06 February 2020

Critique: Intensive care neural networks

Today’s contribution comes from doctoral student William Caicedo. This was presented the Medical Devices and Technology satellite meeting in New Zealand last September. Click to enlarge!


William writes:

Despite being kind of an oddball there (my research deals with machine learning at the software level while most of the presentations where about medical hardware), I was awarded second place in the student poster competition, and I heavily suspect the design played an important role - all of the other posters used the traditional, overstuffed, word heavy design.

This is the “billboard poster” format in portrait mode. As I’ve mentioned before, when you have a big take home message, I don’t see the point of also having a “formal” title. I would remove the title so everything else could be bigger.

In particular, the results would benefit from being much bigger. The labels of the small multiples in the lower right side of the results section are unreadable.

The rest of the results are not much better. The condensed font used for the top diagram looks okay when I see it full size on my laptop are normal computer working distance, but not if I stand back a couple of feet. The lines in the graphs are all narrow and not particularly high contrast.

The top “take home” message has some inconsistent capitalization.

Interpretable Convolutional Neural Networks over-perform traditional techniques for the prediction of mortality in Intensive Care Units

Neither “convolutional neural networks” nor “intensive care units” are proper nouns, so there is no clear reason to capitalize them. Either of these would be more consistent, although I prefer sentence casing:

Interpretable Convolutional Neural Networks Over-perform Traditional Techniques for the Prediction of Mortality in Intensive Care Units (Headline casing)

Interpretable convolutional neural networks over-perform traditional techniques for the prediction of mortality in intensive care units (Sentence casing)

I’m also curious as to the choice of works to emphasize, which William does using colour. I get emphasizing “over-perform” (the result) and “intensive care units” (the setting). But “interpretable”? Why emphasize just one of two adjectives instead of the noun (“neural networks”)?

Similarly, the body of the poster highlights things that didn’t always make sense to me. Highlighting the neural network name, “ISeeU,” is helpful because it draws out the network the poster is describing.

I am guessing the reason for highlighting William’s name was that William was presenting the poster. I’m less clear on the reason for only highlighting William’s given name, rather than his full name.

And the last of these is the changing colour of headings: black for Introduction and Discussion, but red for Methods and “References.”

Would these changes have pushed William’s poster to the first place finish in the poster competition? Alas, we can never know. (P.S.—But if you won the competition and are reading this now, I’d love to see your work, too!)

Related posts

Critique: The Morrison billboard poster

No comments: