% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Masood:1038864,
author = {Masood, Anum and Drzezga, Alexander and Elmenhorst, E. M.
and Foerges, Anna Linea and Lange, Denise and Hennecke, E.
and Baur, D. M. and Kroll, Tina and Neumaier, Bernd and
Aeschbach, D. and Bauer, Andreas and Landolt, H. P. and
Elmenhorst, David and Beer, Simone},
title = {{E}nhancing {U}ltra-{L}ow-{D}ose {B}rain {PET}/{MRI}
{I}mage {Q}uality {U}sing {D}eep {L}earning},
reportid = {FZJ-2025-01680},
year = {2024},
abstract = {Aim:We developed a deep learning model to enhance the image
quality of ultra-low dose brain positron emission tomography
(PET). Significantly reducing the injected dose not only
minimizes radiation risk in subjects but also provides
options for scanning protocols, allowing e.g. for more
follow-up studies. Methods:Our dataset comprises 196 brain
PET/magnetic resonance (MRI) images (MPRAGE) of healthy
volunteers of two sleep deprivation studies, scanned for 100
minutes each using the adenosine A1 receptor ligand
[F-18]CPFPX (mean injected activity: 179.5 MBq, 10 MBq SD).
From our dataset, low-dose images with Dose Reduction
Factors (DRF) of 4, 10, 20, 50, and 100 were created by
modifying the acquired listmode data, corresponding to
injected activities of 45, 18, 9, 3.5 and 1.8 MBq.We first
modified and trained a U-Net model using a dataset of 500
total-body [F-18]FDG PET/MRI images with the same DRFs as
well as the corresponding full-dose images [1]. We then used
a Transfer Learning approach to train the model with our
data. 8 subjects were hold out for testing our model. For
evaluation, we used peak signal-to-noise ratio (PSNR),
Structural Similarity (SSIM), Contrast to Noise Ratio (CNR)
and Normalized Mean Squared Error (NMSE).Results:Our model
can synthesize enhanced low dose images that have
considerably reduced noise compared to the low-dose PET
image and resemble the standard-dose image. We achieved
improved metrics compared to the original U-Net model with
average PSNR of 28.02 (U-Net: 21.23), SSIM of 0.81 (U-Net:
0.53), CNR of 0.72 (U-Net: 0.61) and NMSE of 0.33 (U-Net:
0.67).Conclusions:Our results highlight the potential of our
model for enhancing low-dose brain PET/MRI images, allowing
for scans with reduced injected dose, shorter scan times or
more follow-up studies. References:[1] S Xue et al., Eur J
Nucl Med Mol I 49 (6), 1843 (2022).},
month = {Apr},
date = {2024-04-10},
organization = {62. Jahrestagung der Deutschen
Gesellschaft für Nuklearmedizin (DGN
e. V.), Leipzig (Germany), 10 Apr 2024
- 13 Apr 2024},
subtyp = {After Call},
cin = {INM-2},
cid = {I:(DE-Juel1)INM-2-20090406},
pnm = {5254 - Neuroscientific Data Analytics and AI (POF4-525)},
pid = {G:(DE-HGF)POF4-5254},
typ = {PUB:(DE-HGF)6},
url = {https://juser.fz-juelich.de/record/1038864},
}