Even in an age of reproducibility and quantification we often rely on visual inspection of image datasets. Depending on the image modality a lot of information is often encoded into a single image. For example temporal changes in the BOLD signal in fMRT or FA values in DTI images may be encoded in a color channel on top of a black and white T1 image for anatomical reference.

A huge set of tools is now available for the visualisation and color coding of different image modalities, many of which fail to accurately present the measured data. We specifically looked at FreeSurfer[3], SPM[1], and FSL[2] and show a set of common issues in these tools as well as a simple way to mitigate them.

We openend a set of standard datasets that ship with each of the tools named above and took screenshots of their visualisation using the build-in capture functionality in each software package. By convert the resulting images to the CIELAB color space in which the L* channel represents the perceived lightness of any given color.

We compare the L* channel of the screenshot to a non-color-mapped b/w version of the same dataset acquired from the same software. We compute lightness differences between these two images to show areas where the color-mapped image distorts the original data.

By overwriting the L* channel of the color-mapped image with the lightness information from the b/w representation we fix those distortions and retain any color information that was introduced in the mapping.

All of the tools we testes distort images in nearly all of our tests. color mapping is usually performed in RGB or HSV color spaces, which ignore the intrinsic link between color hue and perceived lightness in the human visual system. The CIELAB color space in contrast is perceptually uniform in all of its components and should therefore be used for these tasks.

Please see figure 1 for an example of distorted data due to different colormaps being applied in FSLView 3.2.0. In each row we show the same slice of a dataset, only varying the colormap. The first column contains the screenshot as taken by FSLView, the second column display the L* channel of the screenshot followed by a visualisation of the difference between L* channel and reference image and the corrected image in the last column. The first row contains the reference image for comparison.Our method of fixing the perceived intensity in color-mapped images through simple replacement of the L* channel could easily be applied in software. Until this algorithm in included in the software packages themselves, the correction can easily be applied afterwards with the help of tool that we released earlier this year[4].