Prostate cancer is one of the leading causes of death from cancer among men. In the last decade, new diagnosis procedures and treatment options have been developed and made possible thanks to there cent progress in prostate imaging modalities. The newest challenges in this field are to detect the smallest tumours and to treat locally to minimise the treatment morbidity. In this thesis, we introduce a set of automatic image processing methods for the guidance and assistance of diagnosis and treatment, in laser-based prostate cancer focal therapies. In the first part of this work, segmentation and computer-aided detection algorithms have been developed for the enhancement of image-based diagnosis of prostate cancer. First, we propose a novel approach that combines Markov Random Fields framework with an Active Shape Model, in order to extract three dimensional outlines of the gland from Magnetic Resonance Imaging (MRI) data. Second, prostate's MRI volume is segmented in to peripheral and central zones: we introduce a method that explores features of multispectral MRI, and is based on belief functions and the modelling of an a priori knowledge as an additional source of information. Finally, computer-aided detection of prostate's peripheral zone tumours is investigated by experimenting novel texture features based on fractal geometry-based. These parameters, extracted from morphological MRI, were tested using both supervised and unsupervised classification methods. The results of these different approaches were studied and compared. The second part of this work addresses the guidance of laser-based focal ablation of prostate tumours. A novel non rigid registration method is introduced for fusion of pre-operative MRI and planning data, and per-operative ultrasound imaging. We test and evaluate our algorithms using simulated data and physical phantoms, which enable comparison to ground truth. Patients' data, combined to expert interpretation,are also used in the evaluations while taking into account the inter-observer variability. The results we obtained show that the methods we developed are satisfyingly accurate, fast and robust to be used in a clinical context. These tools have proven that they enable both time saving and reproducibility in diagnosis and treatment decisions. Multi-centric validation and transfer to the industrial world would bring the contributions of this work to clinical routine, and help improving diagnosis and therapy of prostate cancer.
