This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 138.25.78.25

This content was downloaded on 07/12/2013 at 03:59

Please note that terms and conditions apply.

A feasibility study of PETiPIX: an ultra high resolution small animal PET scanner

View the table of contents for this issue, or go to the journal homepage for more

2013 JINST 8 P12004

(http://iopscience.iop.org/1748-0221/8/12/P12004)

Home Search Collections Journals About Contact us My IOPscience

2013 JINST 8 P12004

PUBLISHED BY IOP PUBLISHING FOR SISSA MEDIALAB

RECEIVED: August 23, 2013REVISED: November 16, 2013

ACCEPTED: November 20, 2013PUBLISHED: December 6, 2013

A feasibility study of PETiPIX: an ultra highresolution small animal PET scanner

K. Li,a,1 M. Safavi-Naeini,a D.R. Franklin,b M. Petasecca,a S. Guatelli,a

A.B. Rosenfeld,a B.F. Huttona,c and M.L.F. Lercha

aCentre for Medical Radiation Physics, University of Wollongong,Northfields Ave, Wollongong, Australia

bFaculty of Engineering and IT, University of Technology Sydney,Broadway, Sydney, Australia

cInstitute of Nuclear Medicine, University College London,Gower Street, London, England

E-mail: kl780@uowmail.edu.au

ABSTRACT: PETiPIX is an ultra high spatial resolution positron emission tomography (PET) scan-ner designed for imaging mice brains. Four Timepix pixellated silicon detector modules are placedin an edge-on configuration to form a scanner with a field of view (FoV) 15 mm in diameter. Eachdetector module consists of 256×256 pixels with dimensions of 55×55×300 µm3. Monte Carlosimulations using GEANT4 Application for Tomographic Emission (GATE) were performed toevaluate the feasibility of the PETiPIX design, including estimation of system sensitivity, angulardependence, spatial resolution (point source, hot and cold phantom studies) and evaluation of po-tential detector shield designs. Initial experimental work also established that scattered photons andrecoil electrons could be detected using a single edge-on Timepix detector with a positron source.Simulation results estimate a spatial resolution of 0.26 mm full width at half maximum (FWHM)at the centre of FoV and 0.29 mm FWHM overall spatial resolution with sensitivity of 0.01%, andindicate that a 1.5 mm thick tungsten shield parallel to the detectors will absorb the majority ofnon-coplanar annihilation photons, significantly reducing the rates of randoms. Results from thesimulated phantom studies demonstrate that PETiPIX is a promising design for studies demandinghigh resolution images of mice brains.

KEYWORDS: Solid state detectors; Gamma camera, SPECT, PET PET/CT, coronary CTangiography (CTA); Image reconstruction in medical imaging

1Corresponding author.

c© 2013 IOP Publishing Ltd and Sissa Medialab srl doi:10.1088/1748-0221/8/12/P12004

2013 JINST 8 P12004

Contents

1 Introduction 1

2 Materials and methods 32.1 Monte Carlo model of PETiPIX 4

2.1.1 System sensitivity 52.1.2 Angular dependence 52.1.3 Spatial resolution 52.1.4 Phantom study 52.1.5 Shielding design 6

2.2 Experimental detection of the scattered photons and the recoil electrons 7

3 Results 73.1 Monte Carlo simulation results 7

3.1.1 System sensitivity 73.1.2 System angular dependence 73.1.3 Spatial resolution 83.1.4 Phantom study 93.1.5 Shielding design 9

3.2 Experimental detection of the scattered photons and the recoil electrons 10

4 Discussion 11

5 Conclusions 13

1 Introduction

Positron emission tomography (PET) is a molecular imaging technique often used for performingin vivo and non-invasive preclinical studies in small animal models. Genetically modified mice areoften used for their genetic similarity with humans, rapid reproductive rate, short life span and easeof breeding [1]. The need to visualise the complex metabolic and physiological processes withinsmall organs or functional regions of laboratory mice demands ever-higher spatial resolution indedicated small animal PET scanners.

