A team of University of Washington researchers including former eScience postdoctoral fellow Anisha Keshavan, along with senior data scientist Ariel Rokem and affiliate Jason Yeatman have recently published a study titled “Combining citizen science and deep learning to amplify expertise in neuroimaging” in Frontiers in Neuroinformatics.
The team wants to understand how the brain develops during childhood and adolescence, and how deviations from normal brain development are related to mental health disorders. A large, open biobank called the Healthy Brain Network plans to release up to 10,000 brain magnetic resonance images of children and adolescents.
The first stage of analyzing these brain images is for researchers to inspect each image individually, and look for problems in image quality that could affect downstream results. But inspecting 10,000 images would take too long for a single team of researchers. So the team developed Braindr, a web application for citizen scientists to inspect and annotate brain images by swiping left or right. Braindr was gamified, so users could unlock various “badges” depending on how many images they scored. With these annotations, researchers trained a neural network to automatically score images at the same level of accuracy as an expert rater.
The success of Braindr led to the development of a generalized, open source, citizen science game platform called Swipes For Science, where researchers can spin up their own citizen science game to annotate large amounts of data with the help of citizen scientists. Swipes for Science is supported by eLife Innovation and early prototypes of Swipes for Science are being tested across various domains including social science, oceanography, natural language processing text annotation, and a variety of other brain imaging studies.