Abstract : This work deals with the segmentation of echographic prostate images acquired intra-operatively simultaneously to the application of high intensity ultrasound (HIFU) for the destruction of tumor. The objective is to precisely delineate the target in order to concentrate the heating induced by ultrasound and to reduce as much as possible their impact on the neighboring structures. After a brief presentation of the clinical and the technical context, in particular the dosimetry, a state-of-the-art is carried out : it shows that although all the most recent and effective methods have been explored, only partial solutions have been obtained due to a large variability of situations to deal with and the relatively poor quality of the images in HIFU devices. Several solutions are then proposed based on discrete deformable contours, improved by the search of gradient based landmark points, coupled or not with optimal search detection. These solutions are tested on about thirty datasets and qualitatively and quantitatively analysed by comparison with expert-defined contours. A preliminary study is also performed for texture characterization using different types of moments (Zernike, Legendre, etc.). The results show that an overall good behaviour is obtained fulfilling the time computation constraints.