Conventional small animal PET scanners are based on scintillation detectors, in which scin-tillator materials such as LSO, LYSO or BGO are optically coupled to position-sensitive photo-multiplier tubes or silicon photomultipliers. These scanners are capable of providing a spatialresolution of 1 to 2 mm FWHM [2–11]. However, achieving sub-millimetre spatial resolutionusing scintillation detectors is a challenge due to the variation of scintillation light collection effi-ciency, intra-crystal Compton or coherent scattering, inter-crystal Compton scattering, and limiteddepth-of-interaction (DoI) information [12, 13]. Simulations suggest that high intrinsic resolution

– 1 –

2013 JINST 8 P12004

(sub-millimetre) can be reached with scintillators if the pixel size is sufficiently small and indivi-dual pixels can be identified unambiguously; however, in practice this results in great complexityand high production costs [14]. A new design for a three dimensional position-sensitive photondetector based on a stack of several two dimensional scintillation detector layers (consisting ofarrays of LYSO crystals) coupled to thin position-sensitive avalanche photodiodes (PSAPD) wasproposed and investigated by Gu et al. [15]. Since the sequence of multiple interactions of photonscan be identified with sufficiently high timing and energy resolution, this detector design has beenshown to achieve submillimetre spatial resolution when coupled to submillimetre crystal elements,making it ideally suited to small animal PET systems [12].

An alternative approach is to eliminate the scintillator altogether, trading lower photon de-tection efficiency for high resolution by exploiting the capabilities of modern CMOS fabricationtechniques. Over 99% of 511 keV annihilation photon interactions with silicon are Compton scat-tered events. Compton scattered photons have a high probability of escaping from the detector;Giovanni et al. have shown that even for a 40 mm thick silicon detector, over 51.8% of singleCompton scattered photons escape undetected [16]. Therefore, the primary detection mechanismis by collection of the recoil electron-hole pair from a Compton event, with some additional low en-ergy gamma photons, scattered within the object and directly detected via the photoelectric effect.Since the energy required to generate one electron-hole pair in silicon is only a few electron-volts,it can provide an excellent energy resolution especially at lower energies (less than 50 keV). Whileits low electron density does not allow the detection of high energy photons, it can be utilised to de-tect and track the recoil electrons which deposit their kinetic energy along their track. Additionally,the charge carriers — especially holes — travel much faster in silicon compared to most alterna-tive semiconductors used for radiation detection (such as CZT), resulting in relatively good timingresolution. Modern CMOS fabrication techniques enable the production of silicon detectors withextremely small pixels, offering excellent intrinsic spatial resolution compared to that which ispossible with scintillation crystals.

In the last decade, several studies [13, 16–22] have demonstrated improved spatial resolu-tion of small animal PET systems through the use of silicon detectors. Park et al. compared theperformance of scintillator-based and non-scintillator PET systems through both simulation andexperimental work [13, 18, 19]. Both simulations and experiments were based on inserting a ringof non-scintillation silicon detectors (with an inner diameter of 4 cm, an axial length of 4 cm andthickness of 1.6 cm) into a conventional scintillator-based small animal PET scanner (with an in-ner diameter of 17.6 cm, an axial length of 16 cm and thickness of 2 cm). This scintillator scanneris based on a single layer of 3×3×20 mm3 BGO crystals. The simulated non-scintillation detec-tors consisted of 16 layers of 300×300×1000 µm3 silicon pads. The implemented experimentalnon-scintillation apparatus is comprised of 16 layers of 1.4×1.4×1 mm3 silicon pads. Spatial res-olution was compared for silicon-silicon (i.e. direct-direct), silicon-BGO (i.e. direct-scintillator)and BGO-BGO (i.e. scintillator-scintillator) coincidences; it was found that using silicon-siliconcoincidences alone provided the best spatial resolution. A spatial resolution of 340-360 µm wasobtained in simulation using silicon detectors with a 18F point source [13, 18]. The experimentalprototype small animal PET scanner achieved a 700 µm FWHM intrinsic detector resolution and980 µm reconstructed resolution [19]. The spatial resolution differs from the previous simulationsbecause of the larger detector pad size, which is the main factor preventing this prototype fromreaching the theoretical limit of resolution imposed by positron range and non-collinearity.

– 2 –

2013 JINST 8 P12004

Park et al. have also demonstrated that the addition of direct-detection silicon detectors issufficient to convert a low resolution PET scanner to a high resolution (1 mm FWHM) small animalPET scanner by forming coincidence pairs from both inner silicon detector and external scintillatorcrystals [13, 20, 21]. In addition to the aforementioned “inserted” design, a small animal PETscanner can also be constructed using direct-detection silicon detectors without any scintillationat all. SiliPET, a prototype small animal PET scanner, is based on four stacks of silicon stripdetectors arranged to form a 5×5×6 cm3 box with a detector pitch of 500 µm [16, 22]. MonteCarlo simulations have predicted this design to be capable of providing a spatial resolution of0.52 mm FWHM at the centre of system.

