Physics in Medicine & Biology

PAPER Related content

- 90Y imaging by TOF and non-TOF PET in
The origin and reduction of spurious extrahepatic a phantom of liver SIRT
Larry van Elmbt, Stefaan Vandenberghe,
counts observed in 90Y non-TOF PET imaging Stephan Walrand et al.

- Phantom study to determine optimal PET

post radioembolization reconstruction parameters for PET/MR
imaging of 90Y microspheres following
radioembolization
To cite this article: Stephan Walrand et al 2018 Phys. Med. Biol. 63 075016 Nichole M Maughan, Mootaz Eldib,
Maurizio Conti et al.

- GATE Monte Carlo simulations for

variations of an integrated PET/MR hybrid
imaging system based on the Biograph
View the article online for updates and enhancements. mMR model
B Aklan, B W Jakoby, C C Watson et al.

This content was downloaded by walrands from IP address 130.104.192.243 on 30/03/2018 at 12:17
 Phys. Med. Biol. 63 (2018) 075016 (15pp) https://doi.org/10.1088/1361-6560/aab4e9

PAPER

The origin and reduction of spurious extrahepatic counts observed

in 90Y non-TOF PET imaging post radioembolization
RECEIVED
15 November 2017
RE VISED
28 February 2018
ACCEP TED FOR PUBLICATION
Stephan Walrand1, Michel Hesse, François Jamar and Renaud Lhommel
7 March 2018 Cliniques Universitaires Saint-Luc, Av Hippocrate 10, 1200 Brussels, Belgium
1
PUBLISHED
Author to whom any correspondence should be addressed.
29 March 2018 E-mail: stephan.walrand@uclouvain.be

Keywords: yttrium-90 PET, TOF reconstruction, random correction, Monte Carlo, bremsstrahlung

Abstract
Our literature survey revealed a physical effect unknown to the nuclear medicine community, i.e.
internal bremsstrahlung emission, and also the existence of long energy resolution tails in crystal
scintillation. None of these effects has ever been modelled in PET Monte Carlo (MC) simulations.
This study investigates whether these two effects could be at the origin of two unexplained
observations in 90Y imaging by PET: the increasing tails in the radial profile of true coincidences, and
the presence of spurious extrahepatic counts post radioembolization in non-TOF PET and their
absence in TOF PET. These spurious extrahepatic counts hamper the microsphere delivery check in
liver radioembolization.
An acquisition of a 32P vial was performed on a GSO PET system. This is the ideal setup to study
the impact of bremsstrahlung x-rays on the true coincidence rate when no positron emission and
no crystal radioactivity are present. A MC simulation of the acquisition was performed using Gate-
Geant4. MC simulations of non-TOF PET and TOF-PET imaging of a synthetic 90Y human liver
radioembolization phantom were also performed.
Internal bremsstrahlung and long energy resolution tails inclusion in MC simulations
quantitatively predict the increasing tails in the radial profile. In addition, internal bremsstrahlung
explains the discrepancy previously observed in bremsstrahlung SPECT between the measure of
the 90Y bremsstrahlung spectrum and its simulation with Gate-Geant4. However the spurious
extrahepatic counts in non-TOF PET mainly result from the failure of conventional random
correction methods in such low count rate studies and poor robustness versus emission-
transmission inconsistency. A novel proposed random correction method succeeds in cleaning the
spurious extrahepatic counts in non-TOF PET.
Two physical effects not considered up to now in nuclear medicine were identified to be at the origin
of the unusual 90Y true coincidences radial profile. TOF reconstruction removing of the spurious
extrahepatic counts was theoretically explained by a better robustness against emission-transmission
inconsistency. A novel random correction method was proposed to overcome the issue in non-TOF
PET. Further studies are needed to assess the novel random correction method robustness.

Introduction

After performing the first 90Y imaging by TOF-PET in humans (Lhommel et al 2009), we additionally observed that,
despite the low positron yield, the PET image displayed a much better spatial resolution than that of conventional
bremsstrahlung-SPECT. In agreement with SPECT, no count was observed outside the radioembolized liver.
Since then, this novel 90Y imaging modality has undergone tremendous clinical development. This modality is
also becoming the standard imaging technique in microspheres delivery checks (Pasciak et al 2014) as well as post
therapy dosimetry assessment (Willowson et al 2015). All studies comparing SPECT and PET in post liver 90Y
radioembolization imaging concluded that, even using standard commercial PET reconstruction softwares, PET
achieves better performances, both for spatial resolution and for quantification (Kao et al 2011, Elschot et al 2012,

© 2018 Institute of Physics and Engineering in Medicine

Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Barber et al 2013, Padia et al 2013). Many studies (van Elmbt et al 2011, Walrand et al 2011, Behrendt et al 2012,
Goedicke et al 2013, Takahashi et al 2015, Maughan et al 2016) have also been conducted in order to understand
and to optimize the performances of 90Y PET imaging. However, despite this abundant work, all the physics
underlying this modality is not yet fully understood.
In this regard, an issue that remains open concerns the origin of the spurious extrahepatic counts often seen in
90
Y non-TOF PET post liver radioembolization, which vanish when using the TOF-PET system (e.g. see Lhom-
mel et al (2009) versus Werner et al (2010)). In some cases, these spurious extrahepatic counts can make it dif-
ficult to identify real, but unwanted extrahepatic microspheres delivery which is mandatory in order to anticipate
side effects. It could also make the quantification of lung shunting challenging. It was early observed (van Elmbt
et al 2011) that, unlike for pure positron emitters, the true coincidences radial profile in 90Y PET exhibits tails
that increase with the radial distance. A recent study (Conti and Eriksson 2016) confirmed that this property is
shared by most positron emitters that have simultaneous single photon emissions, such as de-excitation gamma
rays or bremsstrahlung x-rays. Note that these increasing tails could artificially lead to an over-compensation of
the scatter when using tail-fitting correction methods. As in pure positron emitters, the tails outside the patient
boundary can only come from the scattered events; tail fitting is often used to fine tune the scatter model to the
individual patient geometry.
For 90Y, it was initially hypothesized (van Elmbt et al 2011) that these increasing tails could result from the
creation of e+e− pairs in the crystal by the high energy bremsstrahlung x-rays and that they could also be at the
origin of the observed spurious extrahepatic counts. Three features supported this assumption: the total solid
angle between two opposite crystal pixels increases with the radial distance similarly to the tails’ increase; coming
from a single annihilation, such coincidences cannot be compensated by the random correction, and lastly, these
annihilations occurring in the crystal, TOF information discards them during TOF reconstruction.
However, state-of-the-art Monte Carlo (MC) simulations (Strydhorst et al 2016) of 90Y PET using Gate 7.2
(Jan et al 2004) did not reveal any such significant pair creations in the crystal and did not exhibit any radial pro-
file tail increasing. Triggered by these recent MC simulations, we searched the literature for non-modelled phys-
ics effects in Gate, i.e. not modelled in Geant4 (Agostinelli et al 2003), which could explain these increasing tails.
Specific photon emissions occurring in a single crystal lattice not modelled in Gate, such as coherent
bremsstrahlung, electron channelling emission, parametric x-ray radiation, coherent e+e− pair creation, were
investigated, but discarded due to their emission probability being too low. However, pixelated crystal detectors
exhibit a long energy resolution tail (Bonifacio et al 2010) that is rarely modelled in MC simulation of SPECT
and PET acquisitions. As a result, a fraction of the high energy bremsstrahlung x-rays backscattered by the crys-
tal (mostly around 250 keV according to the Compton formula) will be shifted by this long energy resolution
tail into the usual [450 650] keV energy acquisition window. These long energy resolution tails were not imple-
mented in the previous Gate simulations (Strydhorst et al 2016).
Regarding the 90Y x-rays emission in the patient, we found a production not modelled in the ion source
of Gate and Geant4: the so-called internal-bremsstrahlung (Cengiz and Almaz 2004). Internal-bremsstrahlung
emission is present in all beta decays as a result of the interaction of the emitted beta particle with the electrical
field of the parent nucleus. Internal bremsstrahlung has been accurately measured and modelled in fundamental
nuclear physics as it gives information about the nucleus electrical field and therefore about the nucleons’ distri-
bution. In contrast to the external bremsstrahlung the intensity of which is proportional to the square of atomic
number Z of the crossed medium, the internal-bremsstrahlung intensity only depends on the parent nucleus.
As a result, for most beta emitter (including 32P and 90Y) above 700 keV, the internal-bremsstrahlung intensity
significantly supersedes that of the external-bremsstrahlung in low Z media, such as water or carbon (Liden and
Starfelt 1955).
In this paper, MC simulations showed that these two missing effects quantitatively explain the increasing tails
of the true coincidences radial profile. For this purpose, simulations and acquisitions of a 32P vial using a GSO
PET were used. This was the ideal setup to study the bremsstrahlung component in the true coincidences rate
as no 511 keV emission and no crystal radioactivity were present. Lastly, the spurious extrahepatic activities in
90
Y imaging using LSO, or LYSO, non-TOF PET were proved to mostly arise from the abundant random coinci-
dences produced by the 176Lu radioactivity. As these randoms are not rightly removed by the existing corrections
methods, a novel correction method which succeeded in removing these extrahepatic activities is proposed.

