Skip to content
DraconiticResearch
Search Sources API
203,203 studies31 sourcesUpdated 56 min ago
Peer-reviewedCC BY 4.0Journal of Nuclear Medicine

Fully Automated Segmentation of [18F]FDG and [68Ga]/[18F]PSMA PET/CT Images via Data-Centric Deep Learning

  • Manuel Pires1
  • Sebastian Gutschmayer1
  • Enrica Bergalla2
  • Stephane Chauvie2
  • Dominik Amereller3
  • Elisabetta Abenavoli4
  • Aleksa Lazarevic3
  • Natalia Wegrzyn5
  • Philip Richter5
  • Iustin Tibu5
  • Ken Miller3
  • David Kersting6,7
33 more authors
  • Moon-Sung Kim8
  • Jakob Dexl9,10
  • Michael Ingrisch9,10,11
  • Clemens C. Cyran9
  • Maurice Heimer9
  • Mej Kazazi9
  • Jason Callahan12
  • Rodney Hicks12,13
  • Eduardo Calderón14
  • Jonas Benecke14
  • Fabian Schmidt14,15
  • Roberto Sciagra4
  • Julie Auriac16
  • Romain-David Seban16
  • Fanny Orlhac16
  • Irene Buvat16
  • Sophie Carina Siegmund17
  • Liam Widjaja17
  • Rudolf A. Werner17
  • Giulia Santo18
  • Ariane Kronthaler18
  • Irene J. Virgolini18
  • Yasser Gaber Abdelhafez19,20
  • Hande Nalbant19
  • Miju Cheon19,21
  • Ramsey D. Badawi19
  • Lorenzo Nardo19
  • Anting Li22
  • Ahti Vehmanen22,23
  • Jukka Kemppainen22,23
  • Riku Klén22
  • Thomas Beyer1
  • Lalith Kumar Shiyam Sundar9
23 affiliations
  1. 1QIMP Team, Medical University of Vienna, Vienna, Austria
  2. 2Santa Croce e Carle Hospital, Medical Physics, Cuneo, Italy
  3. 3Department of Biomedical Imaging and Image-Guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Vienna, Austria
  4. 4Division of Nuclear Medicine, Azienda Ospedaliero Universitaria Careggi, Florence, Italy
  5. 5Medical University of Vienna, Vienna, Austria
  6. 6Department of Nuclear Medicine, West German Cancer Center, University Hospital Essen, University of Duisburg-Essen, Essen, Germany
  7. 7German Cancer Consortium, Partner Site, University Hospital Essen, Essen, Germany
  8. 8Institute for AI in Medicine, University Hospital Essen, Essen, Germany
  9. 9Department of Radiology, LMU University Hospital, LMU Munich, Munich, Germany
  10. 10Munich Center for Machine Learning, Munich, Germany
  11. 11relAI–Konrad Zuse School of Excellence in Reliable AI, Munich, Germany
  12. 12Melbourne Theranostic Innovation Centre, North Melbourne, Victoria, Australia
  13. 13St Vincent’s Hospital, Department of Medicine, The University of Melbourne, Fitzroy, Victoria, Australia
  14. 14Department of Nuclear Medicine and Clinical Molecular Imaging, University Hospital Tuebingen, Tuebingen, Germany
  15. 15Werner Siemens Imaging Center, Department of Preclinical Imaging and Radiopharmacy, Eberhard Karls University Tuebingen, Tuebingen, Germany
  16. 16IRIS, Institut Curie, Inserm, CNRS, Université Versailles Saint-Quentin, Universite PSL, Orsay, France
  17. 17Department of Nuclear Medicine, LMU University Hospital, LMU Munich, Munich, Germany
  18. 18Department of Nuclear Medicine, Medical University of Innsbruck, Innsbruck, Austria
  19. 19Department of Radiology, University of California, Davis, Davis, California
  20. 20Nuclear Medicine Unit, South Egypt Cancer Institute, Assiut University, Assiut, Egypt
  21. 21Department of Nuclear Medicine, Veterans Health Service Medical Center, Seoul, Korea
  22. 22Turku PET Centre, University of Turku and Turku University Hospital, Turku, Finland; and
  23. 23Department of Clinical Physiology and Nuclear Medicine, Turku University Hospital, Turku, Finland
Read full textPDFDOICite

Abstract

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.