The PETiPIX small animal PET scanner presented in this paper has been designed to achieveultra high spatial resolution, limited only by positron range and 511 keV photon non-collinearity.It utilises four state-of-the-art Timepix pixellated silicon detectors with ultra small pixel dimen-sions (55×55×300 µm3) in an edge-on configuration. By recording a large number of singleCompton scattered events, a two-dimensional (2D) tomographic image can be reconstructed ac-curately. A feasibility study was conducted by building an accurate PETiPIX model in theGEANT4 Application for Tomographic Emission (GATE) platform. Evaluations of system sen-sitivity, angular dependence, spatial resolution, hot phantom and cold phantom studies show anexcellent capability of PETiPIX to perform PET scans for mouse brains.

The remainder of the paper is organised as follows: section 2 presents the materials andmethods to assess the feasibility of PETiPIX. Section 3 presents the results obtained from thefeasibility tests. A discussion of results and future plans are presented in sections 4 and 5.

2 Materials and methods

The proposed PETiPIX scanner is shown in figure 1. It consists of four Timepix detectors placedin an edge-on configuration mounted on a rotating gantry. Its FoV is 15 mm in diameter, suitablefor conducting PET scans of mouse brains. The Timepix module used in this study is a pixellatedsilicon detector on a 300 µm silicon substrate. It consists of 256×256 pixels, each with an activearea of 55×55 µm2, which can be programmed independently into photon counting mode (Medipixmode), energy measuring mode (time-over-threshold (TOT) mode) or arrival time measurementmode (Timepix mode) [23]. All Timepix modules are synchronised by a common clock signal andtheir data are recorded in list mode format (LMF). Each gamma interaction with energy exceedinga previously set low energy threshold (THL) will trigger an event-log operation. In the LMF file,each logged event contains the Timepix module number, pixel ID and time-stamp for this event.Coincidence pairs are found by post processing all LMF data sets from all four Timepix modulesand used to reconstruct the final image. Using the associated Pixelman software with coincidence-detection plugin, a timing resolution of 10 ns is achievable.

Silicon’s low photon absorption for 511 keV photons, combined with the small dimensionsof Timepix detector module results in single Compton scattering being the most common form ofinteraction between the incident photons and the detector. This type of event can be localised tohave occurred within the expected Compton recoil electron range, typically from several to a fewhundred µm (depending on the energy of the incident photon). Compared to the positron range,which has a mean of approximately 200 µm for a 18F source in tissue, the error caused by the

– 3 –

2013 JINST 8 P12004

14 mm1

4 m

m

TimepixBoardReadout

Data Logger, Clock Source

and Gantry Controller

15 mm

Figure 1. Conceptual drawing of PETiPIX.

Compton recoil electron migration to adjacent pixels is negligible. The fine detector pixellationcombined with its edge on orientation provides extremely precise DoI information. Therefore,spatial resolution is only limited by the positron range and annihilation photon non-collinearity,with the latter of very little significance due to the small scanner diameter.

2.1 Monte Carlo model of PETiPIX

An accurate model of the PETiPIX scanner was simulated in GEANT4 Application for Tomo-graphic Emission (GATE) version 6.1 (with GEANT4 9.4p02) [24]. To simulate PETiPIX, a setof four pixellated silicon slabs was accurately modelled in GATE. Each silicon slab is 300 µmthick and consists of 256×256 pixels, with each pixel measuring 55×55 µm2. The physics pro-cesses, including gamma interactions (Photoelectric effect, Compton scattering and Rayleigh scat-tering), electron ionisation, multiple scattering and bremsstrahlung were modelled using the Stan-dard GEANT4 electromagnetic physics Package, which has previously been shown to be in goodagreement with the NIST reference database [25]. In addition, excellent agreement between GATEsimulations and experimental results has been obtained through a number of studies [26, 27]. Asdiscussed above, a recoil electron from a single Compton scattering event may pass through mul-tiple pixels along its path so that multiple signal pulses are generated from one detector modulewithin a given timing window. All of these pulses are recorded with their energy, position (pixeladdress) and timing information; however, only the event with the highest energy is selected toform one half of a coincidence pair in the image reconstruction process. The energy window usedduring offline coincidence detection ranges from 10 keV to 650 keV, the timing window is set to10 ns and a constant energy resolution of 1.2 keV for all energies is used throughout the simulationstudies. Positron range as well as non-collinearity are taken into account in all system simulations.

