Stochastic Optimisation Methods Applied to Image Reconstruction: results, software and outlook.

Title: Stochastic Optimisation Methods Applied to Image Reconstruction: results, software and outlook.

Abstract: Stochastic Variance Reduction (SVR) optimisation methods have the potential to outperform traditional subset methods for image reconstruction. In this talk, I’ll describe the rationale for SVR and present results for PET image reconstruction from Twyman et al. TMI 2022, as well as preliminary results for other image reconstruction settings, including Motion Incorporated Image Reconstruction for MR, and synergistic reconstruction for PET and SPECT data. Finally, I will introduce a generic implementation in the Open Source Software “Core Imaging Library” (CIL) and give some example results.

Will not be available via zoom.