Thesis: The in vivo measurement is an efficient method to estimate the retention of activity in case of internal contamination. However, it is currently limited by the use of standard physical phantoms for the calibration, reproducing neither the individual morphology of the measured person nor the actual distribution of the contamination, which therefore leads to significant systematic uncertainties on its quantification. To improve the in vivo measurement, an original numerical calibration method based on voxel phantoms created from the medical images of persons and associated with the MCNPX Monte Carlo code of particle transport was developed through the OEDIPE software. In this thesis, the dynamic feature, and often heterogeneous distribution between body organs and tissues of the activity was simulated on the basis of the radionuclide biokinetic models elaborated by the International Commission on Radiological Protection, in order to study its influence on the detection efficiency and to provide correction factors for the current calibration. The method was applied at the medical department of AREVA NC La Hague reprocessing plant and an actual actinide inhalation case was studied a posteriori to check the influence of the biokinetics. This thesis opens new prospects for the estimation of uncertainty in internal dosimetry and provides a new method to analyse contamination cases by personalizing the biokinetic model in order to improve the estimation of the dose.