– 4 –

2013 JINST 8 P12004

2.1.1 System sensitivity

A 37 MBq 18F point source with a diameter of 10 µm was simulated inside a spherical water phan-tom with a diameter of 12 mm, placed at the centre of the scanner’s FoV. The simulation runs for 5seconds of simulated acquisition time. Absolute system sensitivity was calculated as a ratio of thenumber of true coincidences detected per second to the 18F point source activity.

2.1.2 Angular dependence

To avoid geometric distortions, data acquired by the scanner must provide full coverage of the ob-ject (a full 180◦ arc) [28]. The angular coverage of PETiPIX was evaluated by placing a 37 MBq 18Fpoint source at a radial displacement of 3 mm from the centre of FoV in a 12 mm diameter waterphantom. The first simulation was performed with the scanner set at its initial position for 2 secondsof simulated acquisition time. The second simulation was then repeated with exactly the same con-ditions as the first, but with the scanner rotated by 45 degrees. 2D sinograms for each simulationas well as their summation were constructed to evaluate the angular coverage of PETiPIX scanner.

2.1.3 Spatial resolution

Spatial resolution of the proposed scanner was assessed by placing five uniformly spaced37 MBq 18F point sources with 10 µm diameters inside a cylindrical water phantom with a di-ameter of 13 mm. Three sets of simulations were then performed in which the point sources areplaced 1 mm, 0.5 mm, and 0.3 mm apart along the radial direction within the FoV. Each simulationruns for 20 seconds of simulated acquisition time for two scanner positions (initial position of zerodegrees and rotated by 45 degrees).

To demonstrate the capability of resolving the DoI, uniformity of the spatial resolution acrossthe entire FoV was also tested. A 37 MBq 18F point source with a diameter of 10 µm was placedin a water phantom with a diameter of 13 mm. This point source was simulated in six differentpositions, stepped along the radial direction in 1 mm increments from the centre of FoV to 6 mmfrom the centre. Each point source position was simulated for 5 seconds of simulated acquisitiontime with two scanner positions (initial position and rotated by 45 degrees).

Images were reconstructed using the data acquired from these simulations via the two-dimensional filtered backprojection (2D-FBP) method. As the point source profile was expectedto be a cusp-like shape due to the blurring from non-zero positron range, spatial resolution ofPETiPIX was directly measured by the FWHM of the point source profile, without any Gaussiancurve fitting.

2.1.4 Phantom study

To assess the overall image quality of the proposed scanner, two accurately modelled Ultra MicroJaszczak phantoms (hot and cold phantoms) were simulated separately.

The hot Jaszczak phantom was filled with a total activity of 3.5 MBq 18F. It consists of a hollowcylinder, 12 mm in diameter with a length of 0.3 mm, filled with water, accommodating six sectorsof hot rods, 0.3 mm in length with diameters of 0.96 mm, 0.8 mm, 0.68 mm, 0.54 mm, 0.4 mm and0.3 mm. The centre-to-centre spacing within each sector is twice the rod diameter.

The cold Jaszczak phantom consists of a hollow cylinder, 10 mm in diameter with a length of0.3 mm, filled with a total activity of 37 MBq 18F, housing six sectors of cold rods, with dimensions

– 5 –

2013 JINST 8 P12004

Cylindrical water phantom

F-18 Line source

Tungsten

Timepix

Tungsten

t

h

Figure 2. Proposed tungsten shield design, showing simulation configuration.

slightly different to the previously described hot phantom: 0.84 mm, 0.70 mm, 0.59 mm, 0.47 mm,0.35 mm and 0.26 mm.

Phantom data from both phantom simulations were acquired in two scanner positions (initialand 45 degree rotated positions). The hot and cold phantoms were simulated for a total of 1000 and800 seconds for each scanner position, respectively. Images were reconstructed by 2D-FBP as wellas maximum-likelihood expectation maximisation (MLEM) method. To demonstrate the influencebrought by the angular coverage of the data set, images with a single scanner position were alsoreconstructed.

2.1.5 Shielding design

For a source distribution extending in the axial direction or surrounded by a scattering medium ex-tending in this direction, a significant rate of random coincidences will be observed by a transaxialdetector array unless shielding is used to absorb non-coplanar photons.

