ExOTics
Exploring the potential of Optimal Transport for Monte Carlo methods in Medical physics
Embedded Files
PI: Hong-Phuong DANG
Duration: 2026 - 2029
Funding: ANR JCJC grant number ANR-25-CE45-3443-01
Résumé : Ce projet se concentre sur l'intégration de la méthode du transport optimal (TO) dans les simulations de Monte Carlo (MC) en physique médicale, avec un accent particulier sur la dosimétrie, qui calcule la distribution de dose dans le patient. Les simulations MC sont couramment utilisées dans ce domaine, mais elles sont notoirement lentes en raison de l'estimation itérative et stochastique des fonctions de densité de probabilité. Diverses techniques de réduction de la variance (VRT) ont été introduites pour accélérer la MC sans biaiser la simulation. Malgré cela, des limites subsistent en termes de précision et d'incertitude. Le TO offre un cadre mathématique puissant pour mesurer la "distance" entre deux distributions, ce qui en fait un outil prometteur pour améliorer la précision et la fidélité des cartes de dose dans les simulations de MC. Le projet propose trois volets : (1) Intégrer le TO dans les phases d'entraînement de l'apprentissage profond pour prédire les doses, (2) Utiliser le TO comme nouvelle métrique pour évaluer les nouvelles méthodes de simulation de MC et (3) Utiliser le TO pour améliorer le transfert d'apprentissage (TL) et ainsi étendre les capacités des simulations de MC. Ces volets visent à améliorer la précision, la robustesse et l'efficacité des prédictions de dose, ce qui pourrait révolutionner l'approche de l'estimation et de l'analyse des doses, offrant ainsi un potentiel considérable pour la planification du traitement des patients en radiothérapie. Le projet vise spécifiquement à personnaliser et à améliorer la dosimétrie pour la thérapie ciblée par radionucléides du cancer de la prostate.
Summary : This project focuses on the integration of the optimal transport (OT) method into Monte Carlo (MC) simulations in medical physics, with a particular emphasis on dosimetry, which calculates the dose distribution in the patient. MC simulations are commonly used in this field, but they are notoriously slow due to the iterative and stochastic estimation of probability density functions. Various variance reduction techniques (VRT) have been introduced to accelerate MC without biasing the simulation. Despite this, limitations persist in terms of precision and uncertainty. OT provides a powerful mathematical framework for measuring the "distance" between two distributions, making it a promising tool for improving the accuracy and fidelity of dose maps in MC simulations. The project proposes three components: (1) Integrating OT into deep learning training phases to predict doses, (2) Using OT as a new metric to evaluate novel MC simulation methods, and (3) Using OT to enhance transfer learning (TL) and thereby extend MC simulation capabilities. These components aim to improve the precision, robustness, and efficiency of dose predictions, which could revolutionize the approach to dose estimation and analysis, offering significant potential for personalized treatment planning in radiotherapy, specifically for targeted radionuclide therapy of prostate cancer.
Summary : This project focuses on the integration of the optimal transport (OT) method into Monte Carlo (MC) simulations in medical physics, with a particular emphasis on dosimetry, which calculates the dose distribution in the patient. MC simulations are commonly used in this field, but they are notoriously slow due to the iterative and stochastic estimation of probability density functions. Various variance reduction techniques (VRT) have been introduced to accelerate MC without biasing the simulation. Despite this, limitations persist in terms of precision and uncertainty. OT provides a powerful mathematical framework for measuring the "distance" between two distributions, making it a promising tool for improving the accuracy and fidelity of dose maps in MC simulations. The project proposes three components: (1) Integrating OT into deep learning training phases to predict doses, (2) Using OT as a new metric to evaluate novel MC simulation methods, and (3) Using OT to enhance transfer learning (TL) and thereby extend MC simulation capabilities. These components aim to improve the precision, robustness, and efficiency of dose predictions, which could revolutionize the approach to dose estimation and analysis, offering significant potential for personalized treatment planning in radiotherapy, specifically for targeted radionuclide therapy of prostate cancer.
Job - M2 Internship (with PhD prospects) : Optimal Transport for Dose Estimation in Cancer Radiotherapy
[The offer has been filled]
Detailed subject (en+fr) here.
- Supervision: Hong-Phuong Dang and Clément Elvira
- Location: CentraleSupélec (Rennes campus)
- Duration: 4 to 6 months (start between February–April depending on availability)
- Application: Send us CV + cover letter + transcripts from the last three years
Page updated
Google Sites
Report abuse