Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Comparison of Detectability Between Digital LYSO and BGO-Based PET/CT Scanners: A Phantom Study

Sohlman, Marielle (2026) MSFT02 20262
Medical Physics Programme
Abstract
Background and Aim: A new PET/CT scanner using BGO crystals was recently introduced
in the clinic. Unlike scanners with LSO/LYSO crystals, this scanner does not include the time-
of-flight (TOF) property, due to low timing resolution of BGO crystals. To compensate for the
lack of TOF, the reconstructed images can be filtered with an AI model, called Precision DL
(PDL), which has been trained on images with and without TOF. It still remains unknown
how well the AI model can enhance small lesions in a patient compared to a system with TOF,
especially in more challenging conditions, such as larger patients. The aim of this project is to
compare detectability of small lesions between a PET/CT system with true TOF (GE Health-
Care’s... (More)
Background and Aim: A new PET/CT scanner using BGO crystals was recently introduced
in the clinic. Unlike scanners with LSO/LYSO crystals, this scanner does not include the time-
of-flight (TOF) property, due to low timing resolution of BGO crystals. To compensate for the
lack of TOF, the reconstructed images can be filtered with an AI model, called Precision DL
(PDL), which has been trained on images with and without TOF. It still remains unknown
how well the AI model can enhance small lesions in a patient compared to a system with TOF,
especially in more challenging conditions, such as larger patients. The aim of this project is to
compare detectability of small lesions between a PET/CT system with true TOF (GE Health-
Care’s Discovery MI) and a PET/CT system with AI based TOF (GE HealthCare’s OMNI
Legend).

Materials and Methods: To compare detectability of small lesions between the two sys-
tems, a Jaszczak phantom and a 3D printed XL shell to Jaszczak were used. The XL phantom
was divided into two cases, the shell filled with and without background activity. Six hot spheres
in different sizes were placed in Jaszczak, in a sphere-to-background ratio (SBR) of 10:1, where
the spheres had an activity concentration of 12 kBq/ml at start of scan. The PET radiophar-
maceutical used was 18F-FDG and the phantoms were scanned sequentially on the two PET
systems (DMI and OMNI). Images were acquired in listmode and rebinned into ten 90-s frames
per phantom and PET-system, with one bed position and matrix sizes of 256×256 and 384×384.
Phantom images were reconstructed using OSEM + PSF together with a post filter of 3 mm
and BSREM with β value of 500. The reconstructions were made with and without TOF for the
DMI scanner, whereas the BSREM reconstructions for the OMNI scanner were combined with
the three PDL levels. Recovery coefficients (RC) were calculated from peak values in the spheres
and the calculated activity concentrations in the spheres at the start of scan, and signal-to-noise
ratio (SNR) was determined in the background for each phantom. A peak-to-peak (P2P) value
was calculated as the ratio of the peak in a sphere to the peak in the background and was
evaluated as a detectability value. Non-blinded readers did a visual evaluation of a sample of
the images to visually estimate the lesion detectability.

Results: TheresultsshowedthatbothreconstructionmethodswithTOF(OSEMandBSREM)
with the DMI scanner gave higher RC than all reconstructions from the OMNI scanner across
all three phantom cases. The TOF effect was clearly shown for the DMI scanner, while the
PDL filters seemed to give little to none effect on the recovery. A decrease in the background
SNR was seen in reconstructions with TOF compared to those without TOF, while the PDL
models had an increasing effect on the SNR. The P2P values for the DMI tend to be higher
for reconstructions with TOF compared to those without. For the Jaszczak phantom, the DMI
scanner reaches higher P2P values than the OMNI scanner, while the opposite trend is seen for
the XL phantom. In general, higher P2P values were reached for Jaszczak compared to the XL
phantom, which means that more spheres were thought to be seen in the images for the smaller
phantom. This result was confirmed through the visual evaluation, which also resulted in higher
means of detected spheres for the BGO system. The results of the P2P values for both systems
also showed a large spread of the data points within the ten time frames of the XL phantom.

