 Q) ~ 0.08 ro E

o

0.06

z

0.04 0.02 1 1.5 Filter FWHM (mm)

Fig. 4. Nonnalized SO of each low-count group. Short-time group has a large SO, and 3D Gaussian blurring can reduce it significantly.

III. RESULTS

The SO of each low-count group is obtained as a function of the FWHM of30Gaussian filters, and then normalized by the average number of counts of the sub-scans in that group. Fig. 4 shows that both more counts and more blurring can lead to a small SO, e.g., the SO of the 225 s group with 0.44 mm FWHM blurring is similar to the one of the 112.5 s group blurred with 0.88 mm FWHM.

B. Data Acquisition&Image Reconstruction

A rat was injected with 175 MBq (4.72 mCi) Tc-99m tetrofosmin in 0.4 ml. A 30 min high-count scan was accomplished with the U-SPECT-II system at 140 minutes after the injection, and the data were recorded in list-mode. Afterwards, three groups of sub-scans were derived from the data, including four 450 s sub-scans, eight 225 s sub-scans, and sixteen 112.5 s sub-scans.

We reconstructed all images using ordered subset expectation maximization (OS-EM) algorithm[2] (16subsets and 6 iterations) with voxel size equal to (0.1875 mm)3. The reconstructions were subsequently smoothed by employing 30 Gaussian blurring with full width at half maximum (FWHM) ranging from 0.22 mm to 2.2 mm.

C. Data Analysis

The annular part of the left ventricle (LV) wall in each image was divided into eight sectors with central angles of 45 degrees (Fig.3). For each of the three groups of sub-scans, we calculated the standard deviation (SO) of the number of counts in each sector, and defined the average of those SOs as the SO of that group, which was an indication of image statistical error.

Fig. 3. Segmentation of the annular LV wall.

For example, there were sixteen sub-scans in the 112.5 s group. Therefore, the SO of this group was represented as:

SD =!iSD

s

=!i

-1-i(c!')-C('l1,

(1)

8s=l 8s=l 16 -1n=l

in which, C~s) was the number of counts in sectors of the reconstruction from sub-scan nin that group, and C(s) was

the average number of counts in sectorss of all the images in the group.

We select the first sub-scan in each low-count group as a representative sample. Fig. 5 shows short-axis (SA) slices of the filtered cardiac rat images reconstructed from the three low-count samples. Likewise, the blurred slices of the high-count scan are also presented in the figure as references. The statistical errors in the low-count cardiac images have been suppressed by blurring with large FWHM. For instance, the visualization of LV wall in Slice A in Fig. 5 is clear and similar to the reference image in Slice B which is blurred with a smaller Gaussian kernel. Note that the Slice A is derived from a sub-scan within 112.5 seconds, or equivalently (according to the property of Poisson process) a 30 min scan with an injected activity equal to only 11 MBq (0.30 mCi).

IV. CONCLUTIONS

With the help of some additional blurring, a good cardiac rat SPECT image can be obtained from a scan within only 2 minutes or with very low dose such as 0.30 mCi activity injected. The loss of some image details is inevitable during blurring, but the information of the LV wall can still be extracted from the blurred image and represented with better visualization than the unfiltered noisy-image. As a result, the reduction of dose or scan time for animal studies is possible by using the U-SPECT-II system.

REFERENCES

[I] F. van der Have, B. Vastenhouw, R. M. Ramakers, W. Branderhorst,1.

O. Krah, C. Ji,et al.,"U-SPECT-II: an ultra-high-resolution device for molecular small-animal imaging," J. Nucl. Med, submitted for publication.

[2] H. M. Hudson, and R. S. Larkin, "Accelerated image reconstruction using ordered subsets of projection data,"IEEE Trans. Med Imag.,vol. 13, no. 4, pp. 601-609, 1994.