Related

  1. Deep Learning-based Framework for Rapid, Precise and Fully Automated Localization of Cochlea Landmarks in 3D CT Images
    Y. Wang, A. Kral, T. Lenarz +1
    medRxivPreprintCC BY-NC-NDFull textMedical imaging & radiologymedRxiv
  2. Fully automatic airway segmentation in CT images via dual-stage neural network with topology-constrained centerline completion
    Harmony Yuan, Yuan Cheng, Chengxi Yan +4
    Quantitative imaging in medicine and surgeryPeer-reviewedCC BY-NC-NDFull textMedical imaging & radiologyEurope PMC
  3. Automated corneal endothelial cell detection and segmentation in specular microscopy images using Attention and Faster U-Net
    Ashwini Ramchandra Doke, Suja Palaniswamy
    Scientific ReportsPeer-reviewedCC BY-NC-ND 4.0Full textMedical imaging & radiologyCrossref
  4. Fully automated diaphragm segmentation on routine 2D-cine and 3D thoracic MRI: multireader validation
    Samia Boussouar, Alexiane Pasquier, Damien Vaurs +16
    European radiology experimentalPeer-reviewedCC BYFull textSystems & synthetic biologyMedical imaging & radiologyEurope PMCHAL
  5. Fully automated AI-based quantification of tumour volume on PSMA PET-CT images is associated with overall survival in prostate cancer.
    Johan Wittgren, Jacob Ingvar, Olof Enqvist +5
    Clinical physiology and functional imagingPeer-reviewedCC BYFull textMedical imaging & radiologyEurope PMCPMC

Cite

BibTeXRIS
BibTeX
@article{pires2026fully,
  title = {{Fully Automated Segmentation of [18F]FDG and [68Ga]/[18F]PSMA PET/CT Images via Data-Centric Deep Learning}},
  author = {Manuel Pires and Sebastian Gutschmayer and Enrica Bergalla and Stephane Chauvie and Dominik Amereller and Elisabetta Abenavoli and Aleksa Lazarevic and Natalia Wegrzyn and Philip Richter and Iustin Tibu and Ken Miller and David Kersting and Moon-Sung Kim and Jakob Dexl and Michael Ingrisch and Clemens C. Cyran and Maurice Heimer and Mej Kazazi and Jason Callahan and Rodney Hicks and Eduardo Calderón and Jonas Benecke and Fabian Schmidt and Roberto Sciagra and Julie Auriac and Romain-David Seban and Fanny Orlhac and Irene Buvat and Sophie Carina Siegmund and Liam Widjaja and Rudolf A. Werner and Giulia Santo and Ariane Kronthaler and Irene J. Virgolini and Yasser Gaber Abdelhafez and Hande Nalbant and Miju Cheon and Ramsey D. Badawi and Lorenzo Nardo and Anting Li and Ahti Vehmanen and Jukka Kemppainen and Riku Klén and Thomas Beyer and Lalith Kumar Shiyam Sundar},
  journal = {Journal of Nuclear Medicine},
  year = {2026},
  month = nov,
  doi = {10.2967/jnumed.126.272852},
  url = {https://doi.org/10.2967/jnumed.126.272852},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - Fully Automated Segmentation of [18F]FDG and [68Ga]/[18F]PSMA PET/CT Images via Data-Centric Deep Learning
AU  - Manuel Pires
AU  - Sebastian Gutschmayer
AU  - Enrica Bergalla
AU  - Stephane Chauvie
AU  - Dominik Amereller
AU  - Elisabetta Abenavoli
AU  - Aleksa Lazarevic
AU  - Natalia Wegrzyn
AU  - Philip Richter
AU  - Iustin Tibu
AU  - Ken Miller
AU  - David Kersting
AU  - Moon-Sung Kim
AU  - Jakob Dexl
AU  - Michael Ingrisch
AU  - Clemens C. Cyran
AU  - Maurice Heimer
AU  - Mej Kazazi
AU  - Jason Callahan
AU  - Rodney Hicks
AU  - Eduardo Calderón
AU  - Jonas Benecke
AU  - Fabian Schmidt
AU  - Roberto Sciagra
AU  - Julie Auriac
AU  - Romain-David Seban
AU  - Fanny Orlhac
AU  - Irene Buvat
AU  - Sophie Carina Siegmund
AU  - Liam Widjaja
AU  - Rudolf A. Werner
AU  - Giulia Santo
AU  - Ariane Kronthaler
AU  - Irene J. Virgolini
AU  - Yasser Gaber Abdelhafez
AU  - Hande Nalbant
AU  - Miju Cheon
AU  - Ramsey D. Badawi
AU  - Lorenzo Nardo
AU  - Anting Li
AU  - Ahti Vehmanen
AU  - Jukka Kemppainen
AU  - Riku Klén
AU  - Thomas Beyer
AU  - Lalith Kumar Shiyam Sundar
PY  - 2026
DA  - 2026/11/03
JO  - Journal of Nuclear Medicine
DO  - 10.2967/jnumed.126.272852
UR  - https://doi.org/10.2967/jnumed.126.272852
AB  - 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.
SN  - 0161-5505
SN  - 1535-5667
ER  -