Conclusion: The P2P values together with the visual evaluation conclude that a smaller phan-
tom has greater detectability than a larger one. The P2P values show that the benefit of true
TOF is greater for the Jaszczak phantom than the XL phantom, which contradicts the theory.
The benefits of the BGO system were also shown to be larger in the phantom case with no
activity in the XL shell, which is the most patient like phantom case. Further studies are needed
to confirm the effect of true TOF versus the sensitivity of the BGO crystals for larger patients. (Less)
Popular Abstract (Swedish)
Positronsemissionstomografi (PET) är en bildtagningsmetod där man med hjälp av radioaktiva
läkemedel undersöker funktion i kroppens organ. En vanlig diagnos som undersöks med PET är
spridning av cancer, där det oftast använda radioaktiva läkemedlet är fluorodeoxyglukos (FDG),
även kallat radioaktivt socker. Eftersom cancerceller kräver mer socker än celler i normalvävnad
tas FDG i större utsträckning upp i cancercellerna i jämförelse med de normala cellerna. Den
radioaktiva delen i läkemedlet är fluor-18, vilken är en fluorisotop som sönderfaller med positron-
strålning. Från positronen som skickas ut från sönderfallet skapas två fotoner som sänds ut i
en 180° vinkel. Från många positronsönderfall skapas alltså lika många fotonpar... (More)
Positronsemissionstomografi (PET) är en bildtagningsmetod där man med hjälp av radioaktiva
läkemedel undersöker funktion i kroppens organ. En vanlig diagnos som undersöks med PET är
spridning av cancer, där det oftast använda radioaktiva läkemedlet är fluorodeoxyglukos (FDG),
även kallat radioaktivt socker. Eftersom cancerceller kräver mer socker än celler i normalvävnad
tas FDG i större utsträckning upp i cancercellerna i jämförelse med de normala cellerna. Den
radioaktiva delen i läkemedlet är fluor-18, vilken är en fluorisotop som sönderfaller med positron-
strålning. Från positronen som skickas ut från sönderfallet skapas två fotoner som sänds ut i
en 180° vinkel. Från många positronsönderfall skapas alltså lika många fotonpar och det är
dessa fotonpar som undersökningen går ut på att samla in. Insamlingen av fotonparen görs med
en PET kamera, där fotonerna bromsas upp och avger energi som sedan kan utläsas som en
elektrisk signal. Från den insamlade signalen skapas en avbildning av patienten i kameran, där
områden med högt upptag av läkemedlet syns tydligare än områden med lågt upptag.

Materialet i PET-kameran som bromsar upp fotonerna är en kristall, med specifika egenskaper
som gynnar den återskapade bilden på olika sätt. Kristallen ska önskvärt kunna samla in så
många fotonpar som möjligt samt vara bra på att omvandla fotonparet till en optisk signal.
Något som forskningen har strävat efter är att kristallen ska vara snabb med att sända ut den
optiska signalen från att fotonen har avgett sin energi i kristallen. Om kristallen är tillräckligt
snabb i detta steg är det möjligt att mäta tidsskillnaden mellan de två fotonerna i fotonparet.
I sin tur ger detta en mer säker position i bilden, på vart positronsönderfallet skedde i kroppen.
Möjligheten att kunna mäta tidsskillnaden mellan fotonerna kallas time-of-flight (TOF). Ett
PET system med TOF har därför bättre förutsättningar för att noggrannare kunna markera ut
var exempelvis små tumörer är lokaliserade i jämförelse med ett PET system utan TOF. Nyligen
installerades en PET kamera i Lund (OMNI Legend) där TOF inte är möjligt, då kristallerna är
för långsamma. I stället har denna kristall andra fördelaktiga egenskaper, så som god förmåga
att fånga fotonerna samt att den är relativt billig. I kompensation till förlusten av TOF har
företaget som skapade kameran tränat en AI modell som ska efterlikna sann TOF, och på så vis
kan kristallens andra egenskaper utnyttjas utan att resultatet ska försämras av de långsamma
kristallerna.

I detta arbete undersöktes skillnaden i hur väl små lesioner, eller tumörer, kan uppfattas eller
detekteras mellan de två olika PET systemen (OMNI Legend och Discovery MI) för att ta reda
på om de presterar lika bra under lika förutsättningar. En viktig parameter i projektet var
att undersöka hur patientstorleken påverkar detekterbarheten, då lesioner är svårare att korrekt
avbilda i mer utmanande patientgeometrier, så som i större patienter. Som representation av
patienter användes fantom, vilket är en form av vattentät plastbehållare. Ett mindre cylindriskt
fantom och ett 3D-printat skal till det mindre fantomet användes för att representera olika pa-
tientstorlekar. I det mindre fantomet placerades sfärer i olika storlekar, som skulle motsvara
olika stora lesioner. Fantomen och sfärerna fylldes sedan med utspädd 18F-FDG, där sfärerna
hade tio gånger högre koncentration än de stora volymerna, och placerades sedan i de olika
kamerorna för avbildning.

