ADVANCING MICROSCOPY IMAGE ANALYSIS WITH CITIZEN SCIENCE AND MACHINE LEARNING
Helen Spiers is the Biomedical Research Lead of The Zooniverse platform for online citizen science (www.zooniverse.org), as well as research scientist at The Francis Crick Institute. In her role, Helen Spiers develops novel online citizen science research projects in collaboration with international and interdisciplinary groups. These projects typically apply collective intelligence to perform analysis on large volumes of imaging and biomedical data. She is an awardee of Cycle 2 of the CZI program for imaging scientists, where her aim is to generate and disseminate workflows that facilitate the creation of online citizen projects.
What was the inspiration for your project? How did your idea for the CZI project arise?
In short, we had been working in the space of applying online citizen science to image analysis for some time before submitting the fellowship application to CZI. We had encountered and identified a number of ways to extend the work we were doing, and the funding offered by CZI provided an opportunity to make it happen.
At what point in this process did you first hear about CZI and how did you decide to apply for this call?
Primarily word of mouth, I was aware that CZI had funded projects similar to mine in the past, so I was keeping my eyes open for the second call.
You have done impressive work with the Zooniverse project. How were you drawn to this project?
I was drawn to working with the Zooniverse team for lots of different reasons. In conventional academic positions you often have to funnel into a highly specific research area, however, as The Zooniverse hosts projects across humanities, ecology, astronomy, satellite imagery, natural disasters, and many other areas of science, it was possible for me to retain and nurture broader research interests. I was also drawn to working in the area of citizen science as it provides a space for anyone to connect with and contribute to authentic, cutting-edge research in a meaningful way.
How has the CZI support helped you reach your project goals?
CZI has offered helpful support in a range of ways. Perhaps because of CZI’s background and expertise in digital communication and connecting people, they are fantastic in drawing communities together. For instance, they have wonderful annual meetings that involve many opportunities to connect with other researchers and professionals. CZI also offers an extensive training program that includes content across an unusually broad variety of subjects, such as organizational culture, communications and strategic planning. Overall, CZI has provided lots of opportunities to connect with people, to learn, and to advocate for citizen science and how it may be applied it to microscopy.
How do you define democratizing microscopy and how does your project address this aspect?
Primarily, our project helps democratize microscopy through it’s ability to connect people, who are typically members of the public, to research they would otherwise have no way of being involved with. Not only do these projects allow volunteers to contribute directly to real research tasks, but they also provide opportunity for volunteers to sculpt project development and design through involvement in the Zooniverse project review process, and there are opportunities for volunteers and researchers to directly connect through talk forums too. The Zooniverse platform also make it possible for everyone to build a project – there is no requirement for people to be part of the academic community to create and share a project. We also seek to democratize our work as much as possible in relation to the outputs created; everything we generate is shared as openly as possible: the segmentations, the aggregated data, the trained models, the model predictions, and so on.
What are the biggest challenges you have faced, and what are your biggest successes?
One of the biggest challenges in online citizen science is cultivating opportunities to connect with new audiences, and seeking out ways to engage more deeply with existing contributors. Related to this, when we were building our first Etch A Cell project, we were concerned about how engaging and interesting our data would be to the Zooniverse volunteer community. Our projects predominantly involve fairly abstract looking greyscale images of cells, and we were competing with projects with data such as images of galaxies that had never been seen before, adorable penguins or wildlife in the Serengeti. However, we found that volunteers were excited and interested to contribute to Etch A Cell despite our worries, which I would say was a big success! Another challenge we have encountered has been that the task of contributing to the freehand segmentation of cell features typically takes much longer than the tasks often associated with online citizen science: in the range of minutes rather than seconds. Hence, we had concerns whether or not people would be willing to contribute that amount of time. Thankfully, we have found that there is a large community of volunteers who are willing and able to help with the Etch A Cell projects, and the data they have generated has been comparable to that produced by experts and so it could be used for analyses and training machine learning algorithms: another big success!
How do you see your project evolving in the future? Is it limitless in its potential, and if you see it shifting into something else, what would that be?
Technology is changing very quickly, so this is a question that comes up regularly in the context of Zooniverse. Provided we still need to use supervised training for machine learning in this space, there will still be a large need for human-generated data. Even if we generate a model that can handle every feature of interest in our current datasets, the hardware will inevitably change – we will generate novel microscopes and therefore novel data that will need human input to train new models to support future data analysis.
Check out our introductory post, with links to the other interviews here