Résumé:
Linear Dimension Reduction method (LDR) is a classification technique, based on space
projections. It is considered as an alternative technique to principal component analysis and region of
interest methods. These latter methods, usually, used to extract physiological component parameters of
interest in medical imaging.
This project consisted in reviewing the fundamental mathematical aspect of this method, followed
by its algorithmic procedure. This algorithm has been used in this thesis to decompose a sequence of rat
heart dynamic images into tissue and blood components with minimum noise. The dynamic resulting
images were subsequently used in kinetic modeling to assess physiological parameters. The dynamic
images were measured with positron emission tomography (PET) technique.
