Towards reliable use of artificial intelligence to classify otitis media using otoscopic images: Addressing bias and improving data quality

Yixi Xu, Al-Rahim Habib, Graeme Crossland, Hemi Patel, Chris Perry, Kris Bock, Tony Lian, William B Weeks · 2026

Read Towards reliable use of artificial intelligence to classify otitis media using otoscopic images: Addressing bias and improving data quality by Yixi Xu, Al-Rahim Habib, Graeme Crossland, Hemi Patel, Chris Perry, Kris Bock, Tony Lian, William B Weeks (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/plos-pone-0338867.txt.

Description

Yixi Xu and co-workers audit the photo collections used to train artificial intelligence that diagnoses middle ear infections from eardrum images. Biased training photos can make a smart tool fail in new clinics. Their 2026 study probes three public datasets from Chile, the United States, and Türkiye, covering parts of images and shifting colors to see what the models truly rely on.

Two datasets carried heavy bias: models scored high at home yet leaned on irrelevant marks and flopped abroad. A model trained on meaningless features from one set still scored well on the other biased set, proving the flaw. The third set leaned more on real clinical features. Standard photo rules and wider sampling are needed before such tools earn trust.

Catalog record

CREATORYixi Xu, Al-Rahim Habib, Graeme Crossland, Hemi Patel, Chris Perry, Kris Bock, Tony Lian, William B Weeks
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36312
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0338867

Browse the full library · About the Tower of Babel