FD-MSP: feature decoupling network with multi-scale prototypes for domain-adaptive polyp segmentation
Wenqi Zhong, Xu Yang, Shengyuan Liu, Yapeng Wang, Sio-Kei Im · 2026
Read FD-MSP: feature decoupling network with multi-scale prototypes for domain-adaptive polyp segmentation by Wenqi Zhong, Xu Yang, Shengyuan Liu, Yapeng Wang, Sio-Kei Im (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/PMC13584388.txt.
Description
Wenqi Zhong and colleagues build a smarter image tool for finding gut polyps during colonoscopy in this 2026 paper. Polyps can turn into bowel cancer, so software that outlines them helps doctors act early. The catch is that models trained on images from one hospital often stumble on scans from other machines, and polyps range from tiny flat spots to large lumps.
Their system, named FD-MSP, splits what it learns into features that stay the same across machines and features tied to each device. It also keeps model patterns at three size scales and blends them through a gated output stage. Tested on three public colonoscopy datasets, the tool beat earlier cross-domain methods on the new sets. The authors frame the design as a fix for both the machine gap and the size spread at once.
Catalog record
| CREATOR | Wenqi Zhong, Xu Yang, Shengyuan Liu, Yapeng Wang, Sio-Kei Im |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 153190 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |