DuoMod-Net: Logarithmic balancing and geometric refinement for imbalanced semi-supervised medical image segmentation
Wang Bo, Along He, Ting Xue, Yue Zhang, Yi Xiao, Shiyuan Liu, Shaohua Kevin Zhou · 2026
Read DuoMod-Net: Logarithmic balancing and geometric refinement for imbalanced semi-supervised medical image segmentation by Wang Bo, Along He, Ting Xue, Yue Zhang, Yi Xiao, Shiyuan Liu, Shaohua Kevin Zhou (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/pmc-13280726.txt.
Description
Wang Bo and colleagues published a 2026 paper presenting DuoMod-Net, a learning framework for medical image segmentation when few images are labeled and rare structures dominate. In scans of the abdomen, background pixels and common organs swamp small organs, biasing training twice over: gradients favor the majority, and rare classes get thin feature representations. Their design attacks both problems with two components.
The first rebalances the loss on a logarithmic scale, treating the background as a neutral pivot so foreground organs keep their differences. The second exploits disagreement between models as a geometric signal, expanding the feature space during training so decision boundaries retreat to safer positions. That expansion is removed when the model is deployed. Tested with 5, 10, or 20 percent of data labeled, the system improved rare-class accuracy, cut catastrophic misses, and generalized to unseen datasets without retraining. The authors aim it at clinics where annotation is scarcest.
Catalog record
| CREATOR | Wang Bo, Along He, Ting Xue, Yue Zhang, Yi Xiao, Shiyuan Liu, Shaohua Kevin Zhou |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 87 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |