In the world of scientific research, images fall broadly into two categories: the original image and the corrected image. The original is acquired via an imaging device without any corrections applied in software. The corrected image, often referred to as “enhanced,” is often used for presentation and publication. The distinction is crucial when suspicions arise about content in images. Because images are used as visual proof of experimental evidence, certain alterations to the content may be viewed as unethical. Certainly, any additions to the content from other images, or intentional alterations of visual data to accommodate the experimental hypothesis, will always be unethical. The sole means for determining the extent or the existence of alterations or additions lies in looking at the original. A kind of reasoning can then follow: If the original is the indisputable visual evidence for experimental conclusions, only the original should be used. No alterations to visual data should be applied, and no corrections should be made for imaging device inconsistencies or shortcomings. This view operates along incorrect assumptions. Out of many assumptions, the following two are most often voiced (informally):

• Imaging systems and associated instrumentation can always produce a visual representation identical to what was seen by eye (within limitations inherent to two-dimensional renditions of three-dimensional specimens or scenes) if only those using the instrumentation were more knowledgeable.
• Problems with images are often a result of poorly prepared samples or the use of suboptimal dyes and stains: Bench practices need to be improved.

While both of these assumptions are true in many instances, they are based on a false premise that the instrumentation by itself, even when used correctly, produces accurate representations.

Efforts must be taken to present visual data as closely as possible to what was once perceived by eye. In other words, an image offered as proof must be a true representation of what was once seen. Any deviation from a correct representation is a misrepresentation, and the images become varying degrees of inaccurate data.

Accurate representation more often requires post-processing than not, except in instances in which optical densities or intensities (OD/I) are measured. Post-processing is required, for the most part, because of limitations in imaging devices and associated instrumentation. These limitations include, among others, the use of anti-aliasing filters in front of light detectors in many cameras, leading to blurring of images; inclusion of noise in images as a result of approaching detection limits for instrumentation; and variability in the energy source. Thus, for any of these reasons, the original image is corrected to create a better representation of visual data.

Additional reasons for inaccurate representation of the original as a result of imaging and display devices follow:

• Screen calibration. The screen on which the image is viewed may not display intensities and gradients of colors and grayscale levels appropriately (below, left). The gradients/intensities displayed on the screen may be inaccurate. When these images are incorrectly displayed and then presented or published, the inaccuracy is either perpetuated or the colors and contrasts change to even greater misrepresentations. Post-processing in color- and contrast-managed programs (Photoshop) will lead to better representations than the original (below, right).
  
  

    

• Hue shift. If a digital camera or scanner is used and the image is in color, the raw image is subject to a phenomenon known as “hue shift” (below, left) wherein the overall color shifts toward a hue, such as red or green. This can occur even after white-balancing the camera, and the shift differs among brands of cameras, flatbed scanning devices, and other types of scanners. Post-processing corrects for color shift so that the final images are better representations of the original than the raw scans were (below, right).
  
  

    

• Dynamic range. Some important features in the object or sample may be outside the dynamic range of the image recording device. These features show up as pure white or pure black in the image and contain no details (below, left). Often, the researcher will amplify (or diminish) these features when acquiring the image to reveal details in other features that are darker or brighter. Without the ability to alter the relationship of grayscale values (e.g., lighten the darker values in relation to the brighter values), the image is a misrepresentation of the specimen. Again, post-processing is required to create a truer representation of the original (below, right).
  
  

    

When and Where Misrepresentation Takes Place
Each example cited previously is an argument for the use of postprocessing even when instrumentation is used correctly. Still, this argument is not intended to dismiss the necessity for proper preparation of specimens or the need for greater education in regard to the use of imaging equipment and associated instrumentation. These devices must be used correctly before post-processing.

Also, another factor needs to be considered—how to maintain accurately represented images for various outputs, including hard copy (prints and posters), display on computer screens, inclusion in electronic documents, laptop projection at meetings, and publication. Originals must be conformed to each output, because each associated output device or software will modify the appearance of the image to the degree that the image can become inaccurate.

Thus, the potential for misrepresentation of visual evidence lies in three areas: when taking an image (acquisition), when correcting that image (post-processing), and when conforming that image to the output (conformance).

Much has been written about potential errors for a variety of imaging systems, including digital cameras, video cameras, scanning beam systems with photomultiplier (PMT) tubes, and flatbed scanners. In the area of image processing, attention has been drawn either to instances in which the potential for misrepresenting visual data exists or to specific instances in which a researcher purposefully altered visual data to get a desired result. Of course, the potential to alter visual data is always a possibility, but the system of scientific publication demands repetition of experiments to corroborate results, making the system self-correcting. The use of post-processing programs like Photoshop CS3, when done correctly and when users in all labs are trained in these methods, results in changes that should be made versus those that misrepresent visual data. Thus, in this article, references to changes made to an image are often termed as corrections versus more widely used terms like alterations or manipulations.

Conforming images to a specific output is an inevitable part of postprocessing. Changes to the image almost always require adjustments to grayscale or color tonal values, correction of the contrast, and possibly a change in both the format and mode of the image.

