Contribution to dynamic MRI analyze for diagnosis support for the prostate cancer

Prostate cancer is the most common cancer among men. Its development leads to a neo-angiogenesis that changes the capillary network. It is recognized that DCE-MRI is able to distinguish these physiological changes in microcirculation. However, the images are difficult to analyze and interpret. In this thesis, we were interested by the development of robust methods for the analysis of these images. Initially, we were focused on pharmacokinetic parameters quantification methods. A software platform was constructed to implement the multi-step Tofts model. Technical validation was performed using simulated images with knowledge of the ground truth. Clinical validation is in progress in the Radiology department of Lille University Hospital. In parallel, we have explored the application of nonparametric and unsupervised techniques of data processing for time-intensity curve analysis. We have developed an original approach based on spectral classification. This method, based on graph theory, allows the grouping of signals after transformation of the space of representation. Subsequently, these groups of data can be labeled by comparison to the arterial signal serving as reference. Preliminary experiments conducted on simulated data as well as clinical data show the feasibility of the approach. The two approaches are complementary, one giving quantitative parameters and the other segmenting the cancerous areas. (author)