j cell biol 2004 rossner 11 5

7/29/2019 J Cell Biol 2004 Rossner 11 5

1/5

The Journal of Cell Biology, Volume 166, Number 1, July 5, 2004 1115http://www.jcb.org/cgi/doi/10.1083/jcb.200406019

JCBFeature

11

Whats in a picture?The temptation of image manipulation

Mike Rossner

1

and Kenneth M. Yamada

2

1

Managing Editor, The Journal of Cell Biology

2

Editor, The Journal of Cell Biology, and the National Institute of Dental and Craniofacial Research, National Institutes of Health

Its all so easy with Photoshop

1

. In thedays before imaging software becameso widely available, making adjust-ments to image data in the darkroomrequired considerable effort and/or ex-pertise. It is now very simple, and thustempting, to adjust or modify digital

image files. Many such manipulations,however, constitute inappropriatechanges to your original data, andmaking such changes can be classifiedas scientific misconduct. Skilled edito-rial staff can spot such manipulationsusing features in the imaging soft-

ware, so manipulation is also a riskyproposition.Good science requires reliable data.Consequently, to protect the integrityof research, the scientific communitytakes strong action against perceived

scientific misconduct. In the currentdefinition provided by the U.S. gov-ernment: Research misconduct is de-fined as fabrication, falsification, orplagiarism in proposing, performing,or reviewing research, or in reportingresearch results. For example, showinga figure in which part of the image waseither selectively altered or recon-structed to show something that didnot exist originally (for example, add-

ing or modifying a band in a polyacryl-amide gel image) can represent falsifi-cation or fabrication.

Being accused of misconduct ini-tiates a painful process that can disruptones research and career. To avoidsuch a situation, it is important to un-

derstand where the ethical lines aredrawn between acceptable and unac-ceptable image adjustment.

Here we present some general guide-lines for the proper handling of digitalimage data and provide some specificexamples to illustrate pitfalls and inap-propriate practices. There are differentdegrees of severity of a manipulation,depending on whether the alterationdeliberately changes the interpretationof the data. That is, creating a result is

worse than making weak data look bet-

ter. Nevertheless, any manipulationthat violates these guidelines is a mis-representation of the original data andis a form of misconduct. All of the ex-amples we will show here have beencreated by us using Photoshop; al-though they may appear bizarre, it isremarkable that they are actually basedon real cases of digital manipulationdiscovered by a careful examination ofdigital images in a sample of paperssubmitted (or even accepted) for publi-cation in a journal.

Why is it wrong to touch upimages?

If you misrepresent your data, you aredeceiving your colleagues, who expectand assume basic scientific honestythat is, that each image you present isan accurate representation of what youactually observed. In addition, an im-

age usually carries information beyondthe specific point being made. Thequality of an image has implicationsabout the care with which it was ob-tained, and a frequent assumption(though not necessarily true) is that inorder to obtain a presentation-quality

image, you had to carefully repeat anexperiment multiple times.Manipulating images to make figures

more simple and more convincing mayalso deprive you and your colleagues ofseeing other information that is oftenhidden in a picture or other primarydata. Well-known examples include ev-idence of low quantities of other mole-cules, variations in the pattern oflocalization, and interactions orcooperativity.

Journal guidelines

It is surprising that many journals saylittle or nothing in their Instructionsto Authors about which types of digi-tal manipulations are acceptable and

which are not. The following journalsprovide some guidelines, but they vary

widely in comprehensiveness.

Molecular and Cellular Biology

. Sincethe contents of computer-generatedimages can be manipulated for betterclarity, the Publications Board at itsMay 1992 meeting decreed that a de-scription of the software/hardware usedshould be put in the figure legend(s).

Journal of Cell Science

. Image en-hancement with computer software isacceptable practice, but there is a dan-ger that it can result in the presentationof quite unrepresentative data as well asin the loss of real and meaningful sig-nals. During manipulation of images, a

Reprinted with permission from The NIH Catalyst.

Address correspondence to Mike Rossner, Journalof Cell Biology, Rockefeller University Press,1114 1st Ave., 3rd fl., New York, NY 10021.Tel.: (212) 327-8881. Fax: (212) 327-8576.email: [email protected]

1

The general principles presented here apply tothe manipulation of images using any powerfulimage-processing software; however, because ofthe popularity of Photoshop, we refer to severalspecific functions in this application.

