How to… image annotation

Make images understandalble: explain all that is visible
images
how-to
Author

Helena Jambor

Published

August 8, 2023

How to… image annotation

I already wrote why we need to annotate more, how to choose your image format, and how to annotate physical scales. But we also have to annotate symbols, regions, stainings and fluorphores (for general colors wait for next post.

Figure 1: Image annotations, the good, the bad and the ugly from Jambor 2021

Color annotations

Image colors are usually annotated in the color they are shown, this is fast and easy for readers. Problems may arise when image colors are not color-bind save, when the image has yellow stainigs that do not work for text. Alternatively, the color name is written out and linked to the molecule (GFP = green, actin-Cy3 = red etc). In this case the readers has to do a semantic translation of color name to image color, which is slow and error-prone as we all see colors differently.

Whenever you have >2-3 colors, you can safely assume that not all will be able to distinguish them any more. Not everyone is familiar with “lavender” or “salmon”, as we know since the hilarious and VERY exhaustive survey by XKCD.

Figure 2: His/Her color names by Doghouse Diaries and the color space with name-territoires by XKCD

Visibility of Annotations

Just like your image colors, also your annotations should be color-blind save. But some color combinations are also hard to decipher with regular vision.

Figure 3: ouch…

Symbols, arrows, ROIs

Oftentimes lines, arrows, symbols and regions of interest (ROIs) are used to point out important details or areas in an image. It is essential that all annotations are distinguishable and clearly explained. Avoid using the same symbol for several meanings: if arrows are used to point out structures, they should not concurrently be used to also indicate motion.

Figure 4: Examples of annotations in scientific images

Annotate your heart out..

Whatever you may feel important for your readers you should make an effort to help them take it in fast. Below are examples that annotate the position of laser manipulation with dashed-line ROI, color-code the z-stack depth, progressive zoom, and the overlap colors (good to be reminded in RGB space that mixing red, blue and green is WHITE).

Figure 5: Examples of what can be annotated.

Many annotations

If you went a bit overboard with your annotations, you may have to find creative solutions. Here are two examples:

Figure 6: Annotations, how much is too much? Some clever annotation strategy for many labels above, and a practical approach, make figures larger, here a full-page.

Consistent code from image to quantification

Figure 7: A good example of consistent color (left) and label (right) between image and chart.

Figure 8: Colored annotations are hard to see on multi-colored background - the link to the chart above is almost impossible to see (right).

Annotate where?

Ideally the explanation is in the figure itself, as close to the data as possible. Alternatively, explanations can go to the figure legend. Remember however that at times the legend is not even on the same page as the figure, that legends may have word limits, and that a legends in which many annotations have to be explained become tedious to read.

Speaking of.. Figure legends

Figure legends are essential for complete image annotation. They should inform about the specimen, maybe preparation, dyes and all essentials for understanding the image. Which information is in the figure versus the figure legend (and method section) depends on journal requirements and personal preference, but in sum the information should make the figure self-explanatory. A critical analysis by Yu of figure legends showed that up to 70% of the information needed to comprehend the figure was lacking, many times readers needed a complete paper to fully comprehend figures.

Read more

Are figure legends sufficient? Evaluating the contribution of associated text to biomedical figure comprehension. Yu, H., Agarwal, S., Johnston, M., Cohen, A., 2009. J Biomed Discov Collab