EUVIMEDEUVIMED entdecken
Crossref

Hybrid GPU Monte Carlo software toolkit for fast and accurate cone‐beam CT simulation

Hybrides GPU-Monte-Carlo-Software-Toolkit für schnelle und genaue Cone-Beam-CT-Simulation

Jericho O'Connell, Matthew Jacobson, Thomas Harris, Raphael Bruegger, Dianne Ferguson, Rony Fueglistaller, Yue-Houng Hu, Daniel Morf, Vera Birrer, Pablo Corral Arroyo, Mathias Lehmann, Marios Myronakis und 1 weitere

Medical Physics · 2025-08 · 52 (8)

DOI: 10.1002/mp.18011 ↗

Auf einen Blick

GECCO ist ein hybrides GPU-Monte-Carlo-Toolkit für die CBCT-Simulation. In Validierungen zeigte es schnelle Laufzeiten und Abweichungen von 5,7 HU bis 16 HU.

Fragestellung

Kann GECCO CBCT-Simulationen schneller und zugleich genau genug für klinische Anwendungen berechnen?

Methode

GECCO mit 2 Anwendungen: analytische CT-zu-CBCT-Umsetzung und RTIS. Validierung an XCAT-Phantom, virtuellem Catphan und CNS-Patient:in gegen Referenzsimulationen bzw. OBI-CBCT.

Ergebnisse

GECCO simulierte 915 Full-Resolution-Projektionen eines CNS-Falls in 63 min auf einer NVIDIA 4090 GPU. RMSE: 5,7 HU am Catphan, 14 HU für GECCO und 16 HU für RTIS im CNS-Profil.

Schlussfolgerung der Autor:innen

Die Autor:innen stellen GECCO als Toolkit für hochwertige Machine-Learning-Datensätze und RTIS für die Echtzeit-Auswahl von Bildgebungsverfahren vor.

Kernaussagen

  • 915 Full-Resolution-Projektionen eines CNS-Falls in 63 min simuliert
  • RMSE 5,7 HU am Catphan; 14 HU GECCO und 16 HU RTIS im CNS-Profil
  • RTIS mit 917 ms Initialisierung und 36 ms Updatezeit

Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.

Kurz gefasst

Fachartikel in der Zeitschrift Medical Physics, online erschienen am 08.08.2025. Verfasst von Jericho O'Connell und 12 weiteren Autor:innen. Beteiligte Einrichtung: Dana Farber Cancer Institute, Brigham and Women's Hospital, and Harvard Medical School, Boston, Massachusetts, USA. Hauptthemen laut MEDLINE: Cone-Beam Computed Tomography, Software, Monte Carlo Method, Computer Graphics. Die Arbeit nennt 48 Literaturangaben. Gefördert von: NCI NIH HHS, National Cancer Institute.

Auf einen BlickAbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

With the ever‐increasing dependence on cone‐beam CT (CBCT) for clinical decision‐making, faster and more accurate CBCT simulation tools are necessary to develop the next generation of CBCT techniques and algorithms.

Purpose

GECCO, a hybrid GPU‐Monte Carlo (MC) software toolkit was built for the dual purpose of increasing both speed and accuracy in CBCT simulation. GECCO contains two applications: The GECCO application, a rapid converter of CT to CBCT, and the real‐time image simulation (RTIS) application, a purely analytical approach that updates CBCT images in real‐time according to acquisition technique.

Method

The GECCO application calculates primary projections analytically and scatter with GPU‐MC. The analytical model is a faster, less memory‐intensive update to the Fastcat simulation toolkit, combining mono‐energetic projections into a polyenergetic projection. Scatter is calculated using a modified version of GGEMs GPU‐MC, denoised, and interpolated to reduce the number of MC histories needed for a full simulation. Conversely, the RTIS application is solely analytical and maximizes speed, only simulating a single CT slice and storing partial results to enable dynamic updating of the simulation. Validation compared GECCO simulations of an XCAT phantom, a virtual Catphan phantom, and an elementally decomposed planning CT of a central nervous system (CNS) patient to GGEMs MC projections, TrueBeam CBCT of a Catphan, and a registered CNS patient CBCT, respectively.

Results

The GECCO application took 63 min to simulate 915 full‐resolution projection CNS patient CBCT on an NVIDIA 4090 GPU. The virtual Catphan sensitometry module showed an RMSE of 5.7 HU compared to the TrueBeam on‐board imager (OBI). Likewise, RTIS had an initialization time of 917 ms and an update time of 36 ms, three orders of magnitude faster than any other simulation application. Profiles through the CNS patient showed 14 and 16 HU RMSE as compared to OBI for GECCO and RTIS, respectively.

Conclusion

We introduce the GECCO software toolkit, containing the GECCO application for producing high‐quality machine learning datasets and the RTIS application for real‐time selection of imaging techniques for radiation therapists.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Cone-Beam Computed TomographySoftwareMonte Carlo MethodComputer GraphicsTime FactorsImage Processing, Computer-AssistedHumansPhantoms, Imaging

Stichwörter der Autor:innen

CBCThybrid Monte Carlomachine learning

Autor:innen und Institutionen

  • Dana Farber Cancer Institute, Brigham and Women's Hospital, and Harvard Medical School, Boston, Massachusetts, USA.
  • Varian Medical Systems Imaging Laboratory, Baden, Switzerland.

Förderung

  • NCI NIH HHS
  • National Cancer Institute

Zitieren

APA

O'Connell, J., Jacobson, M., Harris, T., Bruegger, R., Ferguson, D., Fueglistaller, R., Hu, Y. H., Morf, D., Birrer, V., Arroyo, P. C., Lehmann, M., Myronakis, M., & Berbeco, R. (2025). Hybrid GPU Monte Carlo software toolkit for fast and accurate cone‐beam CT simulation. Medical Physics, 52(8). https://doi.org/10.1002/mp.18011

Vancouver

O'Connell J, Jacobson M, Harris T, Bruegger R, Ferguson D, Fueglistaller R, et al. Hybrid GPU Monte Carlo software toolkit for fast and accurate cone‐beam CT simulation. Medical Physics. 2025;52(8). doi:10.1002/mp.18011

BibTeX
@article{OConnell2025,
  author = {O'Connell, Jericho and Jacobson, Matthew and Harris, Thomas and Bruegger, Raphael and Ferguson, Dianne and Fueglistaller, Rony and Hu, Yue-Houng and Morf, Daniel and Birrer, Vera and Arroyo, Pablo Corral and Lehmann, Mathias and Myronakis, Marios and Berbeco, Ross},
  title = {Hybrid GPU Monte Carlo software toolkit for fast and accurate cone‐beam CT simulation},
  journal = {Medical Physics},
  year = {2025},
  volume = {52},
  number = {8},
  doi = {10.1002/mp.18011}
}
Kostenlos starten →