TokenUNet: A new case for transformers integration in efficient and interpretable 3D UNets for brain imaging segmentation
Louis Fabrice Tshimanga, Andrea Zanola, Federico Del Pup, Manfredo Atzori · 2026
Read TokenUNet: A new case for transformers integration in efficient and interpretable 3D UNets for brain imaging segmentation by Louis Fabrice Tshimanga, Andrea Zanola, Federico Del Pup, Manfredo Atzori (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/plos-pone-0354511.txt.
Description
Louis Tshimanga and co-authors present in 2026 a leaner design for outlining brain tumours in three dimensional scans. Transformer based models read global context well but demand heavy computing, which many labs lack. Their design keeps a standard convolutional encoder and pools a fixed small set of tokens from its feature maps, so token count no longer grows with image size.
On a major glioma challenge dataset the model matched a leading heavy rival on the standard overlap score while using far less memory, time, and parameters. Its attention maps also showed plainly where it looked. Easier training on modest hardware could widen who can develop such tools.
Catalog record
| CREATOR | Louis Fabrice Tshimanga, Andrea Zanola, Federico Del Pup, Manfredo Atzori |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43625 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |