AACR Presents Results of Using AI-Backed Image Duplication Detection to Curb Academic Fraud
The use of intelligent image comparison tools to detect bad data may be a growing trend in academia
Holden Galusha is the associate editor for Lab Manager. He was a freelance contributing writer for Lab Manager before being invited to join the team full-time. Previously, he was the...
ViewFull Profile.Learn about ourEditorial Policies.
On Thursday, September 8 at the ninth annual International Congress on Peer Review and Scientific Publication, Daniel Evanko, PhD, director of journal operations and systems at the American Association for Cancer Research (AACR) presented on the AACR’s use of artificial intelligence (AI) to detect duplicated images in manuscripts.
As Nature reported in December 2021, the AACR implemented a system to combat image duplication and doctoring in submitted manuscripts. Though oftentimes an honest mistake, image manipulation is a common tactic among those hoping to publish fraudulent studies; by cropping, rotating, or otherwise adjusting images and then reusing them elsewhere, researchers can give the appearance of having gathered more data in support of their findings than in actuality.
Finding duplicated images can be extremely difficult, however. Consequently, image-proofing software like Proofig, which is the AACR’s software of choice, has been developed to augment detection efforts.
In their trial run of Proofig, lasting from January 2021 to May 2022, the AACR processed more than 1,000 manuscripts, of which 208 authors were contacted after the software flagged their submissions and a human confirmed the findings. Of the 208 papers, four were withdrawn and one was rejected.
While AI solutions like Proofig are already extremely sophisticated, there is still room for improvement. For instance, no such program can scan for plagiarized images stolen from other articles; they only compare images within the same study. Furthermore, certain research techniques such as Western blotting produce results with very subtle differences; these small variations are visible to the human eye but are easily missed by AI. As such, significant human intervention is still required to avoid false positives and maintain a good ROI on the software. Ideally, as the algorithms comprising the software become more sophisticated, less human intervention will be required—though such intervention will always be needed to some extent.
Evanko’s presentation of the AACR’s image proofing endeavor may forecast a broader shift across the scientific publishing industry to further incorporate AI into their workflows and combat academic fraud more effectively.