Faster PET-MR image reconstruction by stochastic optimizationMS23

Image reconstruction with convex anatomical priors can be performed by the Primal-Dual Hybrid Gradient (PDHG) algorithm (Chambolle and Pock, 2011). For many imaging modalities, e.g. PET, the problem has a special saddle point structure which is usually not exploited. We propose a stochastic extension of the PDHG which randomly selects subsets of the data in each iteration. Numerical results on clinical PET data show a significant speed-up by this technique.

This presentation is part of Minisymposium “MS23 - Multi-Modality/Multi-Spectral Imaging and Structural Priors (2 parts)
organized by: Matthias J. Ehrhardt (University of Cambridge) , Simon Arridge (University College London) .

Matthias J. Ehrhardt (University of Cambridge)
computed tomography, image reconstruction, inverse problems, nonlinear optimization, stochastic optimization