Skull-stripping induces shortcut learning in MRI-based Alzheimer’s disease classification
Christian Tinauer, Maximilian Sackl, Rudolf Stollberger, Reinhold Schmidt, Stefan Ropele, Christian Langkammer · 2025
Read Skull-stripping induces shortcut learning in MRI-based Alzheimer’s disease classification by Christian Tinauer, Maximilian Sackl, Rudolf Stollberger, Reinhold Schmidt, Stefan Ropele, Christian Langkammer (2025) 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-12722621.txt.
Description
Christian Tinauer and co-authors show in 2025 that a routine cleanup step can teach brain scan AI the wrong lesson. Deep networks sort Alzheimer disease from healthy brains on scans with high accuracy, but what they actually look at is murky. The team varied how scans were prepared, including stripping away the skull, and watched what the model used.
They trained networks on 990 matched scans under several preparation styles. Accuracy, sensitivity, and specificity barely moved, even when images were reduced to black and white shapes. That stability gave the game away: the models were not reading the fine texture of brain tissue. They were reading volume cues, above all the brain outline that skull stripping creates. The authors call this shortcut learning, a Clever Hans effect, and urge teams to use explanation tools before trusting such models in clinics.
Catalog record
| CREATOR | Christian Tinauer, Maximilian Sackl, Rudolf Stollberger, Reinhold Schmidt, Stefan Ropele, Christian Langkammer |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |