μ 2 \mu^{2} Tokenizer: Differentiable Multi-Scale Multi-Modal Tokenizer for Radiology Report Generation
Siyou Li, Pengyao Qin, Huanan Wu, Dong Nie, Arun J. Thirunavukarasu, Juntao Yu, Le Zhang · 2025
Read μ 2 \mu^{2} Tokenizer: Differentiable Multi-Scale Multi-Modal Tokenizer for Radiology Report Generation by Siyou Li, Pengyao Qin, Huanan Wu, Dong Nie, Arun J. Thirunavukarasu, Juntao Yu, Le Zhang (2025) 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-2507-00316.txt.
Description
Siyou Li and co-authors present a medical AI paper, published in 2025, on writing radiology reports straight from CT scans. Report generation is hard twice over: the model must mine a huge scan for what matters, and judging the draft against an expert's report is itself unreliable. Their system threads a new tokenizer between image and language, blending features across scales and across image and text in a way training can tune end to end.
A preference optimization step, steered by an automated report quality scorer, polishes the output. The team also built a five stage pipeline that turns routine reports into question answer teaching data with cited reasoning. Across four large datasets the method beat existing systems, even trained on limited data.
Catalog record
| CREATOR | Siyou Li, Pengyao Qin, Huanan Wu, Dong Nie, Arun J. Thirunavukarasu, Juntao Yu, Le Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-00316