Positron Emission Tomography (PET) is a method of functional imaging, used in particular for drug development and tumor imaging. In PET, the estimation of the arterial plasmatic activity concentration of the non-metabolized compound (the "input function") is necessary for the extraction of the pharmacokinetic parameters. These parameters enable the quantification of the compound dynamics in the tissues. This PhD thesis contributes to the study of the input function by the development of a minimally invasive method to estimate the input function. This method uses the PET image and a few blood samples. In this work, the example of the FDG tracer is chosen. The proposed method relies on compartmental modeling: it deconvoluates the three-compartment-model. The originality of the method consists in using a large number of regions of interest (ROIs), a large number of sets of three ROIs, and an iterative process. To validate the method, simulations of PET images of increasing complexity have been performed, from a simple image simulated with an analytic simulator to a complex image simulated with a Monte-Carlo simulator. After simulation of the acquisition, reconstruction and corrections, the images were segmented (through segmentation of an IRM image and registration between PET and IRM images) and corrected for partial volume effect by a variant of Rousset’s method, to obtain the kinetics in the ROIs, which are the input data of the estimation method. The evaluation of the method on simulated and real data is presented, as well as a study of the method robustness to different error sources, for example in the segmentation, in the registration or in the activity of the used blood samples.
