Multi-head automated segmentation by incorporating detection head into the contextual layer neural network
Edwin Kys, Febian Febian · 2026
Read Multi-head automated segmentation by incorporating detection head into the contextual layer neural network by Edwin Kys, Febian Febian (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-2602-02471.txt.
Description
Edwin Kys and 1 co-author add a detection head to stop medical image models from hallucinating anatomy, in a 2026 arXiv paper. Deep learning auto-segmentation outlines organs for radiotherapy planning. Standard models sometimes paint target structures onto slices where none exist, producing false positives that anatomy rules out.
Their gated multi-head transformer builds on a Swin U-Net design and uses context from nearby slices. A parallel detection head judges, slice by slice, whether the structure is present, and its output gates the segmentation stream. Training uses a slice-wise Tversky loss to handle lopsided class counts. On a prostate data set full of hard edge cases, the gated model's error collapsed compared with a segmentation-only rival. Detection scores tracked real presence closely, nearly wiping out spurious outlines. Valid slices kept their segmentation quality, a key need for use in clinics.
Catalog record
| CREATOR | Edwin Kys, Febian Febian |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 22 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-02471