Finding the false

For Mike Rossner, the impetus to look for image manipulation came suddenly. In 2002, when Rossner was the managing editor at the Journal of Cell Biology, the journal switched to completely electronic submissions. One of the first submissions after the change contained unusable PowerPoint images. In the process of reformatting, Rossner, now executive director of Rockefeller Press, discovered alterations in some of the images. "We decided right then that we were going to monitor all images of all of our accepted manuscripts for evidence of manipulation," he says.

Today, editors at JCB screen images from every accepted manuscript before publication by dialing contrast and brightness up and down in Photoshop, which can reveal tell-tale signs of cutting and pasting invisible under normal settings. But does the technique work? Other journals say that the scientific process naturally ferrets out fakes as scientists find they can't repeat the experiments, then contact the journals, which retract the papers. Is this process as effective? If so, one would expect that journals that don't screen for bad images would retract as many articles with manipulated images as JCB. But they don't.

Of cell biology journals with the ten highest impact factors in 2006, only editors at JCB and European Molecular Biology Organization Journal say they regularly check for image tweaks. Editors at the four Cell Press journals on the list - Cell, Molecular Cell, Cell Metabolism and Developmental Cell - and at Genes and Development do not screen for evidence of image manipulation. At the remaining journals, all published by Nature, editors screen one to two randomly chosen articles per issue. It's a conscious choice not to do more, says Bernd Pulverer, chief editor of Nature Cell Biology. "While in our view the journal has a responsibility to ensure legitimate and robust data is published, we do not aim to take on the role of 'data police,'" he wrote in an E-mail. "It is essential that a level of trust is maintained in the scientific community."

So, how do the retraction rates compare? Since 2002, editors at JCB have looked for evidence of image manipulation in all of their nearly 2,200 accepted articles. They've found slightly more than one percent (25 articles) contain images that suggest deliberate falsification - that is, images that have been retouched to the point of changing their interpretation. Naturally, these papers never made it to press.

To see if eagle-eyed readers were catching that same one percent at other journals, I searched Pubmed for retractions from the other nine top cell biology journals since 2002 and checked if image manipulation played a role in the decision. In some cases, I contacted editors at the journals to clarify why certain articles were retracted.

Rates of retraction due to image manipulation were much lower than one percent, ranging from 0.04% at Nature Medicine to 0.3% at Nature Cell Biology. Molecular Cell, Nature Structural and Molecular Biology and Developmental Cell have not retracted any papers as a result of image doctoring in that time period. The numbers imply there could be many articles with undetected image falsification lurking in the scientific literature.

Juan Carlos Lopez, chief editor of Nature Medicine, confirms that the retraction rate I found is accurate for that journal. He says finances are the main reason Nature Medicine doesn't screen more articles right now, but notes that could change in the future. Pulverer notes that the low retraction rate for his journal "sounds about right," but the journal is likely not going to adopt JCB's strategy, given that scientists who really want to cheat will find a way, even with screening. "We are working on improving the way we display micrographs, and this will do much more to decrease the manipulation rate than screening."

Emilie Marcus, executive editor of Cell Press, says that she does not keep track of retraction numbers, so she could not confirm my findings. Cell's screening policy could also change, she says, but adds she does not think journals should screen for image manipulation. "It is not the journal's full responsibility to impose ethical standards on the scientific community."

Rossner says he hasn't seen any other studies comparing JCB's image-doctoring rejection rates to retraction rates from other journals, and agrees that simply waiting for readers to find problems isn't enough. "The whole reason we started screening images is because our reviewers were not picking up the problems," Rossner says. "So, it stands to reason that readers of published articles are similarly missing these manipulations, and your numbers bear this out."