Note that conformance steps must be taken or images may become misrepresentations of visual data when reproduced to various outputs (below, left). While many scientists rely on hardware devices and software to interpret and interpolate visual data or on color correction professionals at printing presses, at some point the visual data will be changed. Ultimately, greater control over reproduction can be in the hands of scientists (below, right). That kind of control, with the aid of reference material such as the material found in Scientific Imaging with Photoshop, is more likely to prevent perhaps the greatest instance of misrepresentation in research and what has not been publicly addressed: reproduction to outputs.

Two reproduced images: researcher did not conform color to press output before sending image to the publisher (left); color was corrected before publishing (right). Science/NSF Visualization Challenge, 1st place, Photography, 2004: image courtesy of Marna Ericson, PhD.

Further Division of Areas into Categories
Images can be divided further into categories, depending on the intent of the image and the imaging system used for acquisition. Each group demands its own particular acquisition and post-processing treatment.

For example, if the visual data is intended for measurement of OD/I, changes to images acquired from that imaging system must be kept as close to the original image as possible, with only a few “allowable” post-processing methods and only the necessary methods for conformance to outputs (assuming the original was acquired on an imaging system that results in a linear distribution of grayscale or color values).

On the other hand, if the image is intended for visualization (in this sense, a conscious enhancement of visual data to draw attention to experimental phenomena, such as an image destined for a publication’s cover or a cartoon model) or any subsequent quantification (except OD/I), many more changes are allowable. In the case of quantification, these changes are absolutely necessary for separating measured features from the surrounding areas, often referred to as “background.”

The following categories broadly separate the intent and/or the means for acquiring the image:

• OD/I from flatbed scanners. Images made for measurement of brightness/darkness levels. These include electrophoretic samples done in the field of molecular biology (below,left).
• OD/I from camera and scanned beam systems. Images made for measurement of brightness/darkness levels, or for the measurement of color, not acquired from flatbed scanners. These include samples imaged via microscopes, confocal instruments, electron microscopy, x-ray devices, magnetic resonance imaging, and so on.
• Representation. Images made from any imaging device for purposes of creating an accurate representation of what was once seen (bemow, middle).
• Quantification/Visualization. Images made through conscious alteration of data using pseudocoloring, binarizing, and other techniques to separate relevant visual information from the background (below, right).

Three image types: specimen for OD/I (left), representative specimen (middle), and sample intended for quantification with right side binarized (right).

Author Guidelines
Following specific guidelines for acquisition, post-processing, and conformance ensures accurate representation of what was once seen by eye. Because part of the aim of research is to report or publish, the “gold standard” for ethical guidelines is found in the author guidelines from major scientific publications.

But the division of visual data into areas and categories will not be found in author guidelines from major scientific publications, at least to date. General guidelines are set forth in scientific publications to varying degrees of detail, but often these lack specifics for when and where potential for misrepresentation may exist. That lack of clarity leads to varying degrees of interpretation from one investigator to another and inconsistent responses from journal reviewers and editors.

Using Standards and References
In addition to when and where potential for misrepresentation exists, there is also the question of how: How can representative images be made while preserving a consistent approach to imaging?

The answer isn’t simple. In the best of situations, grayscale and color values can be objectively determined by fitting the imaging system (or the image, or the data derived from the image) to an external standard. Ideally, that standard is a calibrated object with known values. The imaging system can then be calibrated to the standard, and presto, all images from that system are also fitted to the known values. This is typical for situations in which OD/I measurements are derived from images acquired with self-calibrating, scanner systems.

However, other situations present difficulties. A calibrated standard may not be available, as in fluorescence imaging. Calibrated standards may work in an ideal world but not in the real world, such as in colors that can be defined by standards but do not exist with the same purity in a sample. The actual specimen may change in grayscale or color value as a result of preparation techniques and inherent factors, making a calibrated standard useless. Labeling of specimens may vary in intensity, making it more important to calibrate to an internal reference that is part of the specimen versus a calibrated standard. For these reasons and others, the use of an external, calibration standard is not always the answer for a consistent approach to imaging, which is what is desired for reproducibility and correct representations.

A consistent external reference—instead of a standard—may need to be substituted for a calibrated standard to provide a predictable reference value against which colors can be corrected or to which energy source intensities or exposure consistencies can be tracked over time. As in the calibrated standard, the external reference can be included with the specimen or taken at the beginning or end of an imaging session.

The reference can be internal: Specimens may have intrinsic values that can be ratioed against each other, or a consistent grayscale or color value may be found within the specimen, eliminating the need for either a calibration standard or external reference.

When no standards or references are available, distributions of grayscale or color values (histograms) can be matched to either a “perfect,” reference image, or images can be fit to the dynamic range of the imaging instrument (when acquiring images) or fit to a common histogram (post-processing). In that manner, all images are uniform.

As long as a reasoned approach is chosen for the type of specimen and the intent of the image, representative images can be produced, and imaging procedures can be duplicated. A summary of the approach to consistent imaging through the use of calibrated standards, internal references, or situations in which no standards or references exist is shown below.