The effects of both phantom scattering and random coincidences on the spatial resolution mustbe therefore carefully considered for the PETiPIX design performance. This is especially importantfor 2D tomographic scanning of an object that extends beyond the width of the scanner field of viewin the axial direction. To study the above effects, a simulation study was conducted in which a linesource with a total activity of 37 MBq 18F and a diameter of 10 µm and length of 16 mm was placedalong the central axis of a 20 mm long cylindrical water phantom with a diameter of 13 mm.

Four sets of simulations were performed using this source/phantom configuration: three inwhich two 14×14×1.5 mm3 tungsten shields were placed 0.5 mm, 1 mm and 2 mm away from bothsurfaces of each Timepix detector, and one control simulation with no tungsten shield present. Eachsimulation was executed for 20 seconds of simulated acquisition time for two scanner positions(angle of gantry rotation θ = 0◦ and θ = 45◦).

The proposed shield design and simulation configuration is illustrated in figure 2. h denotesthe separation of the shield from the detector while the thickness of the shield plates is denoted t(1.5 mm in this simulation).

– 6 –

2013 JINST 8 P12004

Figure 3. Sensitivity changes with low energy threshould.

The acquired data from each simulation was used to construct a sinogram. A sum of theprojections from all angles was calculated to produce a line source profile, in which the eventsoutside the central peak consist mostly of the phantom scattered and random events.

2.2 Experimental detection of the scattered photons and the recoil electrons

To establish the detector’s ability to localise and detect scattered photons and recoil electrons,a 7 MBq 68Ga ( 68Ge generator) positron-emitting point source encased in a 1 mm thick steelcase was placed 15 mm from a single edge-on Timepix detector, such that the resulting 511 keVannihilation photons were incident on the detector’s edge.

The detector was reverse-biased at 100 V at room temperature and coupled to a USB dataacquisition module (FitPIX). The total acquisition time was 5 minutes, with a frame duration of1 second.

3 Results

3.1 Monte Carlo simulation results

3.1.1 System sensitivity

The true coincidence count rate for a 37 MBq 18F point source is 3541 coincidences per second. Theabsolute system sensitivity is calculated as 0.0096% with an energy window of 10 keV–650 keV.The system sensitivity as a function of the low energy threshold is plotted in figure 3. As the lowenergy threshold is increased (narrowing the energy acceptance window), the system sensitivity isdecreased.

3.1.2 System angular dependence

Three sinograms are shown in figure 4. Figure 4(a) is the sinogram generated when the scanner isplaced in its initial position, covering only limited angles corresponding to approximately half ofthe 180◦ arc. The sinogram shown in figure 4(b) is generated when the scanner is rotated by 45◦.The combined sinogram (figure 4(c)) contains a data set with complete angular coverage.

– 7 –

2013 JINST 8 P12004

(a) Initial Position. (b) Rotated Position. (c) Combined Sinogram.

Figure 4. Sinograms of a point source acquired with two different scanner positions, individually andcombined. The dark part along the individual sinogram curves is due to absent projection angles. Thecumulative sinogram contains all angular information.

(a) 1 mm. (b) 0.5 mm. (c) 0.3 mm.

−3 −2 −1 0 1 2 3−0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Radial Displacement (mm)

inte

nsity

Pro

file

(d) 1 mm profile.

−2 −1.5 −1 −0.5 0 0.5 1 1.5 2−0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Radial Displacement (mm)

inte

nsity

Pro

file

(e) 0.5 mm profile.

−2 −1.5 −1 −0.5 0 0.5 1 1.5 2−0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Radial Displacement (mm)

Intensity

Profile

(f) 0.3 mm profile.

Figure 5. Reconstructed point source images and radial profiles are plotted for five point sources placed1 mm, 0.5 mm and 0.3 mm apart.

3.1.3 Spatial resolution

The reconstructed images and profiles of five uniformly spaced point sources with 1 mm, 0.5 mmand 0.3 mm separation are shown in figure 5. The radial cross-sections and reconstructed imagesclearly demonstrate that the system is capable of resolving all point sources.

– 8 –

2013 JINST 8 P12004

Figure 6. Overall spatial resolution (FWHM) versus the radial source position.

The FWHM of the point source placed at the centre of FoV is measured as 0.26 mm in bothradial and tangential directions and is shown figure 6. As the radial offset increases, the spatialresolution only deteriorates slightly, increasing from 0.26 mm to 0.3 mm in the radial direction and0.26 mm to 0.32 mm in the tangential direction. The overall spatial resolution of proposed PETiPIXwas calculated to be 0.29 mm FWHM.

3.1.4 Phantom study

Reconstructed images of Ultra Micro Jaszczk phantoms (hot phantom and cold phantom) by FBPand MLEM algorithm for single (initial position) and combined scanner positions are shown infigure 7 and figure 8. The FBP reconstructed image of the phantom acquired with a single scan-ner position (figure 7(a), 8(a)) is geometrically distorted. These artefacts are largely eliminatedwhen using both data sets corresponding to the two scanner positions. As seen in figure 7(b)and 8(b), the smallest diameter rods (0.3 mm) and the second smallest diameter rods (0.35 mm)are resolved unambiguously. Applying the MLEM algorithm partially eliminates the geometricdistortion seen in 7(a) and 8(a) , with the resulting image shown in figure 7(c) and 8(c). The bestimage was obtained by applying the MLEM to the data set with two scanner positions (figure 7(d)and figure 8(d)).

3.1.5 Shielding design

The line source profiles acquired using three different shield-to-Timepix-surface distances (h = 0.5,1 and 2 mm) are shown in figure 9. In addition, the accompanying profile with no shielding is alsoplotted for comparison. As phantom scattering events and random events both contributed to thetail sections of these line source profiles, it can be seen that when shields were placed 0.5 mmand 1 mm away from the detector surfaces, the total number of random and scattered events werereduced by 75% and 50% respectively, compared to the profile seen in the absence of shielding.However, when the tungsten shields were 2 mm away, the effect of shielding is negligible.

– 9 –

2013 JINST 8 P12004

(a) FBP, single scanner position. (b) FBP, two scanner positions.

(c) MLEM, single scanner position. (d) MLEM, two scanner positions.

Figure 7. Reconstructed images of Ultra Micro Jaszczk hot phantom by FBP and MLEM algorithm forsingle and two scanner positions.

3.2 Experimental detection of the scattered photons and the recoil electrons

Preliminary experimental data recording the low energy single scattered photons and tracks of theCompton recoil electrons by the Timepix detector were analysed; three one-second acquisitionsare shown in figure 10. The frames clearly show that the recoil electrons from single Comptonscattered events are detected within the space of a few pixels, where the average number of pixelscrossed by the recoil electron is approximately 5.

– 10 –

2013 JINST 8 P12004

(a) FBP, single scanner position. (b) FBP, two scanner positions.

(c) MLEM, single scanner position. (d) MLEM, two scanner positions.

Figure 8. Reconstructed images of Ultra Micro Jaszczk cold phantom by FBP and MLEM algorithm forsingle and two scanner positions.

4 Discussion

The Monte Carlo simulations of the proposed PETiPIX scanner geometry demonstrate that it pro-vides a spatial resolution of 0.26 mm FWHM at the centre of the FoV and an overall spatial resolu-tion of 0.29 mm FWHM. As the vast majority (over 95%) of the recorded coincidence pairs wereformed from single Compton scattering events, the lines of response contain accurate informationregarding the location of annihilation events. All simulations were configured to take positronrange and the non-collinearity effect into account and the results demonstrate that a realisation ofPETiPIX would be suitable for performing pre-clinical mouse brain PET studies.

While the sensitivity of the proposed PETiPIX scanner configuration is comparable to that ofa high resolution SPECT system (0.01%), it is relatively low when compared to conventional small

– 11 –

2013 JINST 8 P12004

Figure 9. Radial profiles of the line source plotted for different shield to Timepix distances.

50 100 150 200 250

50

100

150

200

250

(a) Frame 1.

50 100 150 200 250

50

100

150

200

250

(b) Frame 2.

50 100 150 200 250

50

100

150

200

250

(c) Frame 3.

Figure 10. Three 1 s frames acquired by Timepix when placed in edge-on mode 15 mm from a 7 MBq68Ga (68Ge) point source encased in a 1 mm thick steel, showing the tracks of the Compton recoil electrons.Energies ranging from 10 keV to 313.8 keV are observed.

animal PET scanners. This is due to its limited solid angle coverage and low photon absorption ofsilicon at 511 keV. There are a number of methods by which the sensitivity can be increased. Byadding four detector modules to fill the corners of the PETiPIX scanner, the system sensitivity isnearly doubled. A further improvement can be achieved by increasing the thickness of the detector

– 12 –

2013 JINST 8 P12004

substrate to 500 µm and therefore improving its angular coverage. An alternative approach is tofabricate the detector using a higher Z semiconductor material such as CZT, which provides ahigher stopping power compared to silicon.

However, the images resulting from hot phantom simulations presented in this paper demon-strate that PETiPIX can produce high resolution images in approximately 40 minutes (20 minutesfor each scanner position) of scan time for realistic pre-clinical activity levels.

The angular coverage of the data set acquired by PETiPIX will influence the reconstructedimage. Geometric distortions caused by insufficient angular coverage can be eliminated either byrotating the gantry by 45 degrees to perform a second scan, and by applying an iterative imagereconstruction technique, such as the MLEM algorithm as demonstrated in section 3.

In the case of PETiPIX performing 2D tomographic scans for an object extending in the axialdirection, an increasing number of random coincidences can be expected since many non-coplanarannihilation photons will enter the 14×14 mm2 surface of the Timepix detector. The overall spatialresolution and image contrast can therefore be degraded. Simulation results show that by placing1.5 mm thick tungsten shields less than 1 mm away from both detector surfaces (analogous to plac-ing the Timepix detector inside a rectangular tungsten box with only its front edge open), most ofthe non-coplanar annihilation photons can be prevented from reaching the detector, reducing therate of random and non-coplanar coincidences by up to 75%.

Preliminary experimental results using one Timepix detector demonstrate its ability to detectsingle Compton scattering events. Charges with a range of energies from 10.68 keV to 313.80 keVwere detected by one or multiple adjacent pixels, with an average path length of 5 pixels (figure 10).This provides an accurate localisation of the incident photons as well as a means to verify and val-idate the simulation model. Simulations conducted under equivalent conditions resulted in trackscovering approximately 3 pixels. This discrepancy is due to the fact that charge sharing betweenadjacent pixels has not yet been modelled in the GATE simulations; however since PETiPIX willonly consider the highest-energy pixel in computing the line of response, the system performancewill not be affected.

5 Conclusions

This paper presents the initial design and feasibility studies of the proposed PETiPIX small animalPET scanner based on Timepix silicon detector modules placed in coincidence. Monte Carlo simu-lations of the proposed scanner were performed using GATE to evaluate its performance. A spatialresolution of 0.26 mm FWHM at the centre of FoV and 0.29 mm FWHM overall were predicted bythe simulations, and system sensitivity was estimated to be 0.01%.

The high spatial resolution of PETiPIX provides opportunities for pre-clinical small animalbrain research with exceptional levels of detail. Further proposed improvements to the PETiPIXdesign include extending the geometry to increase the FoV by employing more Timepix detec-tor modules, and 3D PETiPIX by placing Timepix modules stacked in a face-on configuration.Additionally, higher Z substrates are currently under investigation and will be addressed in futurepublications. A prototype PETiPIX system with two Timepix modules synchronised with a centralclock in coincidence is currently under development and preliminary results will be reported infuture publications.

– 13 –

2013 JINST 8 P12004

Acknowledgments

The authors wish to thank CSIRO for their financial support.

References

[1] E. Marshall, Genome sequencing. Celera assembles mouse genome; public labs plan new strategy,Science 292 (2001) 822.

[2] S.R. Cherry et al., MicroPET: A High Resolution PET Scanner for Imaging Small Animals, IEEETrans. Nucl. Sci 44 (1997) 1161.

[3] Y.-C. Tai et al., MicroPET II: design, development and initial performance of an improved microPETscanner for small-animal imaging, Phys. Med. Biol. 48 (2003) 1519.

[4] D.P. McElroy, W. Pimpl, B.J. Pichler, M. Rafecas, T. Schuler and S.I. Ziegler, Characterization andReadout of MADPET-II Detector Modules: Validation of a Unique Design Concept for HighResolution Small Animal PET, IEEE Trans. Nucl. Sci 52 (2005) 199.

[5] A. Del Guerra et al., Performance Evaluation of the Fully Engineered YAP-(S)PET Scanner for SmallAnimal Imaging, IEEE Trans. Nucl. Sci 53 (2006) 1078.

[6] Y. Wang, J. Seidel, B.M.W. Tsui, J.J. Vaquero and M.G. Pomper, Performance Evaluation of the GEHealthcare eXplore VISTA Dual-Ring Small-Animal PET Scanner, J. Nucl. Med. 47 (2006) 1891.

