A single model for glioblastoma segmentation with and without T2-FLAIR: independent validation of a targeted dropout strategy
Marco Öchsner, Lena Kaiser, Robert Stahl, Nathalie L. Albert, Thomas Liebig, Robert Forbrig, Jonas Reis · 2026
Read A single model for glioblastoma segmentation with and without T2-FLAIR: independent validation of a targeted dropout strategy by Marco Öchsner, Lena Kaiser, Robert Stahl, Nathalie L. Albert, Thomas Liebig, Robert Forbrig, Jonas Reis (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/PMC13457126.txt.
Description
Marco Öchsner and six colleagues test, in a 2026 imaging study, one brain tumor model that keeps working when a key scan type is missing. Glioblastoma outlining software usually expects a full MRI set, including the FLAIR sequence that lights up swelling. Real clinics sometimes lack it. The team trained a standard segmentation network while randomly blanking the FLAIR channel, a trick called targeted dropout.
With the full scan set, the dropout model matched the ordinary one, scoring about 95 percent overlap with expert outlines. When FLAIR was removed at test time, the ordinary model sagged to about 81 percent, while the dropout model held near 93 percent. Whole-tumor volume error shrank from a forty-six milliliter miss to under one. A separate model trained only on three sequences did just as well, though. The authors credit adapting to fewer inputs, by dropout or by design, as the real fix.
Catalog record
| CREATOR | Marco Öchsner, Lena Kaiser, Robert Stahl, Nathalie L. Albert, Thomas Liebig, Robert Forbrig, Jonas Reis |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |