Improving Image Integrity in Scientific Papers
Complete the form below to unlock access to ALL audio articles.
Protecting the integrity of scientific literature is of the utmost importance. While there have been tools to identify writing issues, such as plagiarism detectors, in operation for many years, there has been no such option for checking image integrity. The result? Thousands of papers that include image errors and issues being submitted to journals, putting the reputations of researchers and publishers at risk. This article sheds light on the issue of image integrity in academic publishing and gives advice on how to reduce the risk.
In academic papers, images are a crucial way of conveying data and illustrating results — one study found that as many as 72 percent of cell biology related papers included image-based figures. According to research by Morten P. Oksvold in Science and Engineering Ethics, image duplications and manipulation in scientific articles impact up to one in four articles in the life sciences. With more than one million papers uploaded to PubMed every year, the scale of the problem is huge.
Interestingly, the vast majority of image issues can be associated with honest mistakes, rather than fraudulent research. Between January 2021 and May 2022, the American Association for Cancer Research (AACR) screened 1,376 papers, of which 208 required author contact to clarify issues, and 4 were withdrawn. In 98 percent of cases, there was no evidence of intentional image manipulation. At the same time, even a mistake in good faith can be a serious mistake that will harm the interpretation of the results.
Types of image integrity issues
Duplication, the most common issue, refers to any form of reusing the same image in different parts of a paper without explicitly stating that it is being repeated. Duplication could be a result of an image being used in the same way twice, or using an altered version that has been rotated, resized, cropped, flipped or rescaled. It could also be that two images include overlapping sections, something more likely when working with microscopy data. Also, with the western blot method it appears that there are many issues of duplication and reuse of the same images.
The risks of duplication
Image integrity issues are a major threat to the reputation of a researcher. Firstly, they may result in a submitted paper or grant application being rejected. However, not all issues will be picked up at peer review — image issues are very difficult to detect by eye. If they are missed and the paper is published, it may be identified post-publication, then it may lead to a long investigation by the publishers and the university and subsequently, the paper may be retracted. A retraction causes irreparable damage to a researcher’s career, as well as the reputation of the journal that published the paper.
Then, there is the issue of further research. Academics often base their investigations on the literature, and if the paper they are relying on contains integrity issues, the new data could also be flawed. In addition, the results may be impossible to replicate, wasting time, materials and money.
Avoiding the introduction of mistakes
So, how are these issues introduced? And why are they so common? Scientific research may be years long and involve a scientist collecting thousands of images — from western blots to microscopy images. Keeping track of images can be a mammoth task, both during collection and storage. A common cause of mistakes later down is improper file naming, and this risk is exacerbated in collaborative teams across different institutions.
The challenge is that once introduced, these mistakes are incredibly difficult to detect by eye. Now consider that a paper could include hundreds of sub-images, and the problem is amplified greatly — this could mean 10,000 comparisons across an entire paper. Researchers therefore often wish to take steps to reduce the risk of accidentally introducing image integrity issues from the beginning. For example: using a file naming system that makes image management more straightforward and labeling images clearly for collaborators and co-authors. For example, including what the sample is, the magnification, date, slide number and any other relevant information. However, there are some risks, such as avoiding overlap during microscopy, that are technically hard to avoid. These must therefore be picked up by thorough checks of the paper.
Checking and detecting
Because it is very difficult and time-consuming for researchers and peer reviewers to manually check images, a growing number of publishers are introducing AI-based image integrity tools to their peer review processes. Computer vision software can scan a manuscript and compare the images in minutes, flagging potential images with issues to the researcher or reviewer, so that they can be reviewed and amended. For example, the AACR, Springer-Nature, JCI, SAGE, RSC and many more publishers have already adopted this technology.
About the author:
Dr. Dror Kolodkin-Gal is founder of Proofiger Ltd, an automated software that uses AI to detect all the sub-images in a scientific paper.
