VAE constrained MR guided PET reconstruction
Résumé
In this work, we investigate a deep learning PET-MR joint reconstruction method based on the ADMM algorithm. The a priori information to regularize the inverse problem is obtained with a VAE trained with high-quality images. Adaptive choice of the Lagrangian parameter ensures good convergence properties of the method. The proposed approach is tested on simple cases. It outperforms the classical MLEM for high noise levels.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| licence |
