Simultaneous Multi-Bed MAP Reconstruction with CT-Guided Directional TV Prior for Y-90 PET SIRT
DOI:
https://doi.org/10.2218/piwjournal.10841Abstract
Text Yttrium-90 (Y-90) microspheres are used clinically for selective internal radiation therapy (SIRT) for unresectable liver cancers and have been proposed for glioblastoma1,2. Accurate dosimetry is critical but remains challenging due to low-count PET data arising from Y-90’s low positron branching ratio, as well as accurate treatment of bremsstrahlung for SPECT. Anatomically guided edge-preserving regularisation has shown promise for improving reconstruction quality under such conditions3. Whole-organ PET scans often exceed the axial field-of-view, requiring acquisition across multiple, partially overlapping bed positions (BPs). A common approach reconstructs each BP separately and merges them via sensitivity-weighted averaging. However, in overlap regions, low counts and edge-preserving priors can interact to produce intensity discontinuities that obscure image features. We introduce a maximum-a-posteriori (MAP) framework that reconstructs all BPs jointly, with a CT-guided directional total variation (DTV) prior function applied over a combined image volume4. We evaluate this method for Y-90 PET following SIRT.
Y-90 SIRT PET data was acquired on a GE Discovery 690 using two BPs, with overlap centred on the liver. We compared two workflows:
- Separate + Fuse – independent reconstructions with sensitivity-weighted merging post-reconstruction.
- Joint – Simultaneous reconstruction of both bed positions using a combined image volume, with separate updates computed for the data-fit terms per bed position and a single joint update for the prior term.
Visual assessment shows that Separate + Fuse obtains abrupt changes in intensity in the overlap region due to noise. The Joint method suppresses these artefacts and yields a smoother, anatomically consistent distribution (Figure 1).
Simultaneous reconstruction of multiple bed positions with a CT-guided DTV prior reduces overlap artefacts and enhances qualitative fidelity in Y-90 SIRT PET.
Please click on the 'PDF' for the full abstract!
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Sam Porter, Daniel Deidda, Daniel McGowan, Simon Arridge, Kris Thielemans

This work is licensed under a Creative Commons Attribution 4.0 International License.