Materials and method

GSO-PET
The GSO-PET is a small animal Mosaic PET system (Huisman et al 2007) (Philips Medical Systems) which is
the last non-lutetium-based crystal PET that our department still owns. The crystal ring has an inner diameter
of 20.4 cm for an axial length of 12 cm made by the repetition of 2  ×  2  ×  10 mm3 GSO crystals (278 along the
circumference, 52 along the axial direction) separated by 0.3 mm-thick reflective material. A 12 mm-thick

2
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

continuous light guide connects all the crystal pixels to six Anger logic modules allowing coincidence acquisitions
bins ranging from  −46 to  +46.

LYSO-PET
The human LYSO-PET is the first generation 650 ps TOF resolution Gemini TF PET (Surti et al 2007) (Philips
Medical Systems). The crystal ring has an inner diameter of 90.4 cm for an axial length of 18 cm made by the
repetition of 4  ×  4  ×  22 mm3 LYSO crystals (644 along the circumference, 44 along the axial direction) separated
by 0.4 mm-thick reflective material. A thick continuous light guide connects all the crystal pixels to 28 Anger logic
modules.

Point sources acquisition setup

On both systems, all source vials were acquired in list mode using the Philips user5 software with the whole
energy range selected. The vials were set in the centre of the field of view. Vials filled with water were also acquired
with the same setup in order to correct for the background (bg) originating from cosmic gamma rays, from
40
K contained in the concrete walls and also from 176Lu for the LYSO-PET. True coincidence counts (cc) were
assumed to be:
Tcc = Pcc (source) − Dcc (source) − (Pcc (bg) − Dcc (bg))
(1)
where T, P and D refer to true, prompt and delayed, respectively.
An 18 h-acquisition of a 1.3 GBq 32P vial was performed on the GSO-PET in order to study the true coinci-
dence radial profile.
A 20 h-acquisition of a 0.2 MBq 111In vial was performed on the GSO-PET in order to measure the energy
spectrum of the 171–243 keV true coincidences. The energy resolution around the backscatter peak of 32P
bremsstrahlung x-rays was derived from this acquisition.
For the LYSO-PET it appeared that a hardware cut-off limits the acquisition window range to [300 850] keV.
A 1 h-acquisition of a 2 MBq 18F was performed to assess the energy resolution. In order to reduce residual
contribution originating from the 176Lu activity, the true coincidences computed using equation (1) were also
restricted to a TOF  <  400 ps and to the line of responses (LORs) crossing the centred vial.

MC simulations
Gate 7.2 was used with the standard opt3 physics list. As Geant4.10 does not yet include internal bremsstrahlung
in the ion source, 32P and 90Y sources were modelled using gamma rays histogram sources corresponding to
the simulated external bremsstrahlung in water and to the measured internal bremsstrahlung. As moderate
statistic simulations revealed that the PET hardware surrounding the crystal had a negligible impact on the
bremsstrahlung component in true coincidences (data not shown), only the pixelated crystal ring was modelled
in the high statistic simulations in order to get an achievable computation time.
The hits file output was used to allow a clear identification of the true coincidences which is not always the
case when using the Gate digitizer (Strydhorst et al 2016). Moreover, the hits file allowed us to track all the physi-
cal effects arising in the crystal that ended up in a true coincidence (true refers here to the fact that the two pho-
tons detected in coincidence originated from the same bremsstrahlung primary photon).

Bremsstrahlung modelling
Figure 1 shows the internal-bremsstrahlung and external-bremsstrahlung emissions of 32P and of 90Y.
The histogram gamma sources used in Gate were obtained by fitting the bremsstrahlung emissions with:
 β γ β γ

(2) B (E) = a e−bE −cE − e−bEm −cEm

where Em is the maximal energy of the beta particle, i.e. 1.7 and 2.3 MeV for 32P and 90Y, respectively.

Phantoms simulation
An elliptical cylinder phantom (35 cm-long axis, 20 cm-short axis, 20 cm-length) filled with water, centred in the
LYSO ring and containing a 10 cm-diameter spherical 90Y source representing the right liver was modelled with
Gate. The spherical ‘liver’ source was centred in the axial direction, but was off-centre in the transverse plane. The
‘active liver’ contained a small 3 cm-diameter spherical source with a specific activity concentration threefold
higher. Positrons and bremsstrahlung x-rays corresponding to a clinical one bed acquisition with a statistics
of 20 min  ×  1GBq were simulated. A homemade digitizer was developed in order to compute the coincidence
events in the [450 650] keV acquisition window using the modelled energy resolution including the long tail.
A TOF Gaussian resolution of 650 ps corresponding to that of our LYSO-PET was randomly added to the TOF
provided by the MC simulation.

