Including figures and images in clinical research enables researchers to illustrate their findings in a clear, effective, and engaging way. Despite their importance, it can be difficult to manually check images for accuracy — introducing the risk of integrity issues. Here Dr. Dror Kolodkin-Gal, founder of image integrity software developer Proofig AI, explains best practice for image checking in clinical research.
Clinical research is the foundation upon which breakthroughs in medical devices and pharmaceuticals are built. To bring a novel therapy to market, researchers must write and publish sufficient evidence proving that the product meets its intended purpose. This requires collecting and sharing data from multiple sources, such as literature searches, statistics from clinical trials, images, and more.
The evidence shared in clinical research must be accurate, or the consequences can be severe. In 2022, a six-month investigation called into question the validity of results in an integral study of Alzheimer’s disease conducted in 20061. Investigators suspect that some of the figures included in the paper were fabricated, undermining the core findings linking a specific amyloid-β protein assembly to Alzheimer’s disease with neurodegeneration.
The study established the dominant amyloid hypothesis of Alzheimer's, suggesting that the primary cause of the disease is the accumulation of Aβ clumps, known as plaques, in brain tissue. As this hypothesis became more widespread, many researchers developed drugs to combat amyloid formation in patients' brains.
According to Science2, the investigator “identified apparently altered or duplicated images in dozens of journal articles,” all of which were attributed to the author of the study in question. Some Alzheimer’s experts now suspect that studies from this researcher have misdirected Alzheimer’s research for 16 years and potential loses of hundreds of millions of dollars.
This incident highlights the potential damage that can come from not being able to recognise images with issues before publication. While intentional misconduct is a rare occurrence in scientific publishing and very hard to determine for sure, including image issues unintentionally is a very common phenomenon that can still fundamentally damage all parties involved, as well as damage research in general.
Proofig AI
The issue with images
Researchers have multiple resources at their disposal to proactively check written content. They can use software to check for readability, grammatical issues, plagiarism, and more to remove issues in the text and increase their chances of publication. Unfortunately, a lack of tools to proactively check for image integrity issues means that they are prevalent in publishing. According to leading image data integrity analyst Jana Christopher MA, the percentage of manuscripts flagged for image related problems ranges from 20 to 35 per cent3.
In most cases, image integrity issues in scientific papers are unintentional because image complexity makes them difficult to detect. Duplication ─ any form of reusing the same image in different parts of the paper without outlining it ─ is a particularly common accidental offence. For example, two microscope images that include an area of overlap would be classed as duplication. Another way duplication can be accidentally introduced is when an image is flipped or cropped during manuscript preparation.
Duplications like these often occur because researchers will collect hundreds to thousands of images of specimens while conducting research, either for their own paper or for collaborative research with scientists from different universities. If these images are not properly managed, it might be difficult to distinguish between files, increasing the risk of unintentional duplication.
The consequences of image integrity issues
Though image duplications might be unintentional, if researchers do not proactively resolve the issues before publication, it can severely harm their reputation.
During the peer review process, editors and reviewers may detect irregularities with the figures and, depending on the severity of the issue, reject the submission. If the infringement is especially severe, the journal and publisher may suspect the author, which can damage the author’s chances of publishing future work with this publishing house.
If these issues continue to go unnoticed once published and lead to incorrect conclusions, other researchers that believe the content is credible, could base experimental procedures or new research on that paper. Working from inaccurate data means that any new data will also be incorrect, negatively impacting the second researcher and wasting any grant funding raised.
If errors are reported post publication, publishers will investigate to determine how the issue occurred, which can take years. During this time, others may doubt the credibility of the researcher, who may find it difficult to publish other content or successfully apply for funding for new research. If the outcome of the investigation leads to a paper retraction, all parties involved including the researchers, the university where the research was conducted and the publishers, can experience reputational damage that is difficult to rebuild.
Solving the issue
Advances in artificial intelligence (AI) and computer vision has enabled software developers to create automated image proofing software, so that researchers, universities and publishers can proactively check images before publication.
This software automatically scans every sub-image of a research paper to detect any images that appear to have any suspected duplications, checking an entire paper in one to two minutes. Each image is checked against itself and against the rest of the paper, to detect anomalies that researchers should amend before publication. Using image proofing technology enables researchers to check images as quickly and efficiently as they currently check for grammar and plagiarism.
The current investigation into integral research on Alzheimer’s emphasizes the importance of credibility that should be gained by using the best quality control tools during the publication process. By understanding how they can more efficiently detect and resolve any image duplications before submission, researchers can reduce the risk to damage their reputation, publishers can avoid publishing problematic manuscripts with mistakes, and the scientific community can ensure the integrity of future research.
