RAFM: Retrieval-Augmented Flow Matching for Unpaired CBCT-to-CT Translation
Xianhao Zhou, Jianghao Wu, Lanfeng Zhong, Ku Zhao, Jinlong He, Shaoting Zhang, Guotai Wang · 2026
Read RAFM: Retrieval-Augmented Flow Matching for Unpaired CBCT-to-CT Translation by Xianhao Zhou, Jianghao Wu, Lanfeng Zhong, Ku Zhao, Jinlong He, Shaoting Zhang, Guotai Wang (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/arxiv-2603-00535.txt.
Description
Zhou and six colleagues turn cone beam CT scans into synthetic CT for radiotherapy dose planning, where beam scans carry heavy artefacts and unreliable density values. Paired scans are often missing or misaligned, and plain rectified flow with random pseudo pairs gives unstable training on small medical sets. Their RAFM method builds retrieval guided pseudo pairs with a frozen DINOv3 encoder plus a global CT memory bank, stabilising unpaired flow training. Under a strict true unpaired protocol on SynthRAD2023, RAFM beats existing methods.
Catalog record
| CREATOR | Xianhao Zhou, Jianghao Wu, Lanfeng Zhong, Ku Zhao, Jinlong He, Shaoting Zhang, Guotai Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 21655 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.00535