3
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 1. Bremsstrahlung emissions of 32P (left) and of 90Y (right). Brown circles: experimental measures of internal-
bremsstrahlung emissions extracted from Liden and Starfelt (1955) for 32P and from Venkataramaiah et al (1980) for 90Y. Violet
squares: external-bremsstrahlung emission in water obtained using the ion source of Gate. Dashed line: theoretical prediction of the
90
Y internal-bremsstrahlung (Cengiz and Almaz 2004). Solid lines: analytical fits using equation (2) with parameter values given in
table 1.

Table 1. Fitted parameters for E and Em expressed in MeV. The two last columns are the bremsstrahlung weights, quantifying the relative
contribution of internal and external bremsstrahlung.

a B β c γ % (>50 keV) % (>800 keV)

32
P E-BREMSS 25.6 10.15 2.704 0.215 1.550 61 17
32
P I-Bremss 25.9 10.44 0.725 0.179 3.359 39 83
90
Y E-BREMSS 40.0 10.15 1.200 0.213 2.000 67 32
90
Y I-Bremss 25.9 9.713 0.493 0.161 2.599 33 68

Rather than simulating the coincidences originating from the 176Lu present in the LYSO ring, a single bed
acquisition of 20 min was acquired on the LYSO-PET with a non-active elliptical cylinder phantom centred in
the FOV with the same dimensions and attenuation coefficient to those modelled in Gate. The purpose of this
phantom was to reproduce the attenuation and scatter that occurred as a result of the 176Lu gamma rays escaping
the crystal and crossing the patient in an actual clinical acquisition. The coincidence events obtained from the
90
Y simulation were then inserted into this acquired list file in order to be reconstructed by the Philips software. A
synthetic noise-free CT scan corresponding to the phantom was also generated.
Three different list files were generated, i.e. including the coincidences originating: from the 90Y positrons
alone, from 90Y positrons and from the 176Lu decays together, from 90Y positrons, from 90Y bremsstrahlung and
from the 176Lu decays all together. Afterwards the list files were reconstructed using the Philips software, i.e. non-
TOF 3D-RAMLA and TOF LOR reconstruction (Wang et al 2006) both with three iterations and eight subsets,
using the synthetic CT scan for the attenuation correction.

Novel random correction method

The simplest random correction method estimates the random coincidences by shifting the time window of
the second detector by a delay much larger than the maximal possible TOF difference between two annihilation
gamma rays. This second acquisition is then simply subtracted from the prompt coincidences acquisition.
Due to inevitable fluctuations linked to the Poisson statistics of radioactive decays, negative values are pro-
duced. If there is no problem in reconstructing negative values with the filtered back-projection algorithm, for
expectation maximization of the maximum likelihood (EM-ML)-based iterative algorithms such as ordered
subset expectation maximization (OSEM) it is mandatory to rule out these negative values, i.e.:
  
n+1
Anj  max(Ti + Ri − Di , 0)
Aj =  cij  n
(3) i cij k cik Ak
i

where cik is the probability that an annihilation occurring in the voxel k will be detected in the acquisition bin i.
Ti, Ri and Di are the true, random and delayed coincidences respectively measured in the bin i. The upper curly
bracket above Ti  +  Ri indicates that obviously true and prompt randoms cannot be measured separately. The
measured prompts were split into trues and randoms in the equation for discussion. Due to the absence of any

4
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

 
time correlation between the two isotope decays, one has Ri ≈ Di. However, due to statistics fluctuations,
i i
bins i where Ri  =  Di are very scarce.
The negative bin truncation in equation (3) can induce significant bias in the reconstruction (Tapp et al 2014,
Walrand et al 2015), especially when the random coincidences contribution is significant such as in 90Y PET
imaging (van Elmbt et al 2011). Modified EM–ML algorithms allowing use of negative values during reconstruc-
tion (Nuyts et al 2002, Lim et al 2018) significantly reduce this truncation bias.
In order to reduce generation of negative values, three random correction methods have been proposed:
smoothing of the delayed coincidences; estimation of the random coincidence using the detectors single rates;
replacing the subtraction present in the numerator of equation (3) by an addition in the denominator (Wang
et al 2006), i.e. replacing equation (3) by:   
n 
A j Ti + Ri
(4) An+1
j = cij  n .
c
i ij i k cik Ak + Di
  
However, it is of prime importance to realize that for bins i for which Ti + Ri = 0, and Di  >  0, all the cited
methods neglect the random contribution Di detected in this bin i, as the numerator in equations (3) and (4)
vanishes for this bin i.
For a typical 20 min acquisition per bed position in a 1 GBq 90Y acquisition, about 1.5  ×  106 random and
1  ×  105 true coincidences originating from the 176Lu and 90Y, respectively, are detected. A standard sinogram
rebinning of 64 angles, 96 bins, 7 tilts and 40 slices is made of 1.7  ×  106 bins. For this case a simple numeric simu-
lation with uniform random drawing shows that on average 6.2  ×  105  ±  103 randoms occurred in bins such as
Di  >  0 with Pi  =  0 and are thus neglected by the existing correction methods.
In order to overcome this issue, we applied a direct random correction on the acquired list file according to
the following way. For each delayed event, the prompt event with the LOR closest to that of the delayed event is
identified. Both events are then removed from the list file. The distance between the LORs of the prompt coinci-
dences p and of the delayed coincidences d is defined as:
 2  2  2  2
(5) dist ( p, d) = ∆ nAp − nAd + ∆ nBp − nBd + zpA − zdA + zpB − zdB
where A, B refer to the two detectors triggering the coincidences, n (0  →  643) is the number of the detector along
the ring, and z (0  →  43) is the position of the detector in the axial direction.
Δ(n) takes into account the cyclic nature of the number n, i.e.:
∆ (n) = |n| if |n| < 322 and ∆ (n) = 322 − |n| otherwise.
(6)

TOF robustness versus emission-transmission inconsistency

The variance reduction in active regions has been intensively studied (see for example Conti 2006, Vunckx et al
2010, Eriksson and Conti 2015). These studies show that the TOF improves the variance in a volume of interest
inside a radioactive volume by a factor proportional to the ratio of such radioactive object size and the spatial
uncertainty associated to the system time resolution. To our knowledge no such study has been performed on
what happens in a non-active region, such as the extrahepatic region in liver radioembolization.
In this section we will analytically compute the non-TOF and TOF reconstruction of the intrinsic LYSO crys-
tal prompts coincidences in the presence or not of an attenuating cylinder in the PET FOV. This will give the
impact of the residual randoms not rightly corrected by equation (4) or (5).
Let us consider a background acquisition using a lutetium crystal-based PET. All prompts are intrinsic ran-
doms that are spatially uniform resulting in the constant non-TOF sinogram:
S (r, ϕ) = s2 2Tw Tacq
(7)
where s is the 176Lu single counts rate, Tw is the coincidence timing window (6 ns) and Tacq is the acquisition
duration.
As successive decays are not correlated in time, the TOF of random coincidences is uniformly distributed. The
corresponding TOF sinogram associated to the FOV is thus:
 √ 
(8) Stof (r, ϕ, t) = s2 Tacq θ 2 R2 − r2 /c − |t|

where c is the speed of light and R is the reconstruction FOV radius (30 cm).
θ(t) is the Heaviside function which takes into account that √if the TOF t is greater than the time needed by
light to cross the FOV length at the sinogram coordinate r, i.e. 2 R2 − r2 /c , the event will be rejected by the TOF
reconstruction algorithm as being localized outside the FOV.
Let us study how these intrinsic crystal randoms are reconstructed without random correction
(Ti = Di = 0, ∀ i in equation (4)) in the case of the ideal continuous 2D PET tomography, i.e. no scatter, no

