More on color blindness

To the Editor:

In a recent issue of Nature Methods¹ Albrecht raised awareness of the need to carefully design graphic data representation so data can also be appropriately perceived by the many individuals who have color perception deficiencies. Appropriate color selection is not only important for effective communication to a wide audience, but there are also equity implications when developing effective teaching methods and communication strategies.

The Nature Methods editors in their reply² suggested using the Vischeck plugins for ImageJ and Photoshop platforms “for recoloring” purposes. By simulating color blindness, the Vischeck plugins are excellent for giving trichromatic ('normal' vision) observers an insight into how dichromatic viewers may see color images. However, on their own, these simulations do not provide a direct solution to the inverse problem of efficiently recoloring images in a way that allows dichromatic observers to discriminate the color-based information. At best, they may help with trial-and-error attempts at recoloring.

It is seriously problematic to recolor any arbitrary image so it can be perceived unambiguously by all types of dichromatic viewers simultaneously. However, in images with a limited number of hues, such as two-channel confocal images (additive colors) or two-dye bright-field histological stains (subtractive colors), digital imaging techniques now exist to reposition the color information so it is optimally targeted to hues that can be discriminated by the available functional retinal receptors. Below we provide suggestions on how to design the recoloring to resolve this problem in additive and subtractive color images.

Wong³ in a subsequent article on graphics design accounting for those with color vision deficiencies suggests using, for additive color images (typically two-channel confocal images), the method proposed by Okabe and Ito (http://jfly.iam.u-tokyo.ac.jp/color/). This consists of representing the image data as magenta-green pairs instead of the unfortunately common red-green pairs that are difficult for protanope and deuteranope (red and green color-blind; RGCB) observers to perceive. However this recoloring strategy in itself introduces another unexpected problem: RGCB observers tend to perceive the original magenta channel as a bluish color and the original green channel as a yellow-brownish color. That means that although the colors can be discriminated, in the absence of figure legends indicating the actual name of the colors, their descriptions become confusing in communication between trichromatic and RGCB observers. We suggest that instead blue-yellow color pairs are preferable to green-magenta ones because (i) they preserve the discrimination between channels, and (ii) both trichromatic and RGCB observers agree on the color names. This strategy can be easily applied to red-green channel pairs by first copying the green-channel data into the blue channel and duplicating the red channel in the green one.

In the case of two-dye histological stains (subtractive colors), we developed two other methods to reposition the original dye colors into the available visual discrimination space⁴ and proposed guidelines for the design of 'RGCB-safe' look up tables for false-coloring greyscale images⁵ (for instance, ultrasound, X-ray and infrared images). Examples of these approaches are available at http://dentistry.bham.ac.uk/cb/. We hope those resources will be useful to authors for designing more inclusive and effective scientific images.