Dual-domain transformer-based learned primal-dual reconstruction for PET imaging
Anton Adelöw, Hamidreza Rashidykanan, Alessandro Guazzo, Massimiliano Colarieti-Tosti · 2026
Read Dual-domain transformer-based learned primal-dual reconstruction for PET imaging by Anton Adelöw, Hamidreza Rashidykanan, Alessandro Guazzo, Massimiliano Colarieti-Tosti (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13473274.txt.
Description
Adelöw and colleagues present 2026 transformer-based methods for rebuilding PET scan images, which are noisy by nature. Earlier learned reconstruction used convolutional networks, whose local view can miss long-range patterns in the raw detector data. The team built three designs that add attention mechanisms, which let the model weigh distant information, inside a learned primal-dual framework.
Two designs beat the standard baseline on error and quality measures at every noise level in synthetic tests. A third design, adding cross-attention between image and data domains, also improved accuracy but slightly lowered a structural similarity score. Notably, models trained only on realistic synthetic data from the team's small-animal scanner still produced high-quality reconstructions from real measurements, easing the need for scarce real training scans.
Catalog record
| CREATOR | Anton Adelöw, Hamidreza Rashidykanan, Alessandro Guazzo, Massimiliano Colarieti-Tosti |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |