SIAM: Head and Brain MRI Segmentation from Few High-Quality Templates via Synthetic Training
Romain Valabregue, Ines Khemir, Eric Badinet, François Rousseau, Guillaume Auzias Reuben Dorent · 2026
Read SIAM: Head and Brain MRI Segmentation from Few High-Quality Templates via Synthetic Training by Romain Valabregue, Ines Khemir, Eric Badinet, François Rousseau, Guillaume Auzias Reuben Dorent (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-2605-02737.txt.
Description
Romain Valabregue and five co-authors present SIAM, a model that segments whole-head MRI scans after training on just six hand-labeled templates. Synthetic training has advanced brain segmentation by letting models learn from generated data in any contrast. Most methods, though, need hundreds of auto-labeled templates, which bake in biases and make adding new structures awkward.
SIAM randomizes both image intensity and shape, so synthetic scans vary in contrast and in anatomy, including cortical thickness and deep nuclei form. It labels sixteen structures at once, covering brain tissue plus fluid, vessels, membranes, skull, and skin, with no preprocessing step. Across eight varied datasets with hundreds of scans, spanning common contrasts and a wide age range, SIAM matched or beat leading methods on brain structures and extended coverage beyond them.
Catalog record
| CREATOR | Romain Valabregue, Ines Khemir, Eric Badinet, François Rousseau, Guillaume Auzias Reuben Dorent |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.02737