Currently, the treatment of prostate cancer under open MRI at the University of Geneva Hospital, applying high dose rate (HDR) brachytherapy, involves three main steps : needle implantation, dosimetric calculations and irradiation. As such, there is no pre-planning of the treatment, the experience of the physician providing the sole basis for deciding upon the number and position of the needles. Moreover, since the implantation and the dosimetry are performed at different places and at different times, it is not possible to modify the needle configuration, in case it is found that the entire tumour volume cannot be sufficiently irradiated. The present research aims at achieving a significant improvement of the cancer treatment by introducing new methods for HDR brachytherapy involving inverse planning procedures and three dimensional dosimetric databases. Thereby, the possibility is offered to the radio-oncologist to define the regions to be treated, to express the desired values for the dose to be imposed or tolerated for each organ, to propose a specific needle configuration and to verify the relevance of the proposition before proceeding to implantation. Finally, a guiding system, coupled to the MR imaging, serves to ensure correspondence between virtual and real treatment. Such possibilities require the development and validation of appropriate routines for dosimetric calculations and the accurate and rapid optimisation of irradiation conditions, for various source types, both gamma and beta emitting. The final product of the present research is a programme called PROTON. Other necessary developments, which are being implemented by a collaborating organisation, viz. the Signal Processing Laboratory at EPFL, concern the graphical interface of the programme, the volume segmentation tools and the implantation guidance. The dosimetric problem is solved by applying detailed Monte Carlo simulations to generate 3-dimensional databases, which are large enough in extent to cover a standard treatment volume with a single source placed at its centre. The dose map corresponding to location of several sources can then be obtained via a linear combination of dose values from the database of the specific source. The chosen sources for the simulations are the 192Ir model microSelectron of Nucletron B. V., two HDR prototypes with 169Yb and 144Ce and the LDR implants 6711 with 1251 from OncoSeed and 200 with 103Pd from TheraSeed. The next step taken is to verify the accuracy and pertinence of the databases, via experimental studies involving a single source and by comparisons with numerical results obtained applying other methods. The experimental verification carried out has been for the 192Ir and 144Ce sources. Radial dose rate distributions were measured in a water phantom employing accurately calibrated detectors of several different types, both active and passive. For 192Ir, a 0.22 cm3 ionisation chamber, cylindrical lithium fluoride thermoluminescent dosimeters (TLD) of 1 mm diameter and 3 mm length, and metal-oxide semiconductor field-effect transistors (MOSFET) were used. In the case of 144Ce, only the ionisation chamber was employed. The targeted accuracy for the experimental verification of the Monte Carlo simulations has been 5 and 15 % for 192Ir and 144Ce, respectively. It has been shown that this goal is achieved for 192Ir with the ionisation chamber and the MOSFETs, a systematic experimental error resulting in a somewhat larger discrepancy in the case of the TLDs. The 144Ce measurements are found to match well with the simulations, the calculation/experiment differences being generally within 10 %. For the comparison of the dosimetric calculations with other numerical results, choice has been made of the commercially available programme PLATO BPS, routinely used at the University of Geneva Hospital for treatment planning with 192Ir, as also a mathematical interpolation based on the TG-43 formalism. The comparisons with PLATO BPS, for both a single source and for the actual treatment of twelve patients treated between December 2003 and July 2004, show good agreement. The same conclusion results from the comparison with the mathematical interpolation, demonstrating thereby the correctness and pertinence of the dosimetric databases. As regards the optimisation of source dwell times at each position, the routine developed is based on application of a simulated annealing algorithm that minimises an objective function while associating, to each optimisation point, a larger penalty if the dose is further from the prescription for the considered organ. Minimal, maximal doses can thus be specified for the tumour and the organs at risk, respectively (the latter being notably the urethra, the rectum and the bladder). The validation of the optimisation routine has involved, once again, the treatment characteristics for the twelve patients treated using PLATO BPS. In this context, it has been shown that PROTON can follow imposed prescriptions much more efficiently. Moreover, if an appropriate database is available, one has the possibility to proceed with the treatment employing a different type of source. In the present study, we have used 169Yb, which provides treatment conditions very similar to those obtained with 192Ir, and 144Ce, which shows certain advantages over 192Ir, e.g. for the case of a small tumour close to an organ at risk. Finally, for the case of an 192Ir treatment, a presentation is made of the integration of the dosimetry and optimisation programme PROTON with the graphical interface currently under development at the Signal Processing Laboratory. The principal steps of the treatment are illustrated, from the visualisation of tomographic planes to the shape definition of the organs, as also from the treatment verification to the display of results. In conclusion, it has been shown that the developed dosimetry and optimisation routines are appropriate and applicable to source dwell time calculations during a HDR interstitial brachytherapy treatment of the prostate. Integrated into a suitable graphical interface, they allow performing a pre-planning of the treatment. It does appear, however, that a mathematical interpolation of the 3-dimensional dosimetric databases would permit notably a significant gain in time during the optimisation procedure. This is an aspect suggested for future development in the context of the present research.