Cancer is responsible every year for the death of 7.6 million people. Treatments improvement is thus of the greatest importance regarding public health. The association of an early diagnosis with an efficient treatment was shown to lead to a significant impact on patients survival rates. Numerous prognostic factors have been identified and are now being used in clinical routine. Nowadays, Positron Emission Tomography (PET) imaging is often used for tumor and metastasis identification because of its established accuracy in numerous cancer models. PET belongs to the functional imaging techniques and may potentially therefore provide information relative to cancer biology. Nevertheless, because of its low spatial resolution, this technique has not been extensively considered for such a purpose. This thesis work aimed at studying quantitative parameters that could be extracted from PET images through texture analysis, in order to characterize tumor heterogeneity. We identified a set of reproducible parameters, robust with respect to partial volume effects as well as segmentation methods that are probably related to the tumor physiology. We have also demonstrated the power of these parameters obtained from diagnostic images for contributing in predicting the therapeutic response as prognostic factors. These new quantitative parameters could in the relatively short term be utilized complementarily to standard oncology factors for patient management purposes.
