Tomographie et géométrie discrètes avec la transformée Mojette

摘要

We explore through this thesis the insights of discrete tomography over classical tomography in continuous space. We use the Mojette transform, a discrete and exact form of the Radon transform, as a link between classical tomography and discrete tomography. We focus especially on the study of the discrete space induced by the Mojette transform operator through four research axis.Axis 1 focuses on the Mojette space properties in regards to discrete affine transforms of digital images. We provide tools to achieve affine transforms directly from the projections of a digital object, without preliminary tomographic reconstruction. This property is well-known for the continuous Radon transform but non-trivial for its sampled versions.Axis 2 seeks for some links between continuous-sampled projections related to medical imaging acquisition modalities and discrete projections derived by the Mojette transform. We implement interpolation schemes to estimate discrete projections from the continuous ones — on either synthetic or real data — and their reconstruction.In axis 3, we provide an algebraic framework for the description and inversion of the Mojette transform. The input data, the projections as well as the operators are modeled as polynomials. Within this framework, the Mojette projection operator advantageously reduce to a Vandermonde matrix.This thesis has been realized at both IRCCyN Lab and Keosys company within the Quanticardi FUI project. Axis 4 focuses on the design and the implementation of a clinical software for the absolute quantification of myocardial perfusion with positron emission tomography.