5
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 2. Approximation (11) of the inverse of the LOR attenuation (10) as a function of the sinogram radial coordinate r for a
centred water cylinder of 13.22 cm radius (the geometric mean of the elliptical phantom axis).

statistical noise. The detection probability c[tof] at the detector coordinates (r,ϕ) of an annihilation occurring in
(x,y,[t]) with or without TOF for a uniform attenuating cylinder is:
    
[tof] −x sinϕ + y cosϕ 
(9) c (r, ϕ, [t] , x, y) = δ (r − x cosϕ − y sinϕ) δ t − U R2o − r2
c/2
where
  √ 2 2
U R2o − r2 = e−2µ Ro − r
(10)

represents the attenuation depending on the cylinder radius Ro and on the attenuation coefficient µ (≈0.09 cm−1
for 511 keV in water).
Note that the intrinsic crystal randoms are not attenuated by the patient body as they result from two distinct
176
Lu decays occurring in the crystal ring. However, the reconstruction algorithm will apply the patient attenu-
ation to intrinsic crystal randoms not rightly removed by the random correction. The relative impact of such
emission-transmission inconsistency on the non-TOF and TOF EML solutions has already been qualitatively
discussed (Conti 2010). In order to be able to quantitatively compute this impact, let us use the approximation:
√ 2 2 1√ 2   1  2
e2µ Ro − r ≈ R − r2 + e2µRo − 1 θ (Ro − |r|) Ro − r 2 .
(11) R R o

Figure 2 shows the approximation:

The non-TOF EM-ML solution is (see annexe A):
 
s2 2Tw Tacq 1  2µRo  1
(12) A (ρ) = + e −1 θ (Ro − |ρ|) .
2 R Ro
An easy calculus (see annexe B) shows that the TOF EM-ML solution that maximizes the likelihood is:

tof s2 Tacq 2
A
(13)(ρ) = π 1 1
.
√ 2 2 2 √ 2 2 2 dϕ c
´
0 π 1 2µR 1
R −ρ cos ϕ+ (e o −1) θ(Ro −|ρ cosϕ|) Ro − ρ cos ϕ
R Ro

In ρ  =  0 the solutions are:

 
s2 2Tw Tacq  2µRo  R
A (0) =
(14) 1+ e −1
2R Ro

s2 2Tacq 2µRo
Atof (0) =
(15) e .
c

6
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 3. Black circles: measured true coincidences energy spectra according to equation (1) for GSO and LYSO PET, and extracted
from Nikolopoulos et al (2013) for the LSO-PET. Blue line: 176Lu-511 keV pile up component simulated with Gate; in B note
the good reproduction of the observed increase before the hardware 850 keV cut-off. Green line: long tail component due to the
scintillation light transport in the crystal pixel (fully covered by the brown curve in A and C). Red lines: conventional Gaussian fits of
the photoelectric peaks. Orange lines: energy spectra obtained by convolution of the gamma ray energies (171 and 245 keV for 111In,
and 511 keV for 18F) with the energy resolution models (equations (7)–(10), not shown for the LSO-PET). For GSO-PET and LYSO-
PET specifications see material and method; for LSO-PET specifications see Nikolopoulos et al (2013).The higher tails amplitude for
the LSO-PET is not linked to the crystal type, but is likely linked to the block design used in place of a continuous light guide for the
GSO and LYSO PETs.

Results

Energy resolution modelling

Figure 3 shows in log scale the true coincidences spectra according to equation (1) for 111In on the GSO PET
(figure 3(A)) and for 18F on the LYSO PET (figure 3(B)). The experimental 18F LSO-PET spectrum (figure 3(C))
was extracted from Nikolopoulos et al (2013).
The energy resolution of the GSO PET was modelled by a sum of a Gaussian and a fourth power exponential:
 1  −(E −E)2 /(2× 0.122 E2.14 ) 4
−(E −E) /(2 (1.13 E)4 )

(16) E res (E, E − E) = e + 0.018e
N(E)
where N(E) is a normalization factor to unity that has to be numerically computed.
The energy resolution of the LYSO PET was modelled by:
1  −(E −E)2 /(2× 0.0532 E2 ) 0.486 2 2

+ 0.254e−|E −E| /(2×0.0022 E ) + PU (E − E)

Eres (E, E − E) =
(17) e
N(E)
where the pile up is:
( ) α E −(E+380)
/(2 ×0.162 (E+380)2 )
PU (E − E) = 0.0014 e−|E −(E+380)|

(18)
with:
α (E) = 1.89 if E > 0, α (E) = 2.48 if E  0.
(19)

Long energy resolution tail: intrinsic crystal activity component

The total 176Lu activity contained in the LYSO volume handled by one single Anger module is about 90 kBq
(Yamamoto et al 2005) while the integration time is 90 ns (Surti et al 2007). As a result, the pile up probability
between any one gamma or x-ray interacting in the crystal and one 176Lu decay occurring in the crystal volume
handled by the Anger module is 90 kBq  ×  90 ns  ≈  0.8%. This pile up shifted the energy of 0.8% of the incident
gamma rays by the energy deposed by the 176Lu decay, and thus just behaves like energy resolution degradation.
Note that for PET using independent 40  ×  40 mm2 photodetector-crystal pixels block design, the 176Lu activity
in such a block is only 5 kBq, lowering the pile up probability down to  ≈0.04%.
The hardware high energy cut-off present on the LYSO-PET limits the long energy resolution tail measure-
ment to 850  −  511  ≈  340 keV. Thus we extended the 18F acquired energy spectrum with the 176Lu energy spec-
trum of the crystal hosting the decay shifted by 511 keV. The 176Lu spectrum was normalized to represent 0.8% of
the total spectrum.
Due to the hardware low energy cut-off, it was not possible to measure this 176Lu spectrum, which was thus
simulated using the ion source of Gate. The spectrum was convolved with the conventional LYSO Gaussian
energy resolution at 511 keV (Surti et al 2007).

7
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 4. (A) Schematic representation of a photoelectric effect (red star) in a crystal pixel (yellow). (B) Photoelectric effects yield
as a function of the DOI which mostly follows the attenuation law e−µ DOI . C: exiting scintillation photons yields as a function of
the DOI (see Bonifacio et al (2010)). The number of scintillation photons directly exiting the crystal (red arrows in (A)) decreases
together with the distance to the photodetector and is not fully compensated by back scintillation photons (blue arrow) partially
reflected by the reflective material wrapping the crystal (gray lines in (A)). (D) Resulting energy spectrum which is mostly the photo-
electric effect yields expressed as a function of the exiting scintillation photons yields (brown lines). The long energy resolution tail
is clearly visible. Note that (D) does not take into account the intrinsic crystal energy resolution, the FWHM of which will change the
spectrum peak to long tail ratio.

Long energy resolution tail: scintillation light transport component

Long energy resolution tails in non-active crystal have been experimentally observed and explained by optical
transport MC simulations (Bonifacio et al 2010). The long tails resulted from the decrease of the number of
scintillation photons reaching the PMTs with the depth of interaction (DOI) due to reflection loss on the lateral
crystal walls. Figure 4 details this mechanism, which was also experimentally proved by changing or removing the
lateral reflector material of a long crystal (Kalinnikov and Velicheva 2015).
The light transport component to the spectrum peak ratio observed using the LSO-PET is about hundred-
fold higher than that observed in the LYSO-PET system. This does not result from the different crystal type, but
likely from the four PMTs-crystal block design used in the LSO-PET. Such a block design suffers from an addi-
tional variation of scintillation light collection depending on the crystal pixels’ location versus the block edges. In
contrast, using a thick continuous light guide, such as in the GSO and LYSO PET systems, all crystal pixels always
lie somewhere between two adjacent PMTs. This makes the scintillation light collection between the crystal pixels
more uniform, reduces the impact of the DOI and also reduces the tail of the light response function (Surti et al
2000).
Although sharing the same continuous light guide technology, the light transport component to the spec-
trum peak ratio observed using the GSO-PET is about fiftyfold higher than that observed in the LYSO-PET sys-
tem. GSO has a poorer intrinsic energy resolution due to its lower scintillation yields. This results in a lowering of
the spectrum peak. Lastly, the reduced crystal thickness (10 mm) in the GSO-PET versus that in the LYSO-PET
(22 mm) increases the fraction of photoelectric effects occurring near the photo detector, and thus also increases
the level of the long energy resolution tail.
In addition to all the previously discussed features, the crystal edge polishing and type of reflector material
also play a role by controlling the number of reflections needed to reach the PMTs, but this information is usually
not provided by the system manufacturer.

32
P radial profile
Figure 5(A) shows the true coincidences radial profile in the [450 650] keV energy window for the 32P acquisition
on the GSO-PET. The best agreement of the Gate simulations (red line) with the measured profile (brown
circles) was obtained by using 94% of the 32P activity indicated by the manufacturer. This is in line with routine

8
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 5. (A) True coincidences radial profiles for the 32P vial acquired on the GSO-PET. Brown circles: measured true coincidences
rates. Colour solid lines: gate simulations with all effects (red), neglecting the internal bremsstrahlung (blue) and neglecting the
long energy resolution tail (green). Black line: total solid angle times vial attenuation. The vertical dashed lines correspond to the
vial edges. (B) Schematic explanation of the radial profile: the high-energy bremsstrahlung x-ray (red line) can be scattered back
by crystal A into the crystal pixel B. The energy depositions in crystal A by the recoil electron and in crystal B by the photoelectron
are detected only as prompt coincidence. The long energy resolution tails shifted a fraction of the backscattered photon into the
acquisition window while the recoil energy can directly fall into the acquisition energy window.

measurement accuracy of a pure beta emitter source. Neglecting the internal bremsstrahlung (blue curve) or the
long resolution energy tail (green curve) both result in a dramatic underestimation of the true coincidences rate
produced by the bremsstrahlung x-rays.
Figure 5(B) shows the geometric explanation of the radial profile as deduced from the Geant4 hits file analy-
sis: the high energy bremsstrahlung x-ray (red line) can be scattered back by crystal A into the crystal pixel B.
The energy depositions in crystal A by the recoil electron and in crystal B by the scattered x-ray are detected as a
prompt coincidence. The long energy resolution tails shifted a fraction of the backscattered x-rays, from external
or internal bremsstrahlung, into the acquisition window, while the recoil electron energy can directly fall into the
acquisition window. The higher abundance of internal bremsstrahlung at high energy compared to that of exter-
nal bremsstrahlung (see figure 1 and table 1) explained its major impact. The increase of the radial profile tail is
simply the consequence of the corresponding increase of the scattering solid angle.

Spurious extrahepatic counts MC simulations

Figure 6 shows the reconstructed central transverse slice of the phantom modelling a typical 1GBq 90Y right
lobe radioembolization. Table 2 shows the coincidence counts (cc) in these simulated phantom acquisitions
originating from the positron annihilation (second column), from the coincidences generating the radial profile
tail increase as described in figure 5(B) (third column) and from the random coincidences between 2 176Lu decays
(last column). Further, the contributions for a LSO-4PMTs bloc-based PET were also estimated using the long
energy resolution tail derived from figure 3(C).
Non-TOF reconstructions with the manufacturer random correction 3D-RAMLA software (Wang et al
2006) showed a clear impact of the randoms originating from the natural 176Lu radioactivity present in the crys-
tal ring (figures 6(C) versus (B)). This clearly underlines the failure of conventional random correction methods
in low statistics acquisitions.
In contrast, the inclusion of the true coincidences originating from the 90Y bremsstrahlung x-rays did not
produce any observable effect in the image (figures 6(F) versus (C)). The manufacturer LOR-TOF reconstruc-
tion software (Wang et al 2006) perfectly cleaned all the spurious extrahepatic counts and achieved an imaging
quality as good as in an ideal acquisition free of 176Lu randoms and bremsstrahlung true coincidences (figures
6(D) versus (A)).
Figures 6(G) and (H) shows the TOF and non-TOF reconstruction after removing the delayed coincidences
in the list file. This corresponds to reconstructions without random correction. As expected, figure 6(H) shows
an increase in spurious extrahepatic counts. In contrast, the TOF reconstruction remains free of spurious extra-
hepatic counts, demonstrating its better robustness against emission-transmission inconsistency.
For non-TOF reconstructions, the novel proposed random correction nicely cleaned the spurious extrahe-
patic counts and provided an imaging quality similar to that obtained in an ideal acquisition (figures 6(E) versus
(B)). MC simulations using the LSO-PET energy resolution showed that even with the much higher tail comp­
onent, the true coincidences originating from the 90Y bremsstrahlung x-rays did not produce any observable
effect (data not shown).

9
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 6. Fused CT/PET reconstructed central transverse slice of the simulated 90Y phantom corresponding to a 1 GBq  ×  20 min
statistics. (A) and (B) TOF and non-TOF reconstruction of an ideal acquisition (i.e. no bremsstrahlung, no 176Lu, no randoms),
respectively. (C) and (F) Non-TOF reconstruction with delayed windows random correction for a real LYSO containing 176Lu,
without and with bremsstrahlung x-rays, respectively. Note the abundant spurious extrahepatic counts induced by the 176Lu
randoms, while the 90Y bremsstrahlung x-rays have a marginal impact. (E) Non-TOF reconstruction after correction of the randoms
with the novel proposed method. (D) TOF reconstruction of clinical situation which is almost as good as the ideal situation
(A). (G) and (H) TOF and non-TOF reconstruction without random correction, clearly showing the better robustness of TOF
reconstruction versus emission-transmission inconsistency.

