AI performance varies considerably across mammography devices: a multi-site and multi-vendor retrospective study
Marcel Blum, Rudolf Morant, Alena Eichenberger, Alexander Geissler, Jonas Subelack, Justus Vogel, David Ehlig · 2026
Read AI performance varies considerably across mammography devices: a multi-site and multi-vendor retrospective study by Marcel Blum, Rudolf Morant, Alena Eichenberger, Alexander Geissler, Jonas Subelack, Justus Vogel, David Ehlig (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-13518669.txt.
Description
Marcel Blum and co-authors present a 2026 study of how one AI reader fares across three brands of mammography machines. Over twenty-two thousand screens of women aged 50 to 69 came from six Swiss sites in 2022 and 2023. A shop AI scored each screen. The team set one shared cut score for all machines, then per-machine scores, and weighed finds and misses both ways.
Standard double reading found about five cancers per thousand screens. The shared AI score found a touch fewer; per-machine scores clawed some back. Per-machine tuning also cut the case load for joint review rounds by near a third. One machine's screens still read poorly under any score. AI sign-off must be per device, the study warns.
Catalog record
| CREATOR | Marcel Blum, Rudolf Morant, Alena Eichenberger, Alexander Geissler, Jonas Subelack, Justus Vogel, David Ehlig |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |