Spatiotemporal PET reconstruction using total variation based priorsCP3

In this talk, we focus on dynamic Positron Emission data reconstruction. We present regularization methods that are based on different edge preserving priors adapted to problems corrupted by Poisson noise. In particular, we consider spatiotemporal Total Variation (TV), Total Generalised Variation (TGV) reconstructions and their extensions to the infimal convolution approach as proposed by Holler and Kunisch. The numerical solutions of the corresponding variational problems are evaluated using Primal-Dual Hybrid Gradient (PDHG) optimisation methods under diagonal preconditioning. We compare them with the state of the art techniques as Expectation Maximization (EM) reconstruction and Filtered backprojection for simulated dynamic brain data. This is a joint work with Clovis Tauber (INSERM, Tours) and Maïtine Bergounioux (MAPMO, Orléans).

This presentation is part of Contributed Presentation “CP3 - Contributed session 3

Evangelos Papoutsellis (Université Francois Rabelais de Tours)
Clovis Tauber (UMRS Inserm U930, Imagerie et Cerveau, Université de Tours)
Maïtine Bergounioux (CNRS - Université d'Orléans UMR 7013)
image reconstruction, inverse problems, medical imaging, pet reconstruction