Resultatet från mätningarna visar på att detekterbarheten av sfärerna minskar med ökad fan-
tomstorlek och att mindre sfärer är svårare att se, trots samma aktivitetskoncentration. Detek-
terbarheten hos det mindre fantomet visade ingen större skillnad mellan kamerorna, medan för
det större fantomet visade sig detekterbarheten vara bättre på kameran utan sann TOF (OMNI
Legend). Detta är enligt teorin inte väntat, då fördelen av sann TOF förväntas vara större för
större patient. Utvärderingen av de applicerade AI modellerna visade ingen större förbättring i
jämförelse med datan utan AI modell. (Less)
Please use this url to cite or link to this publication:
author
Sohlman, Marielle
supervisor
organization
course
MSFT02 20262
year
type
H2 - Master's Degree (Two Years)
subject
language
English
id
9248575
date added to LUP
2026-08-18 11:46:07
date last changed
2026-08-18 11:46:07
@misc{9248575,
  abstract     = {{Background and Aim: A new PET/CT scanner using BGO crystals was recently introduced
in the clinic. Unlike scanners with LSO/LYSO crystals, this scanner does not include the time-
of-flight (TOF) property, due to low timing resolution of BGO crystals. To compensate for the
lack of TOF, the reconstructed images can be filtered with an AI model, called Precision DL
(PDL), which has been trained on images with and without TOF. It still remains unknown
how well the AI model can enhance small lesions in a patient compared to a system with TOF,
especially in more challenging conditions, such as larger patients. The aim of this project is to
compare detectability of small lesions between a PET/CT system with true TOF (GE Health-
Care’s Discovery MI) and a PET/CT system with AI based TOF (GE HealthCare’s OMNI
Legend).

Materials and Methods: To compare detectability of small lesions between the two sys-
tems, a Jaszczak phantom and a 3D printed XL shell to Jaszczak were used. The XL phantom
was divided into two cases, the shell filled with and without background activity. Six hot spheres
in different sizes were placed in Jaszczak, in a sphere-to-background ratio (SBR) of 10:1, where
the spheres had an activity concentration of 12 kBq/ml at start of scan. The PET radiophar-
maceutical used was 18F-FDG and the phantoms were scanned sequentially on the two PET
systems (DMI and OMNI). Images were acquired in listmode and rebinned into ten 90-s frames
per phantom and PET-system, with one bed position and matrix sizes of 256×256 and 384×384.
Phantom images were reconstructed using OSEM + PSF together with a post filter of 3 mm
and BSREM with β value of 500. The reconstructions were made with and without TOF for the
DMI scanner, whereas the BSREM reconstructions for the OMNI scanner were combined with
the three PDL levels. Recovery coefficients (RC) were calculated from peak values in the spheres
and the calculated activity concentrations in the spheres at the start of scan, and signal-to-noise
ratio (SNR) was determined in the background for each phantom. A peak-to-peak (P2P) value
was calculated as the ratio of the peak in a sphere to the peak in the background and was
evaluated as a detectability value. Non-blinded readers did a visual evaluation of a sample of
the images to visually estimate the lesion detectability.

Results: TheresultsshowedthatbothreconstructionmethodswithTOF(OSEMandBSREM)
with the DMI scanner gave higher RC than all reconstructions from the OMNI scanner across
all three phantom cases. The TOF effect was clearly shown for the DMI scanner, while the
PDL filters seemed to give little to none effect on the recovery. A decrease in the background
SNR was seen in reconstructions with TOF compared to those without TOF, while the PDL
models had an increasing effect on the SNR. The P2P values for the DMI tend to be higher
for reconstructions with TOF compared to those without. For the Jaszczak phantom, the DMI
scanner reaches higher P2P values than the OMNI scanner, while the opposite trend is seen for
the XL phantom. In general, higher P2P values were reached for Jaszczak compared to the XL
phantom, which means that more spheres were thought to be seen in the images for the smaller
phantom. This result was confirmed through the visual evaluation, which also resulted in higher
means of detected spheres for the BGO system. The results of the P2P values for both systems
also showed a large spread of the data points within the ten time frames of the XL phantom.

Conclusion: The P2P values together with the visual evaluation conclude that a smaller phan-
tom has greater detectability than a larger one. The P2P values show that the benefit of true
TOF is greater for the Jaszczak phantom than the XL phantom, which contradicts the theory.
The benefits of the BGO system were also shown to be larger in the phantom case with no
activity in the XL shell, which is the most patient like phantom case. Further studies are needed
to confirm the effect of true TOF versus the sensitivity of the BGO crystals for larger patients.}},
  author       = {{Sohlman, Marielle}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Comparison of Detectability Between Digital LYSO and BGO-Based PET/CT Scanners: A Phantom Study}},
  year         = {{2026}},
}