TransUNet-GradCAM: a hybrid transformer-U-Net with self-attention and explainable visualizations for foot ulcer segmentation
Akwasi Asare, Mary Sagoe, Justice Williams Asare, Stephen E Moore · 2026
Read TransUNet-GradCAM: a hybrid transformer-U-Net with self-attention and explainable visualizations for foot ulcer segmentation by Akwasi Asare, Mary Sagoe, Justice Williams Asare, Stephen E Moore (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-13621505.txt.
Description
Outlining a diabetic foot ulcer in a photograph — marking exactly which pixels are wound — guides treatment and tracks healing, but ulcers vary wildly in look and setting. Akwasi Asare and co-authors present in 2026 a hybrid model that pairs a U-Net with a transformer bottleneck and attention gates. They test it with unusual rigor, reporting results across five random seeds.
On the public challenge dataset the model reached a Dice overlap score of 0.80. Ablations showed only the hybrid loss function mattered significantly; the transformer and attention pieces added little in-domain. On an external wound-center cohort it kept about 92 percent of its internal score, generalizing partially. Two explanation methods split the honors: one localized the wound better, the other was more faithful to the model's logic. The model is small, under nine million parameters.
Catalog record
| CREATOR | Akwasi Asare, Mary Sagoe, Justice Williams Asare, Stephen E Moore |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |