How AI is Assisting Publishers in Detecting Scientific Misconduct

Addressing Data Integrity Issues in Research Publications

Scientific publishers are increasingly turning to AI-driven software to detect potential data manipulation in academic papers. With image duplication being a common issue, publishers are leveraging technology to ensure research integrity and prevent fraudulent submissions.

Manipulated images—such as duplicated, flipped, cropped, or stretched figures—can create the illusion of additional data, misleading readers and reviewers. The American Association for Cancer Research (AACR) has identified image duplication as the leading cause of retractions between 2016 and 2020. Retractions not only damage an author's credibility but also undermine the journal’s reputation, highlighting weaknesses in the peer-review process.

To proactively address this issue, publishers like AACR have implemented AI tools such as Proofig AI, an advanced image-checking software. Developed by an Israeli startup of the same name, Proofig AI utilizes artificial intelligence, computer vision, and image processing to analyze figures for potential duplication before papers are accepted for publication.

AI-Powered Screening Enhances Accuracy

From January 2021 to May 2022, AACR deployed Proofig AI to assess 1,367 provisionally accepted manuscripts. The software flagged potential duplication in 208 cases, prompting further review. In most instances, the issues stemmed from honest errors rather than intentional fraud. However, four papers were withdrawn, and one was ultimately rejected due to image integrity concerns.

A significant challenge in detecting manipulated images lies in analyzing Western blots—a common technique used to detect proteins. These images consist of subtle variations in line widths, making it difficult to identify manipulations manually. AI-powered tools, such as Proofig AI and ImageTwin, are designed to assist in this process, as their algorithms can efficiently scan for rotated, stretched, or copied sections within an image set.

AI as a Tool, Not a Replacement for Human Review

While AI tools significantly improve the efficiency of image screening, experts emphasize that human oversight remains essential. Image forensics expert Elisabeth Bik notes that software can flag issues, but human expertise is required to interpret results accurately. Proofig AI’s developers acknowledge that Western blot analysis remains particularly complex, requiring extensive algorithmic refinement.

Publishing organizations, including AACR, Taylor & Francis, and Wiley, are implementing AI-based screening tools at various stages of the publication process. Some, like Taylor & Francis, use the software for targeted investigations when editorial concerns arise, while others integrate it into their review workflow for broader integrity checks.

The Future of AI in Research Integrity

Despite advancements, AI software currently has limitations—it can detect duplicated images within a single paper but cannot yet identify plagiarism across different research articles. To address this, industry experts suggest collaboration among publishers to develop a shared database of problematic images.

Dror Kolodkin-Gal, co-founder of Proofig AI, emphasizes the importance of big data in refining AI detection capabilities. He advocates for publishers to work together in creating comprehensive image databases to enhance fraud detection.

While AI cannot eliminate scientific misconduct entirely, its adoption serves as a deterrent, signaling to researchers that their submissions will be scrutinized for integrity. As AI technology continues to evolve, it is poised to play a critical role in upholding research credibility and reinforcing trust in scientific publications.