Bayesian contributions to radiation dose estimation in biological retrospective dosimetry. - IRSN - Institut de radioprotection et de sûreté nucléaire Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

Bayesian contributions to radiation dose estimation in biological retrospective dosimetry.

Résumé

Scoring of dicentric chromosome aberrations in peripheral blood lymphocytes is considered to be the "gold-standard" biological method to estimate the radiation dose received by individuals after proven or suspected radiation exposure. On the one hand, dose estimation is highly relevant to optimize patient-centered care and predict the health consequences of proven radiation exposure. In this context, two main questions arise: 1) "Given the number of dicentrics observed in some blood lymphocytes of a given individual, what were the estimated absorbed dose and its associated uncertainty?" and 2) "Was the radiation exposure total or partial?" On the other hand, dose estimation from dicentric counts can also be crucial to clarify unclear radiation exposure scenarios. In this context, one important additional question is: 3) "Given the number of dicentrics observed, was the individual really exposed to ionizing radiation?" Frequentist statistical approaches are commonly used to answer the above questions that are then formalized as hypotheses testing and inverse regression problems. Bayesian statistical approaches have also been recently proposed but, up to our knowledge, they do not allow answering question 3) and do not highlight clearly the pros and cons of using Bayesian statistics in biological retrospective dosimetry. Finally, no consensus has been reached so far on the best way to proceed to answer the above questions. In this work, we propose an alternative approach - based on the full Bayesian inference of a Poisson mixture model – that allows providing, in a unique and coherent framework, some rich probabilistic answers to the above three questions, simultaneously. Using simulation studies and cytogenetic data from real radiation accident victims and suspected exposed individuals, we highlight the pros and cons of using Bayesian statistics in biological retrospective dosimetry. A sensitivity analysis to the prior choice on the unknown quantities (e.g., the dose) is performed. Our work show that the benefits from using our Bayesian Poisson mixture model are more pronounced for small doses than for high doses.
0000161579_001.PDF (6.36 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-02635585 , version 1 (27-05-2020)

Licence

Copyright (Tous droits réservés)

Identifiants

  • HAL Id : hal-02635585 , version 1

Citer

Sophie Ancelet, Gaetan Gruel, Eric Gregoire, Aurelie Vaurijoux, Laurence Roy, et al.. Bayesian contributions to radiation dose estimation in biological retrospective dosimetry.. Journée Big Ideas for Small Data Journée Big Ideas for Small Data, May 2019, AgroParisTech, Paris, France. ⟨hal-02635585⟩
38 Consultations
11 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More