[7] P. Sempere Roldan et al., Raytest ClearPET, a new generation small animal PET scanner, Nucl.Instrum. Meth. A 571 (2007) 498.

[8] J.S. Kim et al., Performance Measurement of the microPET Focus 120 Scanner, J. Nucl. Med. 48(2007) 1527.

[9] E.P. Visser et al., Spatial Resolution and Sensitivity of the Inveon Small-Animal PET Scanner, J. Nucl.Med. 50 (2009) 139.

[10] J. Kataoka et al., Development of an APD-Based PET Module and Preliminary ResolutionPerformance of an Experimental Prototype Gantry, IEEE Trans. Nucl. Sci 57 (2010) 2448.

[11] M. Safavi-Naeini et al., Preclinical Studies Using a Prototype High-Resolution PET System withDepth of Interaction, IEEE Nucl. Sci. Symp. Med. Imaging. Conf. (2011) 3305.

[12] C.S. Levin, Promising New Photon Detection Concepts for High-Resolution Clinical and PreclinicalPET, J. Nucl. Med. 53 (2012) 167.

[13] S.-J. Park, W.L. Rogers and N.H. Clinthorne, Design of a very high-resolution small animal PETscanner using a silicon scatter detector insert, Phys. Med. Biol. 52 (2007) 4653.

[14] J.R. Stickel and S.R. Cherry, High-resolution PET detector design: modelling components of intrinsicspatial resolution, Phys. Med. Biol. 50 (2004) 179.

[15] Y. Gu, G. Pratx, F.W.Y. Lau and C.S. Levin, Effects of multiple-interaction photon events in a highresolution PET system that uses 3-D positioning detectors, Med. Phys. 37 (2010) 5494.

[16] G. Di Domenico et al., SiliPET: An ultra-high resolution design of a small animal PET scanner basedon stacks of double-sided silicon strip detector, Nucl. Instrum. Meth. A 571 (2007) 22.

[17] J. Kang et al., A small animal PET based on GAPDs and charge signal transmission approach forhybrid PET-MR imaging, 2011 JINST 6 P08012.

– 14 –

2013 JINST 8 P12004

[18] S.-J. Park, W.L. Rogers and N.H. Clinthorne, Effects of Positron Range and Annihilation PhotonAcolinearity on Image Resolution of a Compton PET, IEEE Trans. Nucl. Sci. 54 (2007) 1543.

[19] S.-J. Park et al., Performance evaluation of a very high resolution small animal PET imager usingsilicon scatter detectors, Phys. Med. Biol. 52 (2007) 2807.

[20] A. Studen et al., A silicon PET probe, Nucl. Instrum. Meth. A 648 (2010) S255.

[21] N.H. Clinthorne et al., Silicon as an unconventional detector in positron emission tomography, Nucl.Instrum. Meth. A 699 (2013) 216.

[22] N. Cesca et al., SiliPET: Design of an ultra-high resolution small animal PET scanner based onstacks of semi-conductor detectors, Nucl. Instrum. Meth. A 572 (2007) 225.

[23] X. Llopart, R. Ballabriga, M. Campbell, L. Tlustos and W. Wong, Timepix, a 65k programmable pixelreadout chip for arrival time, energy and/or photon counting measurements, Nucl. Instrum. Meth. A581 (2007) 485.

[24] S. Jan et al., GATE: a simulation toolkit for PET and SPECT, Phys. Med. Biol. 49 (2004) 4543.

[25] K. Amako et al., Comparison of Geant4 electromagnetic physics models against the NIST referencedata, IEEE Trans. Nucl. Sci 52 (2005) 910.

[26] F. Lamare, A. Turzo, Y. Bizais, C.C.L. Rest and D. Visvikis, Validation of a Monte Carlo simulationof the Philips Allegro/GEMINI PET systems using GATE, Phys. Med. Biol. 51 (2006) 943.

[27] D. Lazaro et al., Validation of the GATE Monte Carlo simulation platform for modelling a CsI(Tl)scintillation camera dedicated to small-animal imaging, Phys. Med. Biol. 49 (2004) 271[physics/0411236].

[28] S.R. Cherry, J.A. Sorenson and M.E. Phelps, Tomographic Reconstruction in Nuclear Medicine, inPhysics in Nuclear Medicine, Saunders, chapter 16 (2003).

– 15 –