Authors pull their own mouth-cancer data paper
A 37-author team withdrew a dataset paper after finding bugs in its script, saying the reported figures may mislead.


Picture made for ABUREAU.
Owning up to a mistake is rare in research, and worth noticing. In June 2025 a team posted A Cytology Dataset for Early Detection of Oral Squamous Cell Carcinoma, described as the first large, multicentre set of annotated mouth-cell slides. On 6 November 2026 the authors withdrew it.
Their note on the paper's page says they found bugs in their script, and that the figures in the manuscript differ a lot from their recent results and may mislead readers. The abstract says the slides came from ten tertiary medical centres in India, were stained by two standard methods, and were labelled by expert pathologists. It gives no slide count, and no download link appears on the page.
The withdrawal does not say the slides themselves are wrong. It says the reported scores cannot be trusted.
For people who own or use data, the point is that a data set and the numbers claimed for it are separate things. A buyer should ask for the test method and the code, not only the headline score, and an owner who corrects the record early protects the value of what they hold.

