The challenges with policing image integrity
Research Square’s VP of Publishing Innovation, Damian Pattinson, recently took part in an inquiry by the UK Parliament that focused on research integrity. Wendy Appleby (Registrar and Head of Student & Registry Services, University College London) and Professor David Hand (representing the Royal Statistical Society) also provided testimony, covering a wide range of topics from the general lack of statistical training for researchers to how institutions investigate allegations of misconduct.
One theme winding throughout the proceedings was image integrity in research manuscripts, something that we are actively focused on. All parties involved agreed that the incidence of improper image manipulation is worthy of more sustained attention, but the subsequent discussion underscored many of the challenges inherent in creating solutions.
Boundaries of acceptable image manipulation. One important question to answer before defining solutions is what counts as improper manipulation of a scientific figure. The Journal of Cell Biology’s best practices are a great start, but there are open questions around how much an image can be adjusted before important information is obscured or misrepresented. And most importantly, that decision will need to be made by a person with some familiarity with the paper’s field of research.
At one point, MP Stephen Metcalfe asked whether manipulation of images should be completely banned, which would certainly eliminate some issues. However, a blanket ban would be impractical because images are often much larger than needed or could be much more effectively presented when formed into a multipanel figure, for example. Altogether, guidance can be given, but in any situation short of a full ban, there will still need to be mechanisms in place to detect improper alterations.
Difficulty implementing standards across the publishing industry. A separate line of discussion in the proceedings centered on the responsibilities of authors. One suggested course of action was to require authors to describe the manipulations they made on their figures and to certify that they did not engage in any attempts to defraud the editors, reviewers, or readers. Such a policy would certainly make it easier to deal with any manipulation later discovered, and it would force authors to more carefully consider what they are doing when preparing figures in the first place.
Unfortunately, the sheer size of the publishing industry is a barrier to the implementation of such a standard, especially one that involves requirements of authors. An analagous situation to requiring author engagement about figures would be a publishing data policy requiring the data underlying a paper to be deposited and available.
The open access giant PLOS instituted a new data policy in 2014, and while it continues to exist with minor clarifications, the rollout was not without controversy. It also created a barrier to publication that led some authors to choose other journals. Perhaps most importantly, while this principled stand appears to have been largely successful, it only represents a small portion of the biomedical articles published each year. Other publishers will need to follow suit, while fully aware that PLOS experienced some growing pains.
Limits of automation. Automated solutions are often proposed for image-related concerns. We and others believe that computing power can be brought to bear on the issue, making it much simpler and more cost effective to screen images for improper manipulation. However, it’s highly unlikely that any fully automated tool could properly assess what is permissible and impermissible regarding figure alterations.
For example, a panel may be duplicated in separate figures as a reference to readers and properly acknowledged by the authors. Background may have been adjusted to remove a smudge, which a trained scientist could easily identify as an artifact and not important data. Computers would see these events similarly to true acts of fraud.
The limitations of automation also raise the issue of cost. Without high levels of automation, human time and effort is require to screen each of the millions of papers submitted to journals each year. While many publishers are willing to pay a small fee to iThenticate for plagiarism detection, it’s unclear whether the same is true for a much larger outlay for image manipulation detection.
Further, while the copying of text can really only occur in one way, the use of automated tools has the potential to create an “arms race,” wherein authors find new ways to skirt the tool’s algorithms.
Awareness of image manipulation as an issue has now moved beyond the publishing industry to the governments that oversee research funding and accreditation of higher education. Despite the many challenges, the importance of the problem is not lost. Therefore, we will likely see some progress on this front in the coming months and years, helping safeguard the scholarly record and ensure that important decisions are based on accurate and sound research results.
To watch the full inquiry session, please visit the UK’s parliamentlive.tv site: http://www.parliamentlive.tv/Event/Index/7ec4392f-6208-438a-b595-1fa778940e02.