Table 2. Count contributions for the phantom simulations.

True e+ cc True bremss. cc Random 176Lu cc

LYSO-PET 136k 6k 1475k

LSO-PET 137k 220k 1475k

TOF robustness versus emission-transmission inconsistency

Figure 7 shows different reconstructions without random correction and without scatter correction of a 20 min
background acquisition performed on the LYSO-PET.

Novel random corrections in the patient

Figure 8 shows reconstructed transverse slices post right liver radioembolization using 1GBq 90Y loaded resin
microspheres. The proposed novel random correction method cleaned a large part of the spurious extrahepatic
counts.

Discussion

Internal bremsstrahlung and a long energy resolution tail are two effects that quantitatively explain the radial
profile tail increase. However, if they have a major impact on the tails, their impact on spurious extrahepatic
counts in 90Y PET imaging post radioembolization is marginal. This is explained by the fact that the backscattered
x-rays in a crystal have to cross the patient to reach the opposite crystal to be detected as a true coincidence.
Due to their low energy (≈250 keV) they underwent a high attenuation in the patient, i.e. a reduction factor of
approximately e0.12×30  =  37.

10
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Figure 7. Central transverse slices (displayed with a common gray scale) of manufacturer’s reconstructions without scatter
correction and without random correction of a 20 min background acquisition performed on the LYSO PET, i.e. just acquiring
intrinsic crystal randoms. (A) and (C) TOF reconstruction without and with attenuation correction, respectively. (B) and (D) Non-
TOF reconstruction without and with attenuation correction, respectively. Numbers in blue represent the reconstructed number of
counts within the boundary (blue line) of the attenuation map used in figure 6 and in the reconstructions (C) and (D) as well. Note
the much lower random counts in TOF reconstructions versus corresponding non-TOF reconstructions.

Figure 8. Fused typical transverse PET/CT slice post radioembolization of the right liver with 1 GBq 90Y loaded resin microspheres.
Two bed positions were acquired for a total time of 40 min on a LYSO-PET/CT system with TOF capabilities. (A) Manufacturer’s
non-TOF 3D-RAMLA OSEM reconstruction and random correction methods. (B) Manufacturer’s non-TOF 3D-RAMLA OSEM
reconstruction after random correction with the proposed method. (C) Manufacturer’s TOF-LOR reconstruction.

As already shown, the intrinsic LYSO or LSO activity induces by itself an increasing radial profile tail (Conti
et al 2017). The beta energy deposition in the crystal pixel holding the 176Lu decay and an escaping 307 or 401 keV
gamma rays energy deposition in another crystal pixel can be detected as a prompt coincidence. However, in a 90Y
therapeutic activity setup this effect is about fivefold lower than that originating from 90Y bremsstrahlung x-rays.
However, one has to keep in mind that their impact could be less negligible in small animal imaging due to
low body attenuation and non-TOF PET use. In humans, these increasing tails could artificially lead to an over-
compensation of the scatter with tail-fitting correction methods, especially in non-TOF PET systems using a four
PMTs-crystal block design which are affected by a higher long energy resolution tail.
The long energy resolution tail of LYSO-PET is well explained by the 176Lu decay piling up with detected
annihilation gamma rays. On the other hand, the natural abundance of the radioactive isotope 152Gd (≈0.2%) is
too low to induce the same effect for the GSO-PET. The stem effect (Therriault-Proulx et al 2013) resulting from
Cerenkov and radioluminescence light emission in the light guide, was also found to be too low. Experiments
(Bonifacio et al 2010) and optical MC transport simulations (Kalinnikov and Velicheva 2015) show that the long
energy resolution tail results from photon loss in multiple reflections on the lateral walls needed to reach the
photo-detector.
Rault et al (2010) already observed an underestimation of Geant4 and MCNPX for the high energy part of
the 90Y bremsstrahlung-spectrum in water versus their measurement performed using a germanium detector.
Without finding the origin of this discrepancy, i.e. the internal-bremsstrahlung emission, Rong et al (2012) and

11
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Table 3. Theoretical counts (equations (14) and (15)) in the centre of the FOV (black numbers) and the measured counts within the
phantom boundary (blue numbers). Numerical values are given relatively to the non-TOF reconstruction without attenuation correction.
Ro was set to the geometric mean of the elliptical phantom axes. Note the good agreement between theoretical and measured values.

µ used in recon Atof (0) ×2R A(0)×2R

(Meas.)
s2 2Tw Tacq (Meas.) s2 2Tw Tacq
−1 2R 1 (1)
0.00 cm cTw ≈ 0.33 (0.40)
0.09 cm−1 2R 2µRo
cTw e ≈ 3.60 (3.54) ≈ R 2µRo
Ro e ≈ 24.51 (24.31)

Dewaraja et al (2017) rightly made the choice to use this measured spectrum in their state-of-the-art bremsstrahl-
ung-SPECT reconstruction, which resulted in improved reconstruction.
MC simulations of the phantom modelling clinical radioembolization setup showed that, using lutetium-
based crystal, the spurious extrahepatic counts mainly originated from the randoms produced by the 176Lu activ-
ity present in the crystal. Using BGO, due to the larger time coincidence window used, above 1 GBq the 90Y
bremsstrahlung x-rays produce as much random coincidences as those produced by 176Lu.
Due to the low count rate, half of these random coincidences are recorded in sinogram bins free of prompt
coincidences. As shown, all the proposed correction methods in non-TOF PET simply neglect this fraction
of random coincidences. In contrast, the phantom simulation and the patient study show that our proposed
method which takes into account all the detected random coincidences, succeeds in cleaning the spurious extra-
hepatic counts. Implementation of this method should help the clinician in excluding real, but unwanted extra-
hepatic microspheres delivery when using non-TOF PET systems.
The random correction used in the TOF LOR reconstruction is similar to that used in the non-TOF
3D-RAMLA reconstruction methods (Wang et al 2006), i.e. the smoothed delayed bins set as an addition in the
denominator of the iterative algorithm such as in equation (4). As shown in the materials and method section, the
much better performance of TOF PET results from its robustness versus the inconsistency between the attenua-
tion map used in the reconstruction and the fact that intrinsic crystal randoms did not undergo any attenuation.
While TOF reconstruction wrongly magnifies the residual randoms by the inverse of the attenuation, non-TOF
reconstruction further exaggerates this wrong magnification by the ratio of the FOV to the attenuating medium
dimensions (see table 3). Conti (2011) already observed and qualitatively discussed this effect, but for the true
coincidences (see figure 4 in Conti 2011).

Conclusions

From our literature survey we found a physical effect forgotten by the nuclear medicine community and by the
high energy physics community as well, i.e. internal bremsstrahlung. This internal bremsstrahlung explains the
discrepancy observed in bremsstrahlung SPECT between the measured 90Y bremsstrahlung x-ray spectrum and
the one simulated by Gate-GEANT4 which does not include internal-bremsstrahlung modelling.
Our MC simulations show that the internal bremsstrahlung and the long energy resolution tail jointly explain
the tails’ increase in the 90Y PET radial profile. Further investigations are needed to investigate how this tails’
increase could impact on tail tuning scatter correction methods.
The spurious extrahepatic counts in non-TOF PET originate from the failure of conventional random cor-
rection methods in low count rate studies. The spurious extrahepatic counts removal in TOF reconstruction
was theoretically explained resulting in a better robustness against emission-transmission inconsistency. A novel
random correction method was proposed to remove the spurious extrahepatic counts in non-TOF PET.
Further studies are needed to assess the novel random correction method’s robustness.

Acknowledgments

The authors thank Dr J Strydhorst and Dr T Carlier for their valuable discussions about MC simulations in 90Y
PET imaging.

Ethics

No human studies and no animal studies were performed for this work.

Annexe A. non-TOF solution

The Poisson log likelihood a measured sinogram S (r, ϕ) and an activity distribution A is:
   
ˆ
(A.1) L (A) = dr dϕ S (r, ϕ) log S (r, ϕ) − S (r, ϕ)

12
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

with:
ˆ
dx dy c (r, ϕ, x, y) A (x, y) = S (r, ϕ).
(A.2)

Let us consider the expression:

 
l (r, ϕ) = S (r, ϕ) log S (r, ϕ) − S (r, ϕ) .
(A.3)
The derivative is:
dl (r, ϕ) S (r, ϕ)
(A.4) = − 1.
dS (r, ϕ) S (r, ϕ)
So if it exists, the exact solution of the equation is:
ˆ
(A.5) dx dy c (r, ϕ, x, y) A (x, y) = S (r, ϕ).

This solution will maximize the likelihood.

Using equations (7), (9) and (11) equation (A.5) becomes:
dx dy δ (r − x cosϕ − y sinϕ) A (x, y)
´
 √  
(A.6)
= s2 2Tw Tacq R1 R2 − r2 + (e2µRo − 1) θ (Ro − |r|) R1o R2o − r2

which has the exact solution:

 
s2 2Tw Tacq 1  2µRo  1
A (ρ) =
(A.7) + e −1 θ (Ro − |ρ|)
2 R Ro
where ρ is the polar coordinate in the transverse plane.
To which the EM-ML algorithm converges regarding equations (A.3) and (A.4).

Annexe B. TOF solution

The back-projection of the comparison between the measured sinogram and the estimated projection in
equation (4) for Ti = Di = 0, ∀ i becomes:
 
ˆ δ (r − x cosϕ − y sinϕ) δ t − −x sinϕ+y
c/2
cosϕ

(B.1) s2 Tacq dr dϕ dt ´  
−X sinϕ+Y cosϕ
dXdY δ (r − X cosϕ − Y sinϕ) δ t − c/2 An (X, Y)

1
ˆ
2  
= s Tacq dϕ ´
(B.2)
dXdY δ ((X − x) cos ϕ + (Y − y)sin ϕ) δ −(X−x)sinϕ+(Y−y)cosϕ
c/2 An (X, Y)

1
ˆ
2  
= s Tacq dϕ ´
(B.3)
dXdY δ (X cosϕ + Y sinϕ) δ −X sinϕ+Y
c/2
cosϕ
An (X + x, Y + y)

1
ˆ
2  
= s Tacq dϕ ´
(B.4) Y
dXdY δ (X) δ c/2 An (X cosϕ − Y sinϕ + x, X sinϕ + Y cosϕ + y)

2 1
ˆ
2
= s Tacq dϕ ´
(B.5)
c dXdY δ (X) δ ( Y) An (X cosϕ − Y sinϕ + x, X sinϕ + Y cosϕ + y)

2 1 N 1
ˆ
= s2 Tacq dϕ
(B.6) n
= π n .
c A (x, y) cTw A (x, y)
The backprojection normalization is:
   
ˆ
−x sinϕ + y cosϕ 
(B.7) dr dϕ dt δ (r − x cosϕ − y sinϕ) δ t − U R2o − r2
c/2
ˆ  
2 2
dϕ U
(B.8) Ro − (x cosϕ + y sinϕ) .

As a result:

13
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

s2 2Tacq π
An+1 (x, y) =   .
(B.9) c 2
dϕ U R2o − (x cosϕ + y sinϕ)
´

Writing x = ρ cosθ and y = ρ sinθ we have:

s2 2Tacq π s2 2Tacq π
 An+1 (ρ) =    =   .
c ´
dϕ U R2o − ρ2 cos2 (ϕ − θ) c ´
dϕ U R2o − ρ2 cos2 ϕ
(B.10)

