Rapid prototyping with CRUK on Trailblazer
In my previous post I talked about the approach that CRUK’s Citizen Science team took on Trailblazer: an experiment to see just how accurate the public could be at identifying cancerous cells in tumour tissue samples. Now I’d like to focus in on what we built. The final Trailblazer project was made up of three parts: Tutorial, Feedback and Mechanic (or Tasks).
The tutorial step was a standard part of PyBossa – the open source crowdsourcing framework we were using. However we had to make some tweaks so we could show the relevant information to our citizen scientists in a way that was as easy as possible to understand.
We had to create a new image carousel to dynamically load different slide information. We also had to integrate the zoom library into each slide; which proved quite a challenge.
After completing a few rounds of qualitative testing we found that our citizen scientists were great at identifying different cells but lacked confidence in doing so. This lack of confidence really drove the next few iterations.
We decided to create a feedback state directly after the tutorial. This was a place where the user could try to complete the task and be presented with immediate feedback on their choices. We found that this was a great place to add any additional image information that could help them when going through to the next stage: the mechanic.
The Mechanic part is also built into the PyBossa framework. By default, as part of each task, the citizen scientist is shown an image and asked to say whether a particular feature was present or not. We needed to customise this functionality so that citizen scientists could identify cancer cells in a 6 by 6 grid overlaid on each image. (I talked about this in more depth last time).
As part of this work we were also able to help the PyBossa team identify a bug with image loading in Internet Explorer. The PyBossa team were really helpful and friendly and it felt great to be able to contribute something to the wider citizen science project in this way.
In 3 months of prototyping we created around 15 different projects. After each release we had rounds of qualitative testing; which we found to be invaluable. When you’re working on an innovative project like Trailblazer, that really pushes the boundaries in terms of what can be done with crowdsourced scientific research, it’s easy for teams to jump to conlcusions about what the user needs. Getting a pair of fresh eyes on the project is a great way to see if the approach is working how you expected and, if not, time to re-evaluate.
Personally what I took from my time on Trailblazer is less about pure coding skills and more about learning about how to release products quickly and iterate on them. This approach is particularly important for teams like the Citizen Science team at CRUK as proving the concept helps secure vital funding and resources. Bells and whistles can come later.
It was also a great experience in terms of working on a team with lots of different skillsets, from data scientists to coders to non-technical people, all pulling together to complete a vital piece of work.