Published July 6, 2004


2/5

12 The Journal of Cell Biology|

Volume 166, Number 1, 2004

positive relationship between the origi-nal data and the resulting electronicimage must be maintained. If a figurehas been subjected to significant elec-tronic manipulation, the specific na-ture of the enhancements must benoted in the legend or in the Materials

and Methods.

The Journal of Cell Biology

. No spe-cific feature within an image may beenhanced, obscured, moved, removed,or introduced. The grouping of imagesfrom different parts of the same gel, orfrom different gels, fields, or exposuresmust be made explicit by the arrange-ment of the figure (e.g., using dividinglines) and in the text of the figure leg-end. Adjustments of brightness, con-trast, or color balance are acceptable ifthey are applied to the whole image

and as long as they do not obscure oreliminate any information present inthe original. Nonlinear adjustments(e.g., changes to gamma settings) mustbe disclosed in the figure legend.

Because the last set of guidelines isby far the most comprehensive we havefound to date (full disclosure: we wrotethem), we will continually refer back tothem in the following discussionsof the use and misuse of digitalmanipulations.

Blots and gels

Gross misrepresentation

The simplest examples of inappropriatemanipulation are show in Fig. 1. De-leting a band from a blot, even if youbelieve it to be an irrelevant back-ground band, is a misrepresentation ofyour data (Fig. 1 A). Similarly, addinga band to a blot, even if you are onlycovering the fact that you loaded the

wrong sample, and you know for surethat such a protein or DNA fragmentor RNA is present in your sample, is a

misrepresentation of your data. In theexample shown in Fig. 1 B, the addi-tional band in lane 3 has been gener-ated by simply duplicating the band inlane 2.

Another example of using Photo-shop inappropriately to create data isillustrated in Fig. 2, in which a wholesingle panel has been replicated (ar-rows) and presented as the loadingcontrols for two separate experiments.

More subtle manipulations

Brightness/contrast adjustments.Adjust-ing the intensity of a single band in ablot constitutes a violation of the

widely accepted guideline that Nospecific feature within an image may beenhanced, obscured, moved, removed,or introduced. In the manipulated im-age in Fig. 3 A, the arrow indicates asingle band whose intensity was re-

duced to produce an impression ofmore regular fractionation. Althoughthis manipulation may not alter theoverall interpretation of the data, it still

constitutes misconduct.While it is acceptable practice to ad-

just the overall brightness and contrastof a whole image, such adjustmentsshould not obscure or eliminate anyinformation present in the original

Figure 1. Gross manipulation of blots. (A) Example of a band deleted from the originaldata (lane 3). (B) Example of a band added to the original data (lane 3).

Figure 2. Gross manipulation of blots. Example of a duplicated panel (arrows).



3/5

Whats in a picture? |

Rossner and Yamada 13

(Fig. 3 B). When you scan a blot, nomatter how strong the bands, there willinvariably be some gray background.

While it is technically within theguidelines to adjust the brightness andcontrast of a whole image, if you over-adjust the contrast so that the back-ground completely drops out (Fig. 3 B,part 2 vs. part 3), this should raise sus-

picions among reviewers and editorsthat other information (especially faintbands) may have dropped out as well.

It may be argued that this guidelineis stricter than in the days before Pho-toshop, when multiple exposures couldbe used to perfect the presentation ofthe data. Perhaps it is, but this is justone of the advantages of the digital ageto the reviewer and editor, who cannow spot these manipulations when in

the past an author would have takenthe time to do another exposure.Think about this when you are doingthe experiment and perform multipleexposures to get the bands at the den-sity you want, without having to over-adjust digitally the brightness and con-trast of the scanned image.

Cleaning up background. It is very

tempting to use the tool variouslyknown as Rubber Stamp or CloneStamp in Photoshop to clean up un-

wanted background in an image (Fig.4). Dont do it. This kind of manipula-tion can usually be detected by some-one looking carefully at the image filebecause it leaves telltale signs. More-over, what may seem to be a back-ground band or contamination mayactually be real and biologically

important and could be recognized assuch by another scientist.

Splicing lanes together. It is clearlyinappropriate manipulation to take aband from one part of a gel and moveit to another part, even if you do notchange its size. But it is within usualguidelines to remove a complete lanefrom a gel and splice the remaininglanes together. This alteration shouldbe clearly indicated, however, by leav-ing a thin white or black line betweenthe gel pieces that have been juxta-posed. Again, it could be argued thatthis guideline is stricter than in thedays before Photoshop when paperphotographs of a gel were cut up andpieces were glued next to each other.This practice, however, usually left ablack line indicating to the reader what

had been done.As it was with gel photographs, it isunacceptable to juxtapose pieces fromdifferent gels to compare the levels ofproteins or nucleic acids. Rerun all ofthe samples on the same gel!

Figure 3. Manipulation of blots: brightness and contrast adjustments. (A) Adjusting the in-tensity of a single band (arrow). B) Adjustments of contrast. Images 1, 2, and 3 show sequen-tially more severe adjustments of contrast. Although the adjustment from 1 to 2 is accept-able because it does not obscure any of the bands, the adjustment from 2 to 3 isunacceptable because several bands are eliminated. Cutting out a strip of a blot with the

contrast adjusted provides the false impression of a very clean result (image 4 was derivedfrom a heavily adjusted version of the left lane of image 1). For a more detailed discussionof gel slicing and dicing, see Nature Cell Biologyeditorial (2).

Figure 4. Manipulation of blots: cleaningup background. The Photoshop RubberStamp tool has been used in the manipu-lated image to clean up the background inthe original data. Close inspection of the im-age reveals a repeating pattern in the leftlane of the manipulated image, indicatingthat such a tool has been used.



4/5

14 The Journal of Cell Biology|

Volume 166, Number 1, 2004

Micrographs

Enhancing a specific feature.An exam-ple of manipulation by enhancement isshown in Fig. 5, in which the intensityof the gold particles has been enhancedby manually filling them in with blackcolor using Photoshop. This type of

manipulation misrepresents your origi-nal data and is thus misconduct. Thereare acceptable ways to highlight a fea-ture such as gold particles, which in-clude arrows or pseudocoloring. Ifpseudocoloring is done with the Col-orize function of Photoshop, it doesnot alter the brightness of individualpixels, but pseudo-coloring should al-

ways be disclosed in the figure legend.Other examples of misconduct in-

clude adjusting the brightness of only aspecific part of an image or erasing

spots. Using the Brightness adjust-ment in Photoshop is considered to bea linear alteration (see below), whichmust be made to the entire image.

Linear vs. nonlinear adjustments. Lin-ear adjustments, such as those forBrightness or Contrast in Photo-shop, are those in which the samechange is made to each pixel accordingto a linear function. It is acceptable(within limits noted above) to applylinear adjustments to a whole image.There are other adjustments in Photo-

shop that can be applied to a whole im-age, but the same change is not madeto each pixel. For example, adjust-ments of gamma output (Color Set-tings in Photoshop) alter the intensityof each pixel according to a nonlinearfunction. Adjustments of Curves or

Levels in Photoshop alter the tonalrange and color balance of an image byadjusting the brightness of only thosepixels at particular intensities and col-ors. Such nonlinear changes are some-times required to reveal important fea-tures of an image; however, the fact

that they have been used should be dis-closed in the figure legend.Digitally altering brightness or con-

trast levels can be misleading with fluo-rescence micrographs. Some authorsmistakenly change the contrast of anexperimental compared with a controlphoto, or change individual panels in atime course, or use different contrastlevels when making merged imagescompared with the original images. Allof these changes in individual picturesused for comparisons can be misrepre-

sentations. On the other hand, certainadjustments such as background sub-traction or using a filter or digital maskmay be needed to extract informationaccurately from complex images. Re-porting the details and logic of suchmanipulations that are applied to im-ages as a whole should resolve concernsabout their use. Standards and guide-lines in the field will continue toevolve, but full disclosure will alwaysbe the safest course.

Misrepresentation of a microscope

field. The reader assumes that a singlemicrograph presented in a figure repre-sents a single microscope field. Com-bining images from separate micro-scope fields into a single micrographconstitutes a misrepresentation of youroriginal data. In the manipulated im-

age in Fig. 6 (top panel), cells havebeen combined from several micro-scope fields into a single micrograph.This manipulation becomes visible

when the contrast of the image is ad-justed so that the inserted images be-

come visible (bottom panel). You maywant to combine images from severalfields into a single micrograph to savespace, but this assembly should beclearly indicated by thin lines betweenthe different pieces.

Resolution

A pixel is a square (or dot) of uniformcolor in an image. The size of a pixelcan vary, and the resolution of an im-age is the number of pixels per unit

area. Although resolution is defined byarea, it is often described using a linearmeasurementdots per inch (dpi).Thus, 300 dpi indicates a resolution of300 pixels per inch by 300 pixels perinch, which equals 90,000 pixels persquare inch (1).

High-resolution digital cameras (in2004) can acquire an image that is 6megapixels in size. This can generatean image of approximately 2400

2400 pixels, or 8 inches

8 inches at

Figure 5. Misrepresentation of immunogold data. The gold particles, which were actuallypresent in the original (left), have been enhanced in the manipulated image (right). Note alsothat the background dot in the original data has been removed in the manipulated image.

Figure 6. Misrepresentation of image data.

Cells from various fields have been juxta-posed in a single image, giving the impres-sion that they were present in the same mi-croscope field. A manipulated panel isshown at the top. The same panel, with thecontrast adjusted by us to reveal the manip-ulation, is shown at the bottom.



5/5

Whats in a picture? |

Rossner and Yamada 15

300 dpi. Note that, with the right set-tings in Photoshop, physical size andresolution can be traded off againsteach other without a gain or loss in theamount of informationthat is, youcan resize an image without alteringthe total number of pixels.

You should be aware of the resolu-tion at which the image was acquiredby the digital camera on your micro-scope. When that file is opened in Pho-toshop, you have the option of settingthe size and resolution of the image.

You should not set the total number ofpixels to be greater than that in theoriginal image; otherwise, the com-puter must create data for you that

were not present in the original, andthe resulting image is a misrepresenta-tion of the original datathat is, the

dpi of an image can only be increasedif the size of the image is reducedproportionately.

It is acceptable to reduce the numberof pixels in an image, which may benecessary if you have a large image athigh resolution and want to create asmall figure out of it. Reducing the res-olution of an image is done in Photo-shop by sampling the pixels in an areaand creating a new pixel that is an aver-age of the color and brightness of thesampled ones. Although this does alter

your original data, you are not creating

something that was not there in thefirst place; you are presenting anaverage.

Other data-management issues

It is crucially important to keep youroriginal digital or analog data exactly as

they were acquired and to record yourinstrument settings. This primary ruleof good scientific practice will allowyou or others to return to your originaldata to see whether any information

was lost by the adjustments made tothe images. In fact, some journal re-viewers or editors request access to suchprimary data to ensure accuracy.

There are other important issuesconcerning data handling that we havenot addressed by focusing on manipu-lations of existing data. Examples in-

clude selective acquisition of data byadjusting the settings on your micro-scope or imager, selecting and report-ing a very unusual result as being repre-sentative of the data, or hiding negativeresults that may contradict your con-clusions. Any type of misrepresentationof experimental data undermines scien-tific research and should be avoided.

Conclusion

Data must be reported directly, notthrough a filter based on what you think

they should illustrate to your audience.

For every adjustment that you make to adigital image, it is important to ask your-self, Is the image that results from thisadjustment still an accurate representa-tion of the original data? If the answer tothis question is no, your actions may beconstrued as misconduct.

Some adjustments are currently con-sidered to be acceptable (such aspseudocoloring or changes to gammasettings) but should be disclosed toyour audience. You should, however,always be able to justify these adjust-ments as necessary to reveal a featurealready present in the original data.

We hope that by listing guidelinesand publicizing examples of transgres-sions, all of us can become more vigi-lant, particularly in guiding junior col-leagues and students away from the

tempting dangers of digital manipula-tion. Just because the tools exist toclean up sloppy work digitally, that isno excuse to do sloppy work.

If you would have redone an experi-ment to generate a presentation-qualityimage in the days before the digital age,you should probably redo it now.

References

1. Rossner, M. and R. ODonnell. 2004. The JCB

will let your data shine in RGB.

J. Cell. Biol.

164:

11.

2. 2004. Gel slicing and dicing: a recipe for disaster.

Nat. Cell Biol.

6:275.


j cell biol 2004 rossner 11 5

Documents