PREreview of A Quantitative Study of Inappropriate Image Duplication in the JournalToxicology Reports

This review reflects comments and contributions from Martyn Rittman and Allie Tatarian. Review synthesized by Stephen Gabrielson.

In this study, both a human reviewer and an AI tool screened papers from the journal Toxicology Reports for inappropriate image duplication.

Major comments:

Having a single human reviewer at both steps (the initial review for duplicated images, and the review of AI-flagged duplications) is a potential source of bias. The safest thing to do is to have three image reviewers - two that conduct an initial review and review the AI-flagged hits, plus a third on hand to act as a tiebreaker in case the two reviewers disagree. Or since this could be a substantial amount of work, having at least an extra reviewer to check the positive hits would be advisable.

Minor comments:

Would like to see more discussion of potential false positives, and what “appropriate” image duplication might look like (especially since the term “inappropriate image duplication” is used throughout).
I would like to know why all papers are taken from a single journal rather than, say, a spread of journals on a similar topic. It leaves the author open to accusations of prejudice against the journal and its editors. A conflict-of-interest statement on whether the author has had previous interactions with the journal would be a useful clarification.
It would be great if the author shares the example of false positive (where the images are mistakenly identified as duplicates). This would give a holistic overview of how good the manual screening was in identifying the duplicate images.
On line 41: "Data duplication between papers, including charts as well as images, has been quantified as high as ~25%" -- is this equivalent to republication without appropriate citation?
On line 66, the author describes that there were cases when images were also compared between other papers by the same research group or author. What prompted this more detailed investigation and why/how were some papers given more scrutiny?

Comments on reporting:

The venn diagram showing how the images were detected is a great idea. Adding the total numbers of papers selected to the study in the venn diagram would be nice, too.

Competing interests