Fully Automated Segmentation of [ 18 F]FDG and [ 68 Ga]/[ 18 F]PSMA PET/CT Images via Data-Centric Deep Learning
Journal of Nuclear Medicine · 2026-11 · 67 (11) · 1843-1852
DOI: 10.2967/jnumed.126.272852 ↗Kurz gefasst
Multizentrische Studie in der Zeitschrift Journal of Nuclear Medicine, online erschienen am 30.09.2026. Verfasst von Manuel Pires und 44 weiteren Autor:innen. Beteiligte Einrichtung: QIMP Team, Medical University of Vienna, Vienna, Austria; manuel.pires@meduniwien.ac.at. Hauptthemen laut MEDLINE: Deep Learning, Positron Emission Tomography Computed Tomography, Fluorodeoxyglucose F18, Image Processing, Computer-Assisted. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar.
Abstract (Originalsprache)
The purpose of this study was to develop and validate lesion identification in oncologic nuclear imaging (LION), an open-source PET-only tumor segmentation pipeline for [18F]FDG and prostate-specific membrane antigen (PSMA)-targeted PET/CT, and to investigate how training data characteristics influence segmentation performance.
Methods
In this retrospective multicenter study, 5209 [18F]FDG PET/CT scans spanning 19 disease types and 2046 PSMA-targeted PET/CT scans were used to train PET-only segmentation models. Tumor segmentation incorporated organs with physiologic uptake as auxiliary classes to enable PET-only inference. Tumor occurrence maps (TOMs) quantified tumor spatial diversity across the training data. For [18F]FDG, disease-specific and mixed-disease models trained on progressively larger subsets were compared to test whether increasing spatial diversity improves generalization. Scanner-related domain shift was analyzed using DINOv2 embeddings. Models were evaluated on multicenter holdout cohorts (616 [18F]FDG scans across 4 diseases; 443 PSMA-targeted prostate cancer scans) and compared with 3 open-source tools.
Results
Organ context improved median Dice from 0.62 to 0.71 for [18F]FDG and from 0.75 to 0.83 for PSMA, on the complete holdout cohorts. Spatial diversity measured by TOMs was strongly associated with Dice (Spearman ρ = 0.80, P = 0.003). A mixed-disease model trained on 500 patients matched the performance of a lymphoma specialist model trained on 3031 cases. DINOv2 embeddings revealed scanner-induced domain shift between same-disease cohorts. LION achieved median Dice scores of 0.71 for [18F]FDG and 0.85 for PSMA and outperformed other open-source approaches on the model-comparison test set, which excluded the AutoPET test cases.
Conclusion
LION enables PET-only automated segmentation for [18F]FDG and PSMA-targeted PET. Training data composition, particularly spatial diversity quantified by TOMs, was strongly associated with segmentation performance.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- QIMP Team, Medical University of Vienna, Vienna, Austria; manuel.pires@meduniwien.ac.at.
- QIMP Team, Medical University of Vienna, Vienna, Austria.
- Santa Croce e Carle Hospital, Medical Physics, Cuneo, Italy.
- Department of Biomedical Imaging and Image-Guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Vienna, Austria.
- Division of Nuclear Medicine, Azienda Ospedaliero Universitaria Careggi, Florence, Italy.
- Medical University of Vienna, Vienna, Austria.
Zitieren
APA
Pires, M., Gutschmayer, S., Bergalla, E., Chauvie, S., Amereller, D., Abenavoli, E., Lazarevic, A., Wegrzyn, N., Richter, P., Tibu, I., Miller, K., Kersting, D., Kim, M. S., Dexl, J., Ingrisch, M., Cyran, C. C., Heimer, M., Kazazi, M., Callahan, J., …, & Shiyam Sundar, L. K. (2026). Fully Automated Segmentation of [ 18 F]FDG and [ 68 Ga]/[ 18 F]PSMA PET/CT Images via Data-Centric Deep Learning. Journal of Nuclear Medicine, 67(11), 1843-1852. https://doi.org/10.2967/jnumed.126.272852
Vancouver
Pires M, Gutschmayer S, Bergalla E, Chauvie S, Amereller D, Abenavoli E, et al. Fully Automated Segmentation of [ 18 F]FDG and [ 68 Ga]/[ 18 F]PSMA PET/CT Images via Data-Centric Deep Learning. Journal of Nuclear Medicine. 2026;67(11):1843-1852. doi:10.2967/jnumed.126.272852
BibTeX
@article{Pires2026,
author = {Pires, Manuel and Gutschmayer, Sebastian and Bergalla, Enrica and Chauvie, Stephane and Amereller, Dominik and Abenavoli, Elisabetta and Lazarevic, Aleksa and Wegrzyn, Natalia and Richter, Philip and Tibu, Iustin and Miller, Ken and Kersting, David and Kim, Moon-Sung and Dexl, Jakob and Ingrisch, Michael and Cyran, Clemens C and Heimer, Maurice and Kazazi, Mej and Callahan, Jason and Hicks, Rodney and Calderón, Eduardo and Benecke, Jonas and Schmidt, Fabian and Sciagra, Roberto and Auriac, Julie and Seban, Romain-David and Orlhac, Fanny and Buvat, Irene and Siegmund, Sophie Carina and Widjaja, Liam and Werner, Rudolf A and Santo, Giulia and Kronthaler, Ariane and Virgolini, Irene J and Abdelhafez, Yasser Gaber and Nalbant, Hande and Cheon, Miju and Badawi, Ramsey D and Nardo, Lorenzo and Li, Anting and Vehmanen, Ahti and Kemppainen, Jukka and Klén, Riku and Beyer, Thomas and Shiyam Sundar, Lalith Kumar},
title = {Fully Automated Segmentation of [ 18 F]FDG and [ 68 Ga]/[ 18 F]PSMA PET/CT Images via Data-Centric Deep Learning},
journal = {Journal of Nuclear Medicine},
year = {2026},
volume = {67},
number = {11},
pages = {1843--1852},
doi = {10.2967/jnumed.126.272852}
}
EUVIMED