References
Agostinelli S et al 2003 Geant4—a simulation tool kit Nucl. Instrum. Methods Phys. Res. A 506 250–303
Barber T W, Yap K S, Cherk M H, Powell A and Kalff V 2013 Comparison of positron emission tomography/CT and bremsstrahlung imaging
following Y-90 radiation synovectomy J. Med. Imaging Radiat. Oncol. 57 567–71
Behrendt F F, Winz O and Mottaghy F M 2012 Study-parameter impact in quantitative 90-yttrium PET imaging for radioembolization
treatment monitoring and dosimetry IEEE Trans. Med. Imaging 2012 0580
Bonifacio D A, Belcari N, Moehrs S, Moralles M, Rosso V, Vecchio S and Del Guerra A 2010 A time efficient optical model for GATE
simulation of a LYSO scintillation matrix used in PET applications IEEE Trans. Nucl. Sci. 57 2483–9
Cengiz A and Almaz E 2004 Internal bremsstrahlung spectra of β-particle emitters using the Monte Carlo method Radiat. Phys. Chem.
70 661–8
Conti M 2006 Effect of randoms on signal-to-noise ratio in TOF PET IEEE Trans. Nucl. Sci. 53 1188–93
Conti M 2010 Why is TOF PET reconstruction a more robust method in the presence of inconsistent data? Phys. Med. Biol. 56 155
Conti M 2011 Focus on time-of-flight PET: the benefits of improved time resolution Eur. J. Nucl. Med. Mol. Imaging 38 1147–57
Conti M and Eriksson L 2016 Physics of pure and non-pure positron emitters for PET: a review and a discussion EJNMMI Phy. 3 8
Conti M, Eriksson L, Rothfuss H, Sjoeholm T, Townsend D, Rosenqvist G and Carlier T 2017 Characterization of 176Lu background in LSO-
based PET scanners Phys. Med. Biol. 62 3700
Dewaraja Y K, Chun S Y, Srinivasa R N, Kaza R K, Cuneo K C, Majdalany B S, Novelli P M, Ljungberg M and Fessler J A 2017 Improved
quantitative 90Y bremsstrahlung SPECT/CT reconstruction with Monte Carlo scatter modeling Med. Phys. 44 6364–76
Elschot M, Vermolen B J, Lam M G, de Keizer B, van den Bosch M A and de Jong H W 2012 Quantitative comparison of PET and Bremss-
SPECT for imaging the in vivo yttrium-90 microsphere distribution after liver radioembolization PLoS One 8 e55742
Eriksson L and Conti M 2015 Randoms and TOF gain revisited Phys. Med. Biol. 60 1613
Goedicke A, Berker Y, Verburg F A, Behrendt F F, Winz O and Mottaghy F M 2013 Study-parameter impact in quantitative 90-yttrium PET
imaging for radioembolization treatment monitoring and dosimetry IEEE Trans. Med. Imaging 32 485–92
Huisman M C, Reder S, WE-Bremsser A W, Ziegler S I and Schwaiger M 2007 Performance evaluation of the Philips MOSAIC small animal
PET scanner Eur. J. Nucl. Med. Mol. Imaging 34 532–40
Jan S et al 2004 GATE: a simulation tool kit for PET and SPECT Phys. Med. Biol. 49 4543–61
Kalinnikov V and Velicheva E 2015 Research of long GSO and LYSO crystals used in the calorimeter developed for the COMET experiment
Funct. Mater. 22 126–34
Kao Y H, Tan E H, Ng C E and Goh S W 2011 Yttrium-90 time-of-flight PET/CT is superior to Bremss-SPECT/CT for
postradioembolization imaging of microsphere biodistribution Clin. Nucl. Med. 36 e186–7
Lhommel R, Goffette P, Van den Eynde M, Jamar F, Pauwels S, Bilbao J I and Walrand S 2009 Yttrium-90 TOF PET scan demonstrates high-
resolution biodistribution after liver SIRT Eur. J. Nucl. Med. Mol. Imaging 36 1696
Liden K and Starfelt N 1955 Internal and external bremsstrahlung accompanying the beta rays of P32 Phys Rev. 97 419
Lim H, Dewaraja Y K and Fessler J A 2018 A PET reconstruction formulation that enforces non-negativity in projection space for bias
reduction in Y-90 imaging Phys. Medi. Biol. 63 035042
Maughan N M, EldI-Bremss M, Conti M, Knešaurek K, Faul D, Parikh P J, Fayad Z A and Laforest R 2016 Phantom study to determine
optimal PET reconstruction parameters for PET/MR imaging of 90Y microspheres following radioembolization Biomed. Phys. Eng.
Exp. 2 015009
Nikolopoulos D, Kottou S, Chatzisavvas N, Argyriou X, Vlamakis E, Yannakopoulos P and Louizi A 2013 A GATE simulation study of the
Siemens biograph DUO PET/CT system Open J. Radiol. 3 56
Nuyts J et al 2002 Reducing loss of image quality due to the attenuation artifact in uncorrected PET whole-body images J. Nucl. Med.
43 1054–62
Padia S A, Alessio A, Kwan S W, Lewis D H, Vaidya S and Minoshima S 2013 Comparison of positron emission tomography and Bremss
imaging to detect particle distribution in patients undergoing yttrium-90 radioembolization for large hepatocellular carcinomas or
associated portal vein thrombosis J. Vasc. Interv. Radiol. 24 1147–53
Pasciak A S, Bourgeois A C, McKinney J M, Chang T T, Osborne D R, Acuff S N and Bradley Y C 2014 Radioembolization and the dynamic
role of 90Y PET/CT Front. Oncol. 4 38
Rault E, Staelens S, Van Holen R, De Beenhouwer J and Vandenberghe S 2010 Fast simulation of yttrium-90 bremsstrahlung photons with
GATE Med. Phys. 37 2943–50
Rong X, Du Y, Ljungberg M, Rault E, Vandenberghe S and Frey E C 2012 Development and evaluation of an improved quantitative 90Y
bremsstrahlung SPECT method Med. Phys. 39 2346–58
Strydhorst J, Carlier T, Dieudonné A, Conti M and Buvat I 2016 A gate evaluation of the sources of error in quantitative 90Y PET Med. Phys.
43 5320–9
Surti S, Karp J S, Freifelder R and Liu F 2000 Optimizing the performance of a PET detector using discrete GSO crystals on a continuous
lightguide IEEE Trans. Nucl. Sci. 47 1030–6
Surti S, Kuhn A, Werner M E, Perkins A E, Kolthammer J and Karp J S 2007 Performance of Philips Gemini TF PET/CT scanner with special
consideration for its time-of-flight imaging capabilities J. Nucl. Med. 48 471–80
Takahashi A, Himuro K, Yamashita Y, Komiya I, Baba S and Sasaki M 2015 Monte Carlo simulation of PET and SPECT imaging of 90Y Med.
Phys. 42 1926–35

14
Phys. Med. Biol. 63 (2018) 075016 (15pp) S Walrand et al

Tapp K N, Lea W B, Johnson M S, Tann M, Fletcher J W and Hutchins G D 2014 The impact of image reconstruction bias on PET/CT 90Y
dosimetry after radioembolization J. Nucl. Med. 55 1452–8
Therriault-Proulx F, Beaulieu L, Archambault L and Beddar S 2013 On the nature of the light produced within PMMA optical light guides in
scintillation fiber-optic dosimetry Phys. Med. Biol. 58 2073
Van Elmbt L, Vandenberghe S, Walrand S, Pauwels S and Jamar F 2011 Comparison of yttrium-90 quantitative imaging by TOF and non-
TOF PET in a phantom of liver selective internal radiotherapy Phys. Med. Biol. 56 6759
Venkataramaiah P, Sanjeeviah H and Sanjeevaiah B 1980 Study of inner bremsstrahlung accompanying beta decay in 185W and 90Y J. Phys. G:
Nucl. Phys. 6 1443
Vunckx K, Zhou L, Matej S, Defrise M and Nuyts J 2010 Fisher information-based evaluation of image quality for time-of-flight PET IEEE
Trans. Med Imaging 29 311–21
Walrand S, Flux G D, Konijnenberg M W, Valkema R, Krenning E P, Lhommel R, Pauwels S and Jamar F 2011 Dosimetry of yttrium-labelled
radiopharmaceuticals for int. therapy: 86Y or 90Y imaging? Eur. J. Nucl. Med. Mol. Imaging 38 57–68
Walrand S, Hesse M, Renaud L and Jamar F 2015 The impact of image reconstruction bias on PET/CT 90Y dosimetry after
radioembolization J. Nucl. Med. 56 494–5
Wang W et al 2006 Systematic and distributed time-of-flight list mode PET reconstruction IEEE Nuclear Science Symp. Conf. Record vol 3 pp
1715–22
Werner M K, Brechtel K, Beyer T, Dittmann H, Pfannenberg C and Kupferschläger J 2010 PET/CT for the assessment and quantification of
90
Y biodistribution after selective internal radiotherapy (SIRT) of liver metastases Eur. J. Nucl. Med. Mol. Imaging 37 407–8
Willowson K P, Tapner M and Bailey D L 2015 A multicentre comparison of quantitative 90Y PET/CT for dosimetric purposes after
radioembolization with resin microspheres Eur. J. Nucl. Med. Mol. Imaging 42 1202–22
Yamamoto S, Horii H, Hurutani M, Matsumoto K and Senda M 2005 Investigation of single, random, and true counts from natural
radioactivity in LSO-based clinical PET Ann. Nucl. Med. 19 109–14

15