Detecting dataset bias in medical AI using a generalized and modality agnostic auditing approach

Nathan Drenkow, Mitchell Pavlak, Keith Harrigian, Ayah Zirikly, Adarsh Subbaswamy, Mohammad Mehdi Farhangi, Nicholas Petrick, Mathias Unberath · 2026

Read Detecting dataset bias in medical AI using a generalized and modality agnostic auditing approach by Nathan Drenkow, Mitchell Pavlak, Keith Harrigian, Ayah Zirikly, Adarsh Subbaswamy, Mohammad Mehdi Farhangi, Nicholas Petrick, Mathias Unberath (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-13507256.txt.

Description

Nathan Drenkow and others launch a broad, type-blind method for spotting data set bias in medical machine minds. In 2026, the paper presents G-AUDIT, which measures shortcut learning risks by checking links among task labels, sensor readings, and patient, setting, and capture traits. Non-typical training data can amplify latent bias and hide it during testing.

Demos across image, text, and tabular medical data sets and multiple machine learning tasks well found shortcuts often missed by by-eye review, such as doctor markings raising false hits for skin lesions. The tool gives builders an ordered audit for bias before use, helping prevent harmful result gaps across patient groups.

Catalog record

CREATORNathan Drenkow, Mitchell Pavlak, Keith Harrigian, Ayah Zirikly, Adarsh Subbaswamy, Mohammad Mehdi Farhangi, Nicholas Petrick, Mathias Unberath
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13507256

Browse the full library · About the Tower of Babel