Medical techniques in full expansion arousing high therapeutic expectations, targeted radionuclide therapies (TRT) consist of administering a radiopharmaceutical to selectively treat tumors. Nowadays, the activity injected to the patient is generally standardized. However, in order to establish robust dose-effect relationships and to optimize treatments while sparing healthy tissues at best, a personalized dosimetry must be performed, just like actual clinical practice in external beam radiotherapy. In that context, this PhD main objective was to develop, using the OEDIPE software, a methodology for personalized dosimetry based on direct Monte Carlo calculations. The developed method enables to calculate the tridimensional distribution of absorbed doses depending on the patient anatomy, defined from CT or MRI data, and on the patient-specific activity biodistribution, defined from SPECT or PET data. Radiobiological aspects, such as differences in radiosensitivities and repair time constants between tumoral and healthy tissues, have also been integrated through the linear-quadratic model. This methodology has been applied to the selective internal radiation therapy (SIRT) which consists in the injection of 90Y-microspheres to selectively treat unresectable hepatic cancers. Distributions of absorbed doses and biologically effective doses (BED) along with the equivalent uniform biologically effective doses (EUD) to hepatic lesions have been calculated from 99mTc-MAA activity distributions obtained during the evaluation step for 18 patients treated at hôpital européen Georges Pompidou. Those results have been compared to classical methods used in clinics and the interest of accurate and personalized dosimetry for treatment planning has been investigated. On the one hand, the possibility to increase the activity in a personalized way has been highlighted with the calculation of the maximal activity that could be injected to the patient while meeting tolerance criteria on organs at risk. On the other hand, the use of radiobiological quantities has also enabled to evaluate the potential added value of fractionated protocols in SIRT. The developed tool can thus be used as a help for the optimization of treatment plans. Moreover, a study has been initiated to improve the reconstruction of post-treatment data from 90Y-SPECT. The estimation from those data of doses delivered during treatment could allow to predict tumoral control and to anticipate healthy tissues toxicity as well as to establish precise dose